Latest / Boring Science / Nuclear Energy_ The Solution to Unlimited Artificial Intelligence _ Boring Science for Sleep

Transcript

Nuclear Energy_ The Solution to Unlimited Artificial Intelligence _ Boring Science for Sleep · Boring Science

• 0:00
  Speaker 1: Hey there. How was your day? Wherever you are in
• 0:03
  Speaker 1: the world, whatever time it is, I'm glad you're here.
• 0:07
  Speaker 1: Take a deep breath. Let your mind settle. In this
• 0:11
  Speaker 1: quiet documentary, we'll explore a question that touches the future
• 0:15
  Speaker 1: of both technology and humanity. Could nuclear energy be the
• 0:19
  Speaker 1: key to powering artificial intelligence, endlessly, safely and sustainably? We'll
• 0:26
  Speaker 1: walk through the science of nuclear power, The growing hunger
• 0:29
  Speaker 1: of AI systems, and the quiet alliance forming between the two.
• 0:34
  Speaker 1: You'll learn how energy shapes intelligence, how fission differs from fusion,
• 0:39
  Speaker 1: and what the world might look like if we choose
• 0:41
  Speaker 1: the atom to fuel our machines. The visuals are here
• 0:44
  Speaker 1: for you, if you want your eyes to be open.
• 0:47
  Speaker 1: They don't need to be very relevant to the speech,
• 0:49
  Speaker 1: always as this is more of a therapy for your sleep,
• 0:52
  Speaker 1: and a good listening experience. You can close your eyes
• 0:56
  Speaker 1: any time. Just listen. Let your thoughts drift across time, power,
• 1:03
  Speaker 1: and possibility. Let's begin. Imagine you're living in an era
• 1:11
  Speaker 1: where artificial intelligence can write poetry, diagnose diseases, and predict
• 1:15
  Speaker 1: stock market fluctuations faster than any human ever could. But
• 1:20
  Speaker 1: here's the catch. All of this incredible computational power comes
• 1:24
  Speaker 1: with a price that's growing more staggering by the day.
• 1:28
  Speaker 1: We're talking about energy consumption on a scale that makes
• 1:31
  Speaker 1: traditional data centres look like nightlights in comparison. Right now,
• 1:37
  Speaker 1: as you're listening to this, data centres across the globe
• 1:41
  Speaker 1: are devouring electricity at rates that would make entire countries jealous.
• 1:46
  Speaker 1: We're looking at a future where artificial intelligence could consume
• 1:50
  Speaker 1: more electricity than some of the world's largest nations. And
• 1:54
  Speaker 1: the numbers? They're absolutely mind-blowing. Current projections show that data
• 2:01
  Speaker 1: center power capacity in the United States alone is set
• 2:05
  Speaker 1: to skyrocket from 33 gigawatts in 2024 to a staggering
• 2:10
  Speaker 1: 120 gigawatts by 2030. That's not just growth. That's a
• 2:17
  Speaker 1: complete transformation of how we think about energy consumption. To
• 2:22
  Speaker 1: put this in perspective, we're talking about electricity usage that
• 2:25
  Speaker 1: could power entire regions. all dedicated to feeding the insatiable
• 2:30
  Speaker 1: appetite of artificial intelligence. But here's where it gets really interesting,
• 2:35
  Speaker 1: and slightly terrifying. The shift from traditional computing to AI-powered
• 2:41
  Speaker 1: systems isn't just about more computers. It's about fundamentally different
• 2:45
  Speaker 1: hardware that consumes energy like nothing we've seen before. Those
• 2:51
  Speaker 1: familiar CPU chips that have powered our computers for decades.
• 2:56
  Speaker 1: They typically run at around 150 to 200 watts, but
• 3:00
  Speaker 1: the GPU chips powering today's AI revolution, they're running at
• 3:04
  Speaker 1: 700 watts for current state-of-the-art models, with next-generation chips expected
• 3:11
  Speaker 1: to push that to 1 to 200 watts. And we're
• 3:14
  Speaker 1: just getting started. Companies like NVIDIA are pushing towards server
• 3:18
  Speaker 1: racks that consume an absolutely astronomical 1 megawatt of power.
• 3:23
  Speaker 1: That's eight times more power than current top-tier systems. We're
• 3:27
  Speaker 1: witnessing the birth of what could be called energy-hungry monsters,
• 3:32
  Speaker 1: and they're multiplying faster than we can build power plants
• 3:34
  Speaker 1: to feed them. The implications are staggering. Global data center
• 3:39
  Speaker 1: electricity consumption is estimated at around 415 terawatt hours in 2024,
• 3:42
  Speaker 1: representing about 1.5% of global electricity consumption. But by 2030,
• 3:52
  Speaker 1: that number is projected to more than double to around
• 3:55
  Speaker 1: 945 terawatt hours. In the United States specifically, data centers
• 4:00
  Speaker 1: consumed about 4.4% of total electricity in 2023, but by
• 4:06
  Speaker 1: 2028 they could be consuming anywhere from 6.7% to 12%
• 4:12
  Speaker 1: of total US electricity. Let me give you a reality
• 4:16
  Speaker 1: check on what these numbers actually mean. Data centres currently
• 4:20
  Speaker 1: have the capacity to consume 508 terawatt hours of electricity annually,
• 4:25
  Speaker 1: if running constantly. That's greater than the total annual electricity
• 4:30
  Speaker 1: production for Italy or Australia. And by 2034, global consumption
• 4:35
  Speaker 1: is expected to top 1,580 terawatt hours, about as much
• 4:40
  Speaker 1: as is used by all of India. Think about that
• 4:44
  Speaker 1: for a moment. We're essentially creating a digital civilisation that
• 4:49
  Speaker 1: consumes as much energy as some of the world's most
• 4:51
  Speaker 1: populous nations. And this isn't some distant future scenario. This
• 4:57
  Speaker 1: is happening right now, accelerating at breakneck speed. The driving
• 5:02
  Speaker 1: force behind this energy explosion isn't just more computers, it's
• 5:07
  Speaker 1: the fundamental shift in how we process information. Traditional computing
• 5:11
  Speaker 1: was relatively efficient, handling tasks in predictable, linear ways. But
• 5:17
  Speaker 1: artificial intelligence particularly the kind of AI that can understand language,
• 5:21
  Speaker 1: generate images, and make complex decisions, requires massive parallel processing.
• 5:28
  Speaker 1: It's the difference between a single person solving a math
• 5:31
  Speaker 1: problem and orchestrating a symphony with thousands of musicians. Every
• 5:36
  Speaker 1: time you ask ChatGPT a question, every time you generate
• 5:40
  Speaker 1: an image with AI, every time a recommendation algorithm suggests
• 5:44
  Speaker 1: your next video to watch, you're triggering computational processes that
• 5:48
  Speaker 1: involve billions of calculations happening simultaneously across specialized hardware. This
• 5:54
  Speaker 1: isn't just number crunching. It's digital orchestration on a scale
• 5:59
  Speaker 1: that's fundamentally changing our relationship with energy. The tech giants
• 6:03
  Speaker 1: are already feeling the heat, literally and figuratively. Google's carbon
• 6:08
  Speaker 1: emissions have risen 48% since 2019, mostly due to data
• 6:10
  Speaker 1: center energy consumption. Microsoft's carbon emissions were almost 31% higher
• 6:19
  Speaker 1: in 2024 than in 2020, largely driven by building data
• 6:24
  Speaker 1: centres for AI and manufacturing AI hardware. These companies, who
• 6:29
  Speaker 1: once prided themselves on green initiatives and carbon neutrality goals,
• 6:34
  Speaker 1: are now grappling with the reality that their AI ambitions
• 6:38
  Speaker 1: are directly conflicting with their environmental commitments. But here's where
• 6:43
  Speaker 1: the story takes a fascinating turn. rather than backing down
• 6:47
  Speaker 1: from AI development or accepting the environmental consequences, these tech
• 6:53
  Speaker 1: titans are making a bold bet on an energy source
• 6:57
  Speaker 1: that many thought was relegated to the history books. Nuclear power.
• 7:03
  Speaker 1: This isn't just about finding more electricity. It's about finding
• 7:07
  Speaker 1: the right kind of electricity. Artificial intelligence doesn't take breaks.
• 7:13
  Speaker 1: It doesn't work nine to five. It doesn't shut down
• 7:16
  Speaker 1: for weekends or holidays. AI systems need constant, reliable, 24-7
• 7:22
  Speaker 1: power delivery. And that's where most renewable energy sources hit
• 7:26
  Speaker 1: a wall. Solar panels don't generate electricity at night. Wind
• 7:32
  Speaker 1: turbines don't spin when the air is still. Even with
• 7:35
  Speaker 1: battery storage, which is expensive and requires significant physical footprint,
• 7:40
  Speaker 1: renewable energy sources struggle to provide the consistent, always on
• 7:45
  Speaker 1: power that AI systems demand. You can't tell a self-driving
• 7:49
  Speaker 1: car to wait for the wind to start blowing, and
• 7:52
  Speaker 1: you can't pause a medical diagnosis AI system because the
• 7:55
  Speaker 1: sun went behind a cloud. This is where nuclear energy
• 8:00
  Speaker 1: enters the picture as not just an option, but potentially
• 8:04
  Speaker 1: the only viable solution to power our AI-driven future. Nuclear
• 8:09
  Speaker 1: power plants can deliver hundreds of megawatts of power continuously,
• 8:14
  Speaker 1: without producing greenhouse gas emissions, regardless of weather conditions or
• 8:19
  Speaker 1: time of day. The numbers are compelling. A single nuclear
• 8:24
  Speaker 1: power plant can generate between 500 to 1,000 megawatts of
• 8:28
  Speaker 1: electricity continuously for decades. That's enough to power hundreds of
• 8:32
  Speaker 1: thousands of homes or several massive AI data centres. And
• 8:37
  Speaker 1: unlike fossil fuel plants, nuclear facilities can run at near
• 8:41
  Speaker 1: maximum capacity, 24 to 7, for months at a time,
• 8:46
  Speaker 1: providing exactly the kind of reliable baseload power that AI
• 8:50
  Speaker 1: systems require. But here's what makes this really interesting. We're
• 8:55
  Speaker 1: not just talking about building new nuclear plants. We're witnessing
• 8:59
  Speaker 1: something unprecedented in the history of energy. Tech companies are
• 9:04
  Speaker 1: literally bringing dead nuclear plants back to life. Microsoft recently
• 9:09
  Speaker 1: signed a deal to restart Three Mile Island, the Pennsylvania
• 9:12
  Speaker 1: power plant that was the scene of the worst commercial
• 9:15
  Speaker 1: nuclear accident in American history. The reactor being reopened wasn't
• 9:19
  Speaker 1: involved in the 1979 accident, but symbolically, this represents a
• 9:25
  Speaker 1: stunning reversal of decades of nuclear scepticism. Under the deal,
• 9:30
  Speaker 1: Microsoft will purchase energy from the restarted plant for 20 years,
• 9:34
  Speaker 1: with the facility providing 835 megawatts of electricity enough to
• 9:40
  Speaker 1: power about 700,000 homes. Never before has a US nuclear
• 9:45
  Speaker 1: plant come back into service after being decommissioned. Never before
• 9:50
  Speaker 1: has all of a single commercial nuclear power plant's output
• 9:54
  Speaker 1: been allocated to a single company. This isn't just energy procurement.
• 9:59
  Speaker 1: It's a fundamental shift in how we think about powering
• 10:02
  Speaker 1: the digital future. Amazon has gone even further, making what
• 10:07
  Speaker 1: might be the most comprehensive nuclear energy commitment in corporate history.
• 10:13
  Speaker 1: they've signed three separate agreements to develop nuclear power projects,
• 10:17
  Speaker 1: including deals to build several small modular reactors with a
• 10:21
  Speaker 1: combined capacity that could exceed 5 gigawatts by 2039. To
• 10:27
  Speaker 1: put that in perspective, that's enough electricity to power nearly
• 10:30
  Speaker 1: 4 million homes continuously. Google, not to be outdone, has
• 10:35
  Speaker 1: partnered with Kairos Power to develop up to 500 megawatts
• 10:39
  Speaker 1: of nuclear capacity using next-generation reactor designs, These aren't your
• 10:44
  Speaker 1: grandfather's nuclear plants. They're using advanced technologies like molten salt
• 10:49
  Speaker 1: cooling systems and ceramic fuel that promise to be safer,
• 10:54
  Speaker 1: more efficient, and more flexible than traditional nuclear designs. What
• 11:00
  Speaker 1: we're witnessing is the emergence of what might be called
• 11:03
  Speaker 1: the nuclear renaissance, driven not by governments or utilities, but
• 11:09
  Speaker 1: by private companies desperate to power their AI ambitions. These
• 11:14
  Speaker 1: aren't environmental activists or nuclear enthusiasts. These are hard-headed business
• 11:20
  Speaker 1: leaders who've done the math and concluded that nuclear power
• 11:24
  Speaker 1: is the only realistic path forward. The secret weapon in
• 11:28
  Speaker 1: this nuclear renaissance is something called Small Modular Reactors, or SMRs.
• 11:34
  Speaker 1: These aren't the massive, custom-built nuclear plants of the past
• 11:39
  Speaker 1: that took decades to build and cost tens of billions
• 11:42
  Speaker 1: of dollars. SMRs are designed to be factory built, standardized
• 11:47
  Speaker 1: and deployed relatively quickly. They have capacities ranging from 20
• 11:52
  Speaker 1: to 300 megawatts, making them perfect for powering individual data
• 11:57
  Speaker 1: centers or small clusters of facilities. The beauty of SMRs
• 12:02
  Speaker 1: lies in their modularity. Need more power? Add another reactor.
• 12:09
  Speaker 1: Want to upgrade your technology? replace individual units rather than
• 12:14
  Speaker 1: entire plants. It's like the difference between building a custom
• 12:19
  Speaker 1: skyscraper and assembling standardised components into the configuration you need.
• 12:26
  Speaker 1: These reactors are designed to be walk-away safe, meaning they
• 12:30
  Speaker 1: can shut down automatically without human intervention if something goes wrong.
• 12:35
  Speaker 1: They use advanced fuel designs that are much more stable
• 12:38
  Speaker 1: than traditional nuclear fuel. and many don't even require large-scale
• 12:42
  Speaker 1: cooling systems or emergency planning zones. But here's the really
• 12:46
  Speaker 1: exciting part. Some of these new reactor designs can be
• 12:50
  Speaker 1: placed almost anywhere. Unlike traditional nuclear plants that require massive
• 12:55
  Speaker 1: cooling towers and exclusion zones, many SMR designs are compact
• 13:00
  Speaker 1: enough to fit on a few acres and safe enough
• 13:03
  Speaker 1: to locate relatively close to population centres. The economics are
• 13:07
  Speaker 1: compelling too. Most SMRs under development could cost less than
• 13:13
  Speaker 1: $ 2 billion compared to more than $ 10 billion for traditional
• 13:16
  Speaker 1: nuclear plants. Their smaller size and standardized design mean they
• 13:21
  Speaker 1: can be built in 3 to 5 years instead of
• 13:24
  Speaker 1: the 6-12 years required for conventional reactors. This isn't just
• 13:28
  Speaker 1: about solving today's energy crisis. It's about fundamentally reimagining how
• 13:33
  Speaker 1: we power civilization. We're moving from a world where energy
• 13:38
  Speaker 1: production was centralized in massive plants located far from population
• 13:43
  Speaker 1: centers to a distributed model where clean, safe nuclear power
• 13:49
  Speaker 1: can be deployed wherever it's needed most. The implications extend
• 13:54
  Speaker 1: far beyond just powering AI systems. If small modular reactors
• 13:58
  Speaker 1: prove successful, they could revolutionize everything from industrial manufacturing to
• 14:04
  Speaker 1: residential communities. Imagine neighborhoods powered by their own small nuclear reactors,
• 14:11
  Speaker 1: completely independent of the electrical grid. Picture factories that never
• 14:15
  Speaker 1: worry about power outages or fluctuating electricity prices because they
• 14:19
  Speaker 1: have their own dedicated nuclear power source. The artificial intelligence
• 14:24
  Speaker 1: revolution is forcing us to confront fundamental questions about energy, environment,
• 14:29
  Speaker 1: and the future of human civilization. We're not just building
• 14:33
  Speaker 1: smarter computers. we're reshaping the entire energy landscape of our planet.
• 14:39
  Speaker 1: Looking at the nuclear energy landscape, we encounter an extraordinary paradox.
• 14:45
  Speaker 1: While artificial intelligence systems demand unprecedented amounts of reliable, clean electricity,
• 14:51
  Speaker 1: the very technology that could satisfy this appetite has remained
• 14:55
  Speaker 1: largely untapped for decades. Nuclear power, once heralded as the
• 15:00
  Speaker 1: energy source of the future, fell from grace following high-profile
• 15:05
  Speaker 1: accidents and mounting construction costs. Yet today, as data centres
• 15:11
  Speaker 1: consume electricity at rates that would power entire nations, a
• 15:15
  Speaker 1: quiet renaissance is taking place in the nuclear industry. The
• 15:19
  Speaker 1: numbers tell a compelling story. Modern AI training runs require continuous,
• 15:26
  Speaker 1: uninterrupted power flows that dwarf traditional computing needs. a single
• 15:32
  Speaker 1: large language model training session can consume as much electricity
• 15:36
  Speaker 1: as a small city uses in a month. Unlike manufacturing
• 15:41
  Speaker 1: plants or office buildings that can adjust their power consumption
• 15:45
  Speaker 1: based on grid availability, AI systems demand constant, reliable electricity
• 15:51
  Speaker 1: 24 hours a day, seven days a week. This creates
• 15:56
  Speaker 1: a unique challenge that traditional renewable energy sources struggle to meet.
• 16:02
  Speaker 1: wind and solar power, while environmentally friendly and increasingly cost-effective,
• 16:08
  Speaker 1: suffer from what engineers call intermittency. The wind doesn't always blow,
• 16:13
  Speaker 1: and the sun doesn't always shine. Battery storage systems, while
• 16:17
  Speaker 1: improving rapidly, remain expensive and require vast physical footprints to
• 16:22
  Speaker 1: store the energy needed for large-scale AI operations. Moreover, These
• 16:28
  Speaker 1: storage systems typically provide only a few hours of backup power,
• 16:33
  Speaker 1: nowhere near enough to sustain AI training runs that can
• 16:36
  Speaker 1: last for weeks or months. This is where nuclear energy's
• 16:41
  Speaker 1: unique characteristics become particularly attractive. Nuclear power plants operate at
• 16:46
  Speaker 1: what's called a high capacity factor, meaning they run at
• 16:49
  Speaker 1: full power nearly continuously, while wind farms might operate at
• 16:53
  Speaker 1: 30% capacity over a year due to variable wind conditions
• 16:58
  Speaker 1: Nuclear plants routinely achieve capacity factors above 90%. This reliability
• 17:05
  Speaker 1: isn't just convenient for AI applications, it's essential. The artificial
• 17:11
  Speaker 1: intelligence boom has created what energy experts describe as a
• 17:15
  Speaker 1: demand profile, unlike anything the electrical grid has encountered before.
• 17:20
  Speaker 1: Traditional industrial loads, even the most energy-intensive ones, typically have
• 17:25
  Speaker 1: predictable patterns. and can often be scheduled during off-peak hours.
• 17:31
  Speaker 1: Steel mills can time their production runs, aluminum smelters can
• 17:35
  Speaker 1: adjust their operations, and even cryptocurrency mining operations can shift
• 17:40
  Speaker 1: their computing loads to take advantage of cheaper electricity prices.
• 17:46
  Speaker 1: AI training, however, operates under different constraints. The massive parallel
• 17:51
  Speaker 1: processing required to train large neural networks means that stopping
• 17:56
  Speaker 1: and restarting operations isn't just inconvenient, it's often technically impossible
• 18:02
  Speaker 1: without losing weeks or months of computational progress. The interconnected
• 18:08
  Speaker 1: nature of modern AI systems means that losing power to
• 18:12
  Speaker 1: even a small portion of the computing cluster can crash
• 18:15
  Speaker 1: the entire training run. This creates what economists call a
• 18:19
  Speaker 1: highly inelastic demand for electricity. Unlike most industrial customers who
• 18:25
  Speaker 1: can reduce their power consumption when prices rise or supply
• 18:29
  Speaker 1: becomes constrained, AI operators must maintain their power draw regardless
• 18:35
  Speaker 1: of cost or grid conditions. This inelasticity has profound implications
• 18:40
  Speaker 1: for how we think about powering the AI revolution. Consider
• 18:45
  Speaker 1: the scale of what's happening. The largest AI training facilities
• 18:50
  Speaker 1: today house hundreds of thousands of specialised processors, each consuming
• 18:55
  Speaker 1: hundreds of watts of power. These facilities operate with power
• 19:00
  Speaker 1: densities that exceed those of traditional data centres by orders
• 19:04
  Speaker 1: of magnitude. A single AI training rack can consume more
• 19:09
  Speaker 1: electricity than an entire suburban neighbourhood. When multiplied across the
• 19:14
  Speaker 1: growing number of AI research facilities worldwide, the total electricity
• 19:19
  Speaker 1: demand becomes staggering. The geographic concentration of AI development adds
• 19:24
  Speaker 1: another layer of complexity. Unlike traditional industries that can be
• 19:28
  Speaker 1: distributed across regions with different energy profiles, AI development tends
• 19:34
  Speaker 1: to cluster in specific locations near universities, technology companies, and
• 19:39
  Speaker 1: skilled workforces. This concentration creates localized demand spikes that constrain
• 19:46
  Speaker 1: regional electrical grids and drive up prices for other users.
• 19:51
  Speaker 1: Nuclear energy offers a solution to these challenges that no
• 19:55
  Speaker 1: other power source can match. A single nuclear power plant
• 19:59
  Speaker 1: can provide the massive continuous electricity supply that AI operations require.
• 20:06
  Speaker 1: The fuel costs for nuclear plants are are remarkably low
• 20:10
  Speaker 1: compared to fossil fuel alternatives, and once built, nuclear facilities
• 20:15
  Speaker 1: can operate for decades with minimal fuel requirements. This long-term
• 20:21
  Speaker 1: stability is particularly attractive to AI companies planning massive computational investments.
• 20:28
  Speaker 1: The economics of nuclear power align well with the economics
• 20:32
  Speaker 1: of AI development. While nuclear plants require substantial upfront capital investment,
• 20:39
  Speaker 1: their operating costs remain stable for decades. This matches the
• 20:43
  Speaker 1: investment profile of AI companies, which typically make large upfront
• 20:48
  Speaker 1: investments in computing infrastructure and then amortise those costs over
• 20:52
  Speaker 1: many years of operation. The predictable, long-term nature of nuclear
• 20:58
  Speaker 1: power costs provides the financial certainty that AI companies need
• 21:02
  Speaker 1: for their strategic planning. Furthermore, nuclear energy offers something that
• 21:07
  Speaker 1: no other power source can provide at scale, carbon-free electricity
• 21:12
  Speaker 1: that doesn't depend on weather conditions or time of day.
• 21:17
  Speaker 1: As technology companies face increasing pressure to reduce their carbon footprints,
• 21:22
  Speaker 1: nuclear power allows them to pursue their AI ambitions without
• 21:25
  Speaker 1: compromising their environmental commitments. This is particularly important as AI
• 21:31
  Speaker 1: systems become more powerful and energy-intensive. The revival of interest
• 21:36
  Speaker 1: in nuclear power for AI applications represents more than just
• 21:40
  Speaker 1: a practical solution to an energy challenge. It signals a
• 21:44
  Speaker 1: fundamental shift in how we think about electricity generation and consumption.
• 21:49
  Speaker 1: For decades, the trend in electricity markets has been towards smaller,
• 21:54
  Speaker 1: more distributed generation sources. Solar panels on rooftops, small wind turbines,
• 22:01
  Speaker 1: and battery storage systems have promised a future where electricity
• 22:05
  Speaker 1: generation is democratised and distributed. The massive power requirements of
• 22:11
  Speaker 1: AI systems challenge this distributed model. While rooftop solar might
• 22:17
  Speaker 1: suffice for residential needs, no collection of small renewable sources
• 22:22
  Speaker 1: can provide the concentrated, reliable power that AI training requires.
• 22:28
  Speaker 1: This has led to renewed interest in large-scale, centralised power generation.
• 22:33
  Speaker 1: with nuclear energy emerging as the most viable option. The
• 22:38
  Speaker 1: technical requirements of AI systems also favor nuclear power over
• 22:42
  Speaker 1: other alternatives. The computing processes used in AI training are
• 22:47
  Speaker 1: extremely sensitive to power quality issues. Voltage fluctuations, frequency variations,
• 22:53
  Speaker 1: and even brief interruptions can cause expensive equipment failures or
• 22:58
  Speaker 1: data corruption. Nuclear plants with their massive rotating generators and
• 23:03
  Speaker 1: sophisticated control systems, provide the stable, high-quality electricity that sensitive
• 23:10
  Speaker 1: AI equipment requires. The timeline for AI development also favours
• 23:15
  Speaker 1: nuclear solutions. While the construction of new nuclear plants can
• 23:20
  Speaker 1: take many years, the development of truly powerful AI systems
• 23:25
  Speaker 1: also requires long-term thinking and planning. Companies investing in AI
• 23:30
  Speaker 1: research today are making bets on technologies that may not
• 23:34
  Speaker 1: reach full commercial deployment for a decade or more. This
• 23:37
  Speaker 1: long-term perspective aligns well with the development timelines for nuclear
• 23:41
  Speaker 1: power projects. The synergy between nuclear energy and artificial intelligence
• 23:47
  Speaker 1: extends beyond simple power supply. Both technologies represent humanity's ability
• 23:53
  Speaker 1: to harness fundamental forces of nature for beneficial purposes. Nuclear
• 23:57
  Speaker 1: power taps into the binding energy of atomic nuclei, while
• 24:02
  Speaker 1: AI systems attempt to replicate the information processing capabilities of
• 24:06
  Speaker 1: biological neural networks. Both require sophisticated engineering, advanced materials, and
• 24:11
  Speaker 1: careful safety protocols. The combination of these technologies offers the
• 24:16
  Speaker 1: possibility of addressing some of humanity's most pressing challenges. Climate change,
• 24:22
  Speaker 1: resource scarcity, disease, and poverty all require solutions that combine
• 24:27
  Speaker 1: massive computational power with clean, abundant energy. Nuclear-powered AI systems
• 24:35
  Speaker 1: could accelerate the development of new medicines, optimize energy distribution networks,
• 24:41
  Speaker 1: design more efficient transportation systems, and model complex environmental systems.
• 24:48
  Speaker 1: As we stand at this technological crossroads, The decisions we
• 24:52
  Speaker 1: make about energy infrastructure will shape the future of artificial
• 24:56
  Speaker 1: intelligence and, by extension, human civilisation itself. The question isn't
• 25:02
  Speaker 1: whether AI will continue to demand more electricity, but rather
• 25:07
  Speaker 1: how we will choose to generate that electricity. Nuclear power
• 25:11
  Speaker 1: offers a path forward that combines the reliability, scale and
• 25:16
  Speaker 1: environmental benefits necessary to support the AI revolutions. while addressing
• 25:20
  Speaker 1: the urgent need for clean energy solutions. The marriage of
• 25:23
  Speaker 1: nuclear energy and artificial intelligence represents more than a technological partnership.
• 25:30
  Speaker 1: It embodies humanity's capacity to turn scientific understanding into practical
• 25:35
  Speaker 1: solutions for complex challenges. As we move forward into an
• 25:40
  Speaker 1: age of unprecedented computational power and artificial intelligence, nuclear energy
• 25:45
  Speaker 1: stands ready to provide the clean, reliable electricity that will
• 25:49
  Speaker 1: power the next chapter of human technological advancement. The dance
• 25:54
  Speaker 1: between artificial intelligence and nuclear power has barely begun, yet
• 25:58
  Speaker 1: it's already reshaping how we think about energy production and
• 26:01
  Speaker 1: consumption in ways that would have been unimaginable just a
• 26:04
  Speaker 1: few years ago. We've established that AI's hunger for electricity
• 26:09
  Speaker 1: is unprecedented, with data centres projected to consume as much
• 26:13
  Speaker 1: power as entire nations by the end of this decade.
• 26:18
  Speaker 1: we've seen how traditional renewable sources struggle to meet the constant,
• 26:22
  Speaker 1: reliable demands of AI systems that never sleep, never pause,
• 26:27
  Speaker 1: and never stop learning. Now we must dive deeper into
• 26:30
  Speaker 1: the heart of this energy revolution to understand not just
• 26:33
  Speaker 1: what nuclear power offers, but how it fundamentally changes the
• 26:36
  Speaker 1: entire equation of artificial intelligence development. The relationship between AI
• 26:43
  Speaker 1: and nuclear energy isn't simply about finding a bigger battery,
• 26:47
  Speaker 1: or a more powerful generator. It's about aligning two of
• 26:51
  Speaker 1: humanity's most transformative technologies in a way that amplifies the
• 26:55
  Speaker 1: potential of both. When we examine this partnership closely, we
• 26:59
  Speaker 1: discover that nuclear power doesn't just solve AI's energy problems.
• 27:04
  Speaker 1: It unlocks entirely new possibilities for what artificial intelligence can accomplish.
• 27:09
  Speaker 1: The constant, predictable flow of nuclear energy creates a stable
• 27:13
  Speaker 1: foundation upon which AI systems can build increasingly sophisticated capabilities,
• 27:19
  Speaker 1: while AI's massive energy demands provide the economic justification for
• 27:24
  Speaker 1: nuclear expansion that the industry has desperately needed. Consider the
• 27:30
  Speaker 1: profound implications of this symbiotic relationship. For decades, nuclear power
• 27:36
  Speaker 1: has struggled with public perception, regulatory hurdles and economic challenges
• 27:42
  Speaker 1: that have limited its growth despite its environmental benefits, the
• 27:46
  Speaker 1: nuclear industry has been waiting for a compelling use case
• 27:49
  Speaker 1: that would drive demand for new reactor construction and justify
• 27:53
  Speaker 1: the enormous capital investments required. AI has provided exactly that catalyst.
• 28:01
  Speaker 1: When Microsoft signs a 20-year power purchase agreement to restart
• 28:05
  Speaker 1: Three Mile Island, when Amazon commits billions to small modular
• 28:09
  Speaker 1: reactor development, When Google partners with advanced reactor companies, they're
• 28:14
  Speaker 1: not just buying electricity. They're providing the economic anchor that
• 28:19
  Speaker 1: makes nuclear expansion financially viable. This economic transformation extends far
• 28:26
  Speaker 1: beyond individual corporate deals. The AI boom is creating a
• 28:30
  Speaker 1: new class of energy consumers with unique characteristics that perfectly
• 28:34
  Speaker 1: match nuclear power's strengths. Unlike traditional industrial users who might
• 28:38
  Speaker 1: reduce consumption during economic downturns, AI data centres represent a stable,
• 28:44
  Speaker 1: growing and predictable source of electricity demand. These facilities operate continuously,
• 28:51
  Speaker 1: consuming power at consistent rates that align beautifully with nuclear
• 28:55
  Speaker 1: plants' preference for steady baseload operation. The result is a
• 28:59
  Speaker 1: marriage of technologies where each partner's strengths complement the other's needs.
• 29:05
  Speaker 1: The geographic implications of this partnership are equally fascinating. Traditional
• 29:10
  Speaker 1: data centres have typically been located near population centres or
• 29:15
  Speaker 1: internet infrastructure hubs, but the marriage of AI and nuclear
• 29:19
  Speaker 1: power is creating new possibilities for distributed computing. Small modular
• 29:25
  Speaker 1: reactors can be deployed in remote locations where land is
• 29:28
  Speaker 1: cheaper and regulatory approval may be easier to obtain. This
• 29:34
  Speaker 1: geographic flexibility allows AI companies to build massive computational facilities.
• 29:40
  Speaker 1: without straining urban electrical grids or competing with residential users
• 29:44
  Speaker 1: for power. The result could be a new map of
• 29:48
  Speaker 1: technological development with AI centers clustered around nuclear facilities in
• 29:53
  Speaker 1: previously underdeveloped regions. But the technical synergies run even deeper
• 29:59
  Speaker 1: than economics and geography. Nuclear power plants generate electricity through
• 30:05
  Speaker 1: a process that's remarkably similar to how AI systems process information.
• 30:12
  Speaker 1: Both involve complex feedback loops, precise control systems, and the
• 30:17
  Speaker 1: management of enormous amounts of data. Nuclear plants have been
• 30:22
  Speaker 1: pioneers in automated control systems, Using sophisticated algorithms to monitor
• 30:28
  Speaker 1: reactor conditions, adjust control rod positions, and optimize power output,
• 30:35
  Speaker 1: these systems generate massive amounts of operational data that AI
• 30:39
  Speaker 1: can analyze to improve plant efficiency, predict maintenance needs, and
• 30:44
  Speaker 1: enhance safety protocols. The potential for AI to revolutionize nuclear
• 30:48
  Speaker 1: plant operations is staggering. Machine learning algorithms can analyze patterns
• 30:53
  Speaker 1: in reactor behavior that human operators might miss, predicting equipment
• 30:57
  Speaker 1: failures before they occur and optimizing fuel usage to extend
• 31:00
  Speaker 1: reactor life. AI can process real-time data from thousands of
• 31:06
  Speaker 1: sensors throughout a nuclear facility, identifying subtle anomalies that could
• 31:11
  Speaker 1: indicate developing problems. The result is safer, more efficient nuclear
• 31:16
  Speaker 1: plants that can operate with higher capacity factors and lower
• 31:20
  Speaker 1: maintenance costs. This technological convergence is already beginning to manifest
• 31:26
  Speaker 1: in the design of new nuclear facilities. The small modular
• 31:30
  Speaker 1: reactors being developed by companies like NuScale, Kairos Power, and
• 31:35
  Speaker 1: Xenergy incorporate AI-driven control systems from the ground up. These
• 31:41
  Speaker 1: systems can automatically adjust reactor parameters to match electricity demand
• 31:46
  Speaker 1: from AI data centers. creating a responsive energy supply that
• 31:51
  Speaker 1: can scale up or down based on computational needs. The
• 31:55
  Speaker 1: result is a level of grid flexibility that traditional nuclear plants,
• 32:00
  Speaker 1: with their preference for steady-state operation, have never been able
• 32:04
  Speaker 1: to provide. The safety implications of this AI nuclear partnership
• 32:10
  Speaker 1: are particularly compelling. Nuclear safety has always depended on multiple
• 32:15
  Speaker 1: redundant systems, and conservative operating procedures designed to prevent accidents.
• 32:21
  Speaker 1: AI adds another layer of protection by continuously monitoring plant
• 32:25
  Speaker 1: conditions and predicting potential problems with unprecedented accuracy. Machine learning
• 32:31
  Speaker 1: algorithms can process vast amounts of historical operating data to
• 32:36
  Speaker 1: identify patterns that precede equipment failures or safety events. This
• 32:41
  Speaker 1: predictive capability allows plant operators to take preventive action before
• 32:45
  Speaker 1: problems develop, reducing both safety risks and maintenance costs. Furthermore,
• 32:51
  Speaker 1: AI can enhance nuclear security by monitoring for cyber threats,
• 32:56
  Speaker 1: unauthorised access and other security concerns. The same pattern recognition
• 33:01
  Speaker 1: capabilities that allow AI to optimise reactor operations can also
• 33:06
  Speaker 1: detect unusual activities that might indicate security breaches. This is
• 33:10
  Speaker 1: particularly important as nuclear facilities become increasingly digitised and connected
• 33:16
  Speaker 1: to broader energy networks. AI-powered security systems can provide real-time
• 33:22
  Speaker 1: threat detection and response capabilities that human security personnel could
• 33:27
  Speaker 1: never match. The environmental benefits of the AI nuclear partnership
• 33:32
  Speaker 1: extend far beyond simply reducing carbon emissions from electricity generation.
• 33:39
  Speaker 1: Nuclear power's reliability means that AI systems can operate continuously
• 33:44
  Speaker 1: without the need for backup power from fossil fuel generators.
• 33:50
  Speaker 1: This eliminates the carbon emissions associated with standby power systems
• 33:54
  Speaker 1: and reduces the overall environmental footprint of AI operations. Additionally,
• 34:00
  Speaker 1: the stable power supply from nuclear plants reduces the need
• 34:04
  Speaker 1: for energy storage systems, which have their own environmental impacts
• 34:08
  Speaker 1: from battery production and disposal. The partnership also creates opportunities
• 34:13
  Speaker 1: for AI to optimize the entire nuclear fuel cycle. Machine
• 34:18
  Speaker 1: learning algorithms can analyze uranium ore deposits to optimize mining operations,
• 34:24
  Speaker 1: reduce waste, and minimize environmental impact. AI can optimize fuel
• 34:29
  Speaker 1: fabrication processes to create more efficient reactor fuel. And it
• 34:35
  Speaker 1: can help manage nuclear waste by predicting the long-term behavior
• 34:38
  Speaker 1: of storage systems and optimizing disposal strategies. The result is
• 34:43
  Speaker 1: a more sustainable nuclear industry that maximizes the value of
• 34:48
  Speaker 1: every uranium atom while minimizing environmental impact. Perhaps most importantly,
• 34:54
  Speaker 1: the AI nuclear partnership is accelerating innovation in both fields.
• 35:00
  Speaker 1: The massive computational power available from AI systems is enabling
• 35:05
  Speaker 1: new approaches to nuclear reactor design, safety analysis, and material science,
• 35:10
  Speaker 1: AI can simulate complex nuclear physics phenomena that would be
• 35:14
  Speaker 1: impossible to model with traditional computational methods. This capability is
• 35:19
  Speaker 1: accelerating the development of advanced reactor designs, improved safety systems,
• 35:24
  Speaker 1: and new nuclear materials with enhanced performance characteristics. The economic
• 35:30
  Speaker 1: spillover effects of this partnership are already becoming apparent. The
• 35:35
  Speaker 1: demand for AI-powered nuclear facilities is driving investment in nuclear
• 35:40
  Speaker 1: supply chains, creating jobs in reactor manufacturing, fuel fabrication and
• 35:45
  Speaker 1: plant construction. Communities that host new nuclear facilities benefit from stable,
• 35:52
  Speaker 1: high-paying jobs and increased tax revenues. The nuclear industry's revival
• 35:58
  Speaker 1: is also spurring innovation in related fields, from advanced materials,
• 36:03
  Speaker 1: science to robotics and automation. International implications of the AI
• 36:08
  Speaker 1: nuclear partnership are equally significant. Countries that can successfully integrate
• 36:14
  Speaker 1: AI and nuclear power will have substantial advantages in the
• 36:17
  Speaker 1: global economy. They'll be able to host the massive computational
• 36:22
  Speaker 1: facilities needed for AI development while maintaining low-carbon energy systems.
• 36:28
  Speaker 1: This creates new forms of technological sovereignty and economic competitiveness
• 36:32
  Speaker 1: that extend far beyond traditional measures of national power The
• 36:37
  Speaker 1: regulatory landscape is evolving rapidly to accommodate this new partnership.
• 36:42
  Speaker 1: Nuclear regulatory agencies are developing new frameworks for licensing AI-enhanced
• 36:48
  Speaker 1: reactor control systems and evaluating the safety implications of AI
• 36:52
  Speaker 1: nuclear integration. These regulatory innovations are creating new standards for
• 36:58
  Speaker 1: how AI systems can be safely incorporated into nuclear facilities.
• 37:03
  Speaker 1: paving the way for broader adoption of these technologies. As
• 37:07
  Speaker 1: we look toward the future, the AI nuclear partnership promises
• 37:10
  Speaker 1: to transform both industries in ways we're only beginning to understand.
• 37:15
  Speaker 1: The stable, abundant power from nuclear plants will enable AI
• 37:19
  Speaker 1: systems to tackle increasingly complex problems, from climate modelling to
• 37:23
  Speaker 1: materials discovery to space exploration. Meanwhile, AI's optimisation capabilities will
• 37:30
  Speaker 1: make nuclear power safer, more efficient and more economical than
• 37:35
  Speaker 1: ever before. The convergence of these two transformative technologies represents
• 37:41
  Speaker 1: more than just a solution to AI's energy needs. It's
• 37:45
  Speaker 1: the foundation for a new era of human capability, where
• 37:49
  Speaker 1: the limitless potential of artificial intelligence is matched by the abundant,
• 37:54
  Speaker 1: clean energy needed to realise that potential. The dance between
• 38:00
  Speaker 1: AI and nuclear power is just beginning, but it's already
• 38:04
  Speaker 1: clear that their partnership will reshape our world in ways
• 38:07
  Speaker 1: we're only starting to imagine. Looking at the progression of
• 38:12
  Speaker 1: our nuclear energy and AI power discussion, we've established the
• 38:16
  Speaker 1: unprecedented energy demands of artificial intelligence, examined how renewable sources
• 38:22
  Speaker 1: struggle with reliability, and explored the nuclear renaissance among tech giants.
• 38:28
  Speaker 1: Now we delve into the technical heart of nuclear power's
• 38:32
  Speaker 1: promise for AI infrastructure. When we examine the technical specifications
• 38:37
  Speaker 1: driving this nuclear AI convergence, the numbers are staggering. Traditional
• 38:42
  Speaker 1: data centres operated with CPUs consuming 150 to 200 watts
• 38:47
  Speaker 1: per chip, but today's AI workloads demand GPUs running at
• 38:51
  Speaker 1: 700 watts for current state-of-the-art models. The trajectory is even
• 38:55
  Speaker 1: more dramatic. Next generation chips expected in 2024 will consume 1,200 watts,
• 39:02
  Speaker 1: while Nvidia's Blackwell lineup pushes power consumption to unprecedented levels.
• 39:08
  Speaker 1: The GB200 NVL72 consumes 120 kilowatts, the upcoming GB300 racks
• 39:15
  Speaker 1: will consume 140 kilowatts, and the future Rubin generation is
• 39:19
  Speaker 1: expected to boost thermal design power by 50% to 180
• 39:22
  Speaker 1: kilowatts per rack. The upgraded Vera Rubin will double this
• 39:28
  Speaker 1: to $ 360. 360kW per rack by 2027, with NVIDIA ultimately
• 39:34
  Speaker 1: targeting supersized 1MW server racks representing 8 times more power
• 39:39
  Speaker 1: than current GB200 racks. This exponential growth in power requirements
• 39:44
  Speaker 1: creates a perfect storm for nuclear energy adoption. Power demand
• 39:49
  Speaker 1: for AI inference tasks is projected to increase at a
• 39:53
  Speaker 1: 122% compound annual growth rate through 2028, as providers work
• 39:58
  Speaker 1: to serve billions of requests and process hundreds of trillions
• 40:02
  Speaker 1: of tokens. The scale becomes clearer when we consider that
• 40:06
  Speaker 1: data centres currently have the capacity to consume 508 terawatt
• 40:12
  Speaker 1: hours of electricity annually if running constantly. greater than the
• 40:16
  Speaker 1: total annual electricity production for Italy or Australia. By 2034,
• 40:23
  Speaker 1: global consumption is expected to top 1,380 terawatt hours, about
• 40:29
  Speaker 1: as much as is used by all of India. The
• 40:32
  Speaker 1: technical demands of AI workloads create unique challenges that traditional
• 40:36
  Speaker 1: power sources struggle to meet. Unlike conventional computing, AI training
• 40:42
  Speaker 1: and inference require continuous, uninterrupted power supply. Any fluctuation or
• 40:48
  Speaker 1: outage can corrupt training runs that may have been running
• 40:51
  Speaker 1: for weeks or months, representing millions of dollars in lost
• 40:55
  Speaker 1: computation and time. This reliability requirement makes nuclear power particularly attractive,
• 41:03
  Speaker 1: as nuclear plants typically achieve capacity factors of 90% or
• 41:06
  Speaker 1: higher meaning they generate electricity at or near their maximum
• 41:11
  Speaker 1: capacity 90% of the time. The heat generated by these
• 41:17
  Speaker 1: massive AI chips creates another technical challenge. Modern GPU clusters
• 41:22
  Speaker 1: generate enormous amounts of waste heat that must be continuously
• 41:27
  Speaker 1: removed to prevent thermal throttling and equipment failure. This cooling
• 41:32
  Speaker 1: requirement can consume 30-50% of the total power budget for
• 41:36
  Speaker 1: an AI data centre. Nuclear plants, with their vast cooling
• 41:42
  Speaker 1: systems already in place, can potentially provide both electrical power
• 41:46
  Speaker 1: and cooling capacity through cogeneration systems, dramatically improving overall efficiency.
• 41:53
  Speaker 1: Small modular reactors represent a paradigm shift in nuclear technology
• 41:58
  Speaker 1: specifically suited to AI infrastructure needs. Unlike traditional nuclear plants
• 42:03
  Speaker 1: that generate 1,000 megawatts or more, SMRs are designed to
• 42:08
  Speaker 1: produce 20 to 300 megawatts, making them ideal for powering
• 42:13
  Speaker 1: individual data centres or small clusters. The modular design allows
• 42:18
  Speaker 1: for factory construction and transport to sites, reducing construction time
• 42:23
  Speaker 1: from the 6 to 12 years required for traditional reactors
• 42:27
  Speaker 1: to just 3 to 5 years. This timeline aligns better
• 42:32
  Speaker 1: with the rapid deployment schedules demanded by AI companies racing
• 42:36
  Speaker 1: to scale their infrastructure. The technical specifications of SMRs make
• 42:41
  Speaker 1: them particularly well suited for AI applications. Most SMR designs
• 42:47
  Speaker 1: incorporate passive safety systems that rely on natural forces like
• 42:51
  Speaker 1: gravity and convection rather than active systems requiring pumps and
• 42:56
  Speaker 1: external power. This inherent safety reduces the complexity of operations
• 43:02
  Speaker 1: and the potential for system failures that could interrupt power
• 43:06
  Speaker 1: supply to critical AI workloads. The smaller emergency planning zones
• 43:12
  Speaker 1: required for SMRs, typically two miles compared to 10 miles
• 43:17
  Speaker 1: for traditional reactors, also make them more suitable for deployment
• 43:21
  Speaker 1: near population centers, where data centers are often located. Advanced
• 43:27
  Speaker 1: reactor designs being developed specifically for industrial applications offer even
• 43:32
  Speaker 1: more promising technical solutions. High-temperature gas-cooled reactors can operate at
• 43:37
  Speaker 1: temperatures up to 950 degrees Celsius, enabling not only electricity generation,
• 43:43
  Speaker 1: but also process heat for industrial applications. Molten-salt reactors, like
• 43:50
  Speaker 1: those being developed by Kairos Power in partnership with Google,
• 43:54
  Speaker 1: operate at atmospheric pressure and can provide both electrical power
• 43:59
  Speaker 1: and high-temperature heat for cooling systems or other industrial processes.
• 44:05
  Speaker 1: The electrical characteristics of nuclear power align perfectly with AI
• 44:08
  Speaker 1: infrastructure requirements. Nuclear plants provide what electrical engineers call firm power,
• 44:15
  Speaker 1: meaning they can generate electricity on demand regardless of weather
• 44:18
  Speaker 1: conditions or time of day. This contrasts sharply with renewable
• 44:22
  Speaker 1: sources that provide variable power, dependent on wind and solar conditions.
• 44:28
  Speaker 1: While battery storage can help smooth out renewable intermittency, current
• 44:32
  Speaker 1: battery technologies can only store a few hours of energy,
• 44:36
  Speaker 1: far less than the days or weeks of continuous power
• 44:40
  Speaker 1: required for major AI training runs. The quality of nuclear
• 44:45
  Speaker 1: power also meets the demanding specifications of AI infrastructure. Nuclear
• 44:51
  Speaker 1: plants generate electricity with extremely stable frequency and voltage characteristics,
• 44:57
  Speaker 1: essential for the precise timing requirements of AI computations. The
• 45:02
  Speaker 1: massive rotating generators in nuclear plants also provide grid stability
• 45:06
  Speaker 1: services through their inherent inertia, helping to maintain electrical system
• 45:11
  Speaker 1: stability even during sudden load changes as AI systems ramp
• 45:16
  Speaker 1: up or down. From a grid integration perspective, nuclear power
• 45:21
  Speaker 1: offers unique advantages for AI infrastructure. Large AI data centres
• 45:28
  Speaker 1: can represent sudden, massive loads on the electrical grid potentially
• 45:32
  Speaker 1: causing stability issues for other customers. Nuclear plants can be
• 45:37
  Speaker 1: designed to provide dedicated power to AI facilities through private
• 45:42
  Speaker 1: transmission lines, isolating these loads from the broader grid. This
• 45:47
  Speaker 1: approach reduces the need for expensive grid upgrades while ensuring
• 45:51
  Speaker 1: that AI operations don't interfere with other electricity customers. The
• 45:57
  Speaker 1: technical economics of nuclear power for AI applications become compelling
• 46:03
  Speaker 1: when considering the total cost of ownership. While nuclear plants
• 46:08
  Speaker 1: require high upfront capital investment, their operating costs are very
• 46:12
  Speaker 1: low and predictable over decades. For AI companies planning infrastructure
• 46:17
  Speaker 1: investments with 20 to 30 year lifespans, the long-term cost
• 46:21
  Speaker 1: stability of nuclear power provides significant financial advantages over volatile
• 46:26
  Speaker 1: fossil fuel prices or the uncertain costs of renewable energy
• 46:30
  Speaker 1: plus storage. the waste heat generated by nuclear plants, typically
• 46:35
  Speaker 1: considered a liability, becomes an asset in AI applications. Modern
• 46:41
  Speaker 1: AI data centers require enormous amounts of cooling, often using
• 46:45
  Speaker 1: as much energy for cooling as for computation. Nuclear plants
• 46:50
  Speaker 1: could potentially provide both electricity and cooling through absorption chillers
• 46:54
  Speaker 1: powered by waste heat, dramatically improving overall system efficiency. Technical
• 46:59
  Speaker 1: challenges remain, however. The licensing and regulatory approval process for
• 47:03
  Speaker 1: nuclear reactors, even SMRs, involves extensive technical documentation and safety analysis.
• 47:11
  Speaker 1: The Nuclear Regulatory Commission requires detailed analysis of every system component,
• 47:17
  Speaker 1: safety procedure and emergency response plan. This process, while necessary
• 47:23
  Speaker 1: for safety, can take years and requires significant technical expertise
• 47:28
  Speaker 1: that many AI companies lack. The technical integration of nuclear
• 47:33
  Speaker 1: power with AI infrastructure also presents challenges. Nuclear plants are
• 47:40
  Speaker 1: designed for steady-state operation, while AI workloads can vary dramatically
• 47:45
  Speaker 1: depending on training schedules, inference demand and system maintenance. Advanced
• 47:51
  Speaker 1: control systems will be needed to manage the interaction between
• 47:54
  Speaker 1: nuclear power generation and AI power consumption. potentially including energy
• 47:59
  Speaker 1: storage systems to buffer load variations. Cybersecurity represents another critical
• 48:05
  Speaker 1: technical challenge. AI data centers are already prime targets for
• 48:09
  Speaker 1: cyberattacks due to the valuable intellectual property they contain. Nuclear
• 48:14
  Speaker 1: plants connected to AI facilities would need robust cybersecurity measures
• 48:19
  Speaker 1: to prevent attacks that could compromise both nuclear safety and
• 48:23
  Speaker 1: AI operations. The technical complexity of securing both nuclear control
• 48:28
  Speaker 1: systems and AI infrastructure simultaneously requires new approaches to cybersecurity
• 48:35
  Speaker 1: that are still being developed. The technical promise of nuclear
• 48:38
  Speaker 1: power for AI infrastructure is substantial, but realizing this potential
• 48:44
  Speaker 1: requires overcoming significant engineering and regulatory challenges. The convergence of
• 48:50
  Speaker 1: nuclear technology and artificial intelligence represents one of the most
• 48:55
  Speaker 1: ambitious technical undertakings of our time, with implications extending far
• 49:00
  Speaker 1: beyond just powering computers to fundamentally reshaping how we generate
• 49:05
  Speaker 1: and use energy in the digital age. The mathematics of
• 49:09
  Speaker 1: nuclear power generation reveal why this technology holds such promise
• 49:13
  Speaker 1: for artificial intelligence applications. A single uranium fuel pellet, no
• 49:18
  Speaker 1: larger than a fingertip, contains as much energy as a
• 49:21
  Speaker 1: ton of coal. This energy density becomes absolutely crucial when
• 49:26
  Speaker 1: we consider the voracious appetite of modern AI systems. The
• 49:31
  Speaker 1: latest generation of AI training runs require sustained power delivery,
• 49:35
  Speaker 1: measured not in kilowatts, but in megawatts, operating continuously for
• 49:41
  Speaker 1: weeks or months at a time. Traditional power sources struggle
• 49:46
  Speaker 1: with this combination of scale and consistency. Solar panels produce
• 49:51
  Speaker 1: zero electricity at night, precisely when many AI training operations
• 49:56
  Speaker 1: prefer to run due to cooler ambient temperatures. Wind turbines
• 50:00
  Speaker 1: sit idle during calm weather, potentially interrupting critical computational processes
• 50:06
  Speaker 1: that cannot simply be paused and restarted. but nuclear reactors
• 50:10
  Speaker 1: operate with the steadiness of a metronome, delivering consistent baseload
• 50:14
  Speaker 1: power regardless of weather conditions, time of day, or seasonal variations.
• 50:20
  Speaker 1: The reliability factor becomes even more critical when we examine
• 50:23
  Speaker 1: the true cost of power interruptions in AI operations. A
• 50:28
  Speaker 1: single outage during a large language model training run can
• 50:31
  Speaker 1: waste weeks of computational work and millions of dollars in
• 50:35
  Speaker 1: lost progress. These training processes cannot simply resume where they
• 50:41
  Speaker 1: left off like a paused video. The intricate mathematical relationships
• 50:46
  Speaker 1: between billions of parameters require continuous, uninterrupted processing to maintain
• 50:52
  Speaker 1: computational integrity. This is where small modular reactors enter the
• 50:56
  Speaker 1: picture as potential game-changers for the AI industry. Unlike the
• 51:01
  Speaker 1: massive nuclear plants of previous generations, SMRs offer a fundamentally
• 51:06
  Speaker 1: different approach to nuclear power generation. Think of traditional nuclear
• 51:10
  Speaker 1: plants as massive cruise ships, powerful but requiring enormous infrastructure
• 51:15
  Speaker 1: and years of construction. SMRs are more like sleek speedboats,
• 51:20
  Speaker 1: smaller and more manoeuvrable, but still carrying the fundamental advantages
• 51:25
  Speaker 1: of nuclear technology. The modular design philosophy behind SMRs represents
• 51:31
  Speaker 1: a manufacturing revolution in nuclear technology. Rather than constructing each
• 51:37
  Speaker 1: reactor as a unique engineering project, SMRs can be mass-produced
• 51:42
  Speaker 1: in factories, then transported to their installation sites. This factory-based
• 51:48
  Speaker 1: approach promises to dramatically reduce both construction time and costs
• 51:54
  Speaker 1: while improving quality control and safety standards. For AI applications,
• 52:00
  Speaker 1: the sizing of SMRs proves particularly advantageous. Most SMR designs
• 52:06
  Speaker 1: target power outputs between 50 and 300 megawatts, perfectly matching
• 52:11
  Speaker 1: the power requirements of large data centers or AI training facilities.
• 52:17
  Speaker 1: This means a single SMR could potentially power an entire
• 52:20
  Speaker 1: AI campus, providing the dedicated, reliable energy supply that these
• 52:25
  Speaker 1: computational powerhouses demand. The geographic flexibility of SMRs also aligns
• 52:32
  Speaker 1: well with AI industry needs. Traditional nuclear plants require massive
• 52:37
  Speaker 1: cooling systems, often necessitating locations near large bodies of water. SMRs,
• 52:44
  Speaker 1: with their smaller size and advanced cooling technologies, can be
• 52:47
  Speaker 1: deployed in a much wider range of locations. This flexibility
• 52:52
  Speaker 1: allows AI companies to locate their facilities closer to population centres,
• 52:57
  Speaker 1: reducing data transmission delays, or in regions with favourable climates
• 53:01
  Speaker 1: for cooling computational equipment. Safety considerations, always paramount in nuclear technology,
• 53:08
  Speaker 1: have been fundamentally reconsidered in SMR design. Many SMR concepts
• 53:14
  Speaker 1: incorporate passive safety systems that rely on physics rather than
• 53:19
  Speaker 1: human intervention or mechanical systems. If something goes wrong, the
• 53:23
  Speaker 1: reactor naturally shuts down and cools itself without requiring external
• 53:28
  Speaker 1: power or operator action. This inherent safety approach provides an
• 53:32
  Speaker 1: additional layer of security for AI facilities, where operational continuity
• 53:38
  Speaker 1: is absolutely critical. The economic argument for nuclear-powered AI becomes
• 53:45
  Speaker 1: even more compelling when we consider the long-term trends in
• 53:48
  Speaker 1: both industries. AI computational requirements continue to grow exponentially with
• 53:55
  Speaker 1: each new generation of models requiring significantly more training power
• 54:00
  Speaker 1: than the last. Simultaneously, the push for carbon-neutral operations intensifies
• 54:07
  Speaker 1: across the technology sector. Nuclear power offers a path to
• 54:12
  Speaker 1: meet both challenges simultaneously, providing abundant clean energy that can
• 54:16
  Speaker 1: scale with AI's growing demands. The construction timeline for SMRs
• 54:21
  Speaker 1: also aligns better with the rapid development cycles of the
• 54:24
  Speaker 1: AI industry. While traditional nuclear plants require a decade or
• 54:28
  Speaker 1: more from conception to operation, SMRs promise construction timelines of
• 54:33
  Speaker 1: three to five years. This compressed timeline allows AI companies
• 54:39
  Speaker 1: to plan nuclear-powered facilities that can come online in sync
• 54:42
  Speaker 1: with their computational expansion plans. International competition adds another layer
• 54:48
  Speaker 1: of urgency to the nuclear AI equation. Countries and companies
• 54:53
  Speaker 1: that successfully deploy nuclear-powered AI infrastructure will gain significant advantages
• 54:59
  Speaker 1: in the global AI race. The ability to train larger
• 55:03
  Speaker 1: models more efficiently, operate AI systems more cost-effectively, and maintain
• 55:08
  Speaker 1: operations independent of grid constraints could determine which nations lead
• 55:13
  Speaker 1: in artificial intelligence development. China's aggressive expansion of both nuclear
• 55:19
  Speaker 1: power and AI capabilities illustrates this strategic thinking. The Chinese
• 55:25
  Speaker 1: government views nuclear-powered data centres as critical infrastructure for maintaining
• 55:30
  Speaker 1: competitiveness in the global AI landscape. Meanwhile, American companies are
• 55:37
  Speaker 1: exploring similar approaches, recognising that energy independence and reliability could
• 55:42
  Speaker 1: become decisive factors in AI leadership. the symbiotic relationship between
• 55:48
  Speaker 1: nuclear power and AI extends beyond simple energy provision. AI
• 55:54
  Speaker 1: systems can potentially optimize nuclear plant operations, predicting maintenance needs,
• 56:00
  Speaker 1: optimizing fuel usage, and enhancing safety monitoring. This creates a
• 56:06
  Speaker 1: positive feedback loop where nuclear power enables more sophisticated AI development,
• 56:11
  Speaker 1: which in turn improves nuclear plant efficiency and safety. However,
• 56:16
  Speaker 1: the path forward is not without significant challenges. Regulatory frameworks
• 56:21
  Speaker 1: designed for traditional nuclear plants may not easily accommodate the smaller,
• 56:26
  Speaker 1: more distributed model that SMRs represent. Public acceptance of nuclear
• 56:31
  Speaker 1: technology remains mixed, despite decades of safe operation in many countries.
• 56:37
  Speaker 1: The initial capital costs for SMR deployment, while lower than
• 56:41
  Speaker 1: traditional nuclear plants, still represent substantial investments that require careful
• 56:47
  Speaker 1: financial planning. The integration of nuclear power with AI infrastructure
• 56:52
  Speaker 1: also raises new technical questions. How do we design data
• 56:57
  Speaker 1: centers that can take full advantage of nuclear power's characteristics?
• 57:02
  Speaker 1: What backup systems are necessary when primary power is already
• 57:06
  Speaker 1: highly reliable? How do we balance the need for energy
• 57:09
  Speaker 1: security with the flexibility that modern AI operations require? These challenges,
• 57:16
  Speaker 1: while significant, are not insurmountable. The convergence of nuclear technology
• 57:22
  Speaker 1: and artificial intelligence represents one of the most promising pathways
• 57:27
  Speaker 1: towards sustainable, scalable AI development. As both technologies continue to
• 57:33
  Speaker 1: mature and costs continue to decline, the nuclear AI combination
• 57:39
  Speaker 1: may well become the foundation for the next generation of
• 57:42
  Speaker 1: artificial intelligence capabilities. The story of nuclear power and AI
• 57:48
  Speaker 1: is still being written, but the early chapters suggest a
• 57:51
  Speaker 1: future where clean, abundant energy enables unprecedented advances in artificial intelligence,
• 57:59
  Speaker 1: while AI technologies help unlock the full potential of nuclear
• 58:03
  Speaker 1: power generation. This partnership could define the next era of
• 58:07
  Speaker 1: technological progress, creating possibilities we can barely imagine today. Looking
• 58:13
  Speaker 1: at the nuclear energy solution for AI's unprecedented power demands,
• 58:17
  Speaker 1: we've witnessed a remarkable transformation in how technology giants approach
• 58:22
  Speaker 1: energy procurement. The statistics are staggering yet sobering. Global data
• 58:30
  Speaker 1: center electricity consumption reached 415 terawatt hours in 2024, representing
• 58:33
  Speaker 1: roughly 1.5% of all electricity consumed worldwide. but here's where
• 58:43
  Speaker 1: the numbers become truly eye-opening. This consumption is projected to
• 58:48
  Speaker 1: more than double by 2030, reaching nearly 945 terawatt hours.
• 58:54
  Speaker 1: To put this in perspective, that's equivalent to the entire
• 58:57
  Speaker 1: annual electricity production of Italy or Australia being consumed just
• 59:01
  Speaker 1: by data centres. The driving force behind this exponential growth
• 59:08
  Speaker 1: lies in the fundamental shift from traditional computing to artificial
• 59:12
  Speaker 1: intelligence processing. Where conventional data centres relied on CPUs consuming
• 59:18
  Speaker 1: 150 to 200 watts per chip, the new generation of
• 59:22
  Speaker 1: AI hardware demands significantly more power. Current state-of-the-art GPU models
• 59:28
  Speaker 1: run at 700 watts, while 2024's next-generation chips are expected
• 59:34
  Speaker 1: to consume 1,200 watts each. NVIDIA's latest Blackwell lineup pushes
• 59:40
  Speaker 1: these boundaries even further, with the GB200 NVL72 consuming 120
• 59:43
  Speaker 1: kilowatts and the upcoming GB300 RACs demanding 140 kilowatts. The
• 59:46
  Speaker 1: trajectory is clear. NVIDIA's future Rubin generation will boost thermal
• 59:57
  Speaker 1: design power by 50% to 180 kW per rack, with
• 1:00:01
  Speaker 1: the upgraded Vera Rubin doubling this to 360 kW by 2027.
• 1:00:07
  Speaker 1: This escalation isn't merely about individual components becoming more power-hungry.
• 1:00:12
  Speaker 1: The entire architecture of AI computing is transforming. Power demand
• 1:00:18
  Speaker 1: for inference tasks, the process of AI models making predictions
• 1:00:23
  Speaker 1: or generating responses, is projected to increase at a compound
• 1:00:27
  Speaker 1: annual growth rate of 122% through 2028. This means the
• 1:00:33
  Speaker 1: electricity required to serve billions of AI requests and process
• 1:00:38
  Speaker 1: hundreds of trillions of tokens will grow exponentially, far outpacing
• 1:00:43
  Speaker 1: the capacity of traditional renewable energy sources to provide consistent,
• 1:00:48
  Speaker 1: reliable power. The challenge extends beyond raw power consumption to
• 1:00:54
  Speaker 1: the nature of AI workloads themselves, unlike traditional computing tasks
• 1:01:00
  Speaker 1: that can be scheduled during off-peak hours or distributed across
• 1:01:03
  Speaker 1: different time zones, AI applications demand constant availability. Training. Large
• 1:01:12
  Speaker 1: language models requires weeks or months of continuous processing, with
• 1:01:17
  Speaker 1: interruptions potentially invalidating entire training runs. Inference serving must respond
• 1:01:23
  Speaker 1: to user requests within milliseconds, 24 hours a day, 7
• 1:01:28
  Speaker 1: days a week. This creates a fundamental mismatch with renewable
• 1:01:32
  Speaker 1: energy sources that depend on weather patterns and daylight cycles.
• 1:01:37
  Speaker 1: Solar power, despite its dramatic cost reductions over the past decade,
• 1:01:41
  Speaker 1: remains inherently intermittent. Even the most advanced solar installations can
• 1:01:46
  Speaker 1: only generate electricity during daylight hours. and their output varies
• 1:01:51
  Speaker 1: significantly with weather conditions, seasonal changes and geographical location. Wind
• 1:01:58
  Speaker 1: energy faces similar challenges, with output fluctuating based on atmospheric
• 1:02:03
  Speaker 1: conditions that can be unpredictable over short time periods. While
• 1:02:07
  Speaker 1: these renewable sources can theoretically provide much of the electricity
• 1:02:11
  Speaker 1: needed for AI operations, their intermittency creates critical gaps that
• 1:02:15
  Speaker 1: must be filled by other power sources. Battery storage technology,
• 1:02:21
  Speaker 1: often proposed as the solution to renewable intermittency, faces significant
• 1:02:25
  Speaker 1: limitations when applied to AI data center requirements. Current battery
• 1:02:31
  Speaker 1: energy storage systems typically store one to two hours of energy,
• 1:02:35
  Speaker 1: designed primarily for grid stabilization and peak shaving, rather than
• 1:02:40
  Speaker 1: extended backup power. Industry experts suggest that storage technologies in
• 1:02:46
  Speaker 1: the 36 to 48 hour range would be most practical
• 1:02:50
  Speaker 1: for data center operators. But even these extended storage solutions
• 1:02:55
  Speaker 1: cannot address the seasonal variations in renewable energy production that
• 1:03:00
  Speaker 1: occur in many regions. The economic implications are equally challenging.
• 1:03:06
  Speaker 1: While the levelized cost of electricity from solar and wind
• 1:03:10
  Speaker 1: has decreased substantially, the total cost of renewable energy systems,
• 1:03:15
  Speaker 1: including storage, remains significantly higher than baseload power sources. Moreover,
• 1:03:22
  Speaker 1: the land requirements for renewable installations capable of powering large
• 1:03:26
  Speaker 1: AI data centers are enormous. A single hyperscale data center
• 1:03:32
  Speaker 1: consuming 100 megawatts of power would require approximately 200 to
• 1:03:37
  Speaker 1: 400 square kilometers of solar panels, depending on geographic location
• 1:03:42
  Speaker 1: and efficiency assumptions. These constraints have led major technology companies
• 1:03:47
  Speaker 1: to fundamentally reconsider their energy strategies. Microsoft's groundbreaking agreement to
• 1:03:54
  Speaker 1: restart Three Mile Island represents more than just a power
• 1:03:58
  Speaker 1: purchase agreement. It's a strategic recognition that AI's energy demands
• 1:04:03
  Speaker 1: require a different approach to electricity procurement. The reactor that
• 1:04:07
  Speaker 1: will be restarted was not involved in the 1979 accident,
• 1:04:11
  Speaker 1: and will provide 835 megawatts of electricity exclusively to Microsoft
• 1:04:17
  Speaker 1: for 20 years. This represents enough power to supply approximately 700,000 homes,
• 1:04:24
  Speaker 1: but Microsoft will dedicate this entire output to supporting its
• 1:04:28
  Speaker 1: AI operations. The financial commitment is substantial. Constellation Energy will
• 1:04:34
  Speaker 1: invest approximately $ 1. 6 billion over four years to restart the facility,
• 1:04:40
  Speaker 1: With the project dependent on federal subsidies in the form
• 1:04:43
  Speaker 1: of tax breaks earmarked for nuclear power in the 2022
• 1:04:47
  Speaker 1: Inflation Reduction Act, this marks an unprecedented milestone in nuclear
• 1:04:52
  Speaker 1: energy history. Never before has a decommissioned nuclear plant been
• 1:04:57
  Speaker 1: brought back into service, and never before has all of
• 1:05:00
  Speaker 1: a single commercial nuclear power plant's output been allocated to
• 1:05:05
  Speaker 1: a single company. Amazon has taken an even more comprehensive
• 1:05:09
  Speaker 1: approach to nuclear energy investment. The company has signed three
• 1:05:13
  Speaker 1: separate agreements to develop nuclear power projects, including partnerships with
• 1:05:18
  Speaker 1: Energy Northwest and X-Energy to build several small modular reactors
• 1:05:22
  Speaker 1: in Washington state and an agreement with Dominion Energy in
• 1:05:26
  Speaker 1: Virginia to explore SMR development. Amazon's partnership with Energy Northwest
• 1:05:33
  Speaker 1: will develop four SMRs starting with 320 megawatts and potentially
• 1:05:38
  Speaker 1: expanding to 960 megawatts. Additionally, Amazon led a $ 500 million
• 1:05:44
  Speaker 1: financing round to support X Energy's advanced SMR designs for
• 1:05:49
  Speaker 1: up to 5 gigawatts of nuclear energy, with plans to
• 1:05:53
  Speaker 1: build multiple SMRs producing at least 5 gigawatts total by 2039.
• 1:05:59
  Speaker 1: Google has pursued a strategic partnership approach with nuclear energy
• 1:06:02
  Speaker 1: developers the company signed a corporate agreement with Kairos Power
• 1:06:06
  Speaker 1: to use multiple SMRs for 24-7 carbon-free energy to power
• 1:06:11
  Speaker 1: Google's data centres and offices. The first SMR is set
• 1:06:16
  Speaker 1: to launch by 2030, with additional units added through 2035,
• 1:06:21
  Speaker 1: supplying up to 500 megawatts total. Kairos Power's reactors use
• 1:06:27
  Speaker 1: a molten salt cooling system and ceramic pebble-like fuel. creating
• 1:06:32
  Speaker 1: a low-pressure nuclear solution that offers enhanced safety characteristics compared
• 1:06:37
  Speaker 1: to traditional reactor designs. The appeal of nuclear energy for
• 1:06:43
  Speaker 1: AI applications extends beyond mere power generation capacity. Nuclear plants
• 1:06:50
  Speaker 1: provide what engineers call baseload power, electricity, that can be
• 1:06:54
  Speaker 1: generated continuously regardless of weather conditions, time of day or
• 1:06:58
  Speaker 1: seasonal variations. This reliability is crucial for AI operations that
• 1:07:04
  Speaker 1: cannot tolerate interruptions. Moreover, nuclear plants typically operate at capacity
• 1:07:11
  Speaker 1: factors above 90%, meaning they generate electricity at or near
• 1:07:15
  Speaker 1: their maximum capacity more than 90% of the time. This
• 1:07:20
  Speaker 1: compares favourably to solar installations, which typically achieve capacity factors
• 1:07:24
  Speaker 1: of 15-25%, and wind farms, which averaged 25 to 35%
• 1:07:31
  Speaker 1: capacity factors. The environmental benefits of nuclear energy for AI
• 1:07:36
  Speaker 1: applications are equally compelling. Nuclear power plants produce electricity without
• 1:07:42
  Speaker 1: greenhouse gas emissions during operation, making them an attractive option
• 1:07:45
  Speaker 1: for technology companies committed to carbon neutrality goals. Google's carbon
• 1:07:51
  Speaker 1: emissions have risen 48% since 2019, mostly due to data
• 1:07:53
  Speaker 1: center energy consumption while Microsoft's carbon emissions were almost 31%
• 1:08:01
  Speaker 1: higher in 2024 than in 2020, primarily due to building
• 1:08:06
  Speaker 1: data centers for AI and manufacturing AI hardware like semiconductors
• 1:08:10
  Speaker 1: and servers. However, the nuclear renaissance in AI energy procurement
• 1:08:16
  Speaker 1: faces significant challenges. The cost of building new nuclear power
• 1:08:20
  Speaker 1: plants has increased nearly 20% above expectations due to construction
• 1:08:25
  Speaker 1: delays and regulatory complexities. Nuclear projects have become increasingly complex
• 1:08:31
  Speaker 1: to execute, inducing delays and higher costs, particularly due to
• 1:08:36
  Speaker 1: safety and regulatory considerations following the 2011 Fukushima accident. Three-quarters
• 1:08:44
  Speaker 1: of recent nuclear construction projects have experienced significant delays, with
• 1:08:49
  Speaker 1: newer reactors averaging 60% higher costs than earlier ones. Small
• 1:08:54
  Speaker 1: modular reactors represent a potential solution to these construction challenges.
• 1:09:00
  Speaker 1: SMRs are advanced nuclear reactors with power capacities up to
• 1:09:03
  Speaker 1: 300 megawatts electric, about one-third the capacity of traditional nuclear
• 1:09:08
  Speaker 1: power reactors. Their modular design allows for factory construction and
• 1:09:13
  Speaker 1: transportation to installation sites, potentially reducing construction timelines from the
• 1:09:18
  Speaker 1: 6 to 12 years required for traditional reactors to 3
• 1:09:23
  Speaker 1: to 5 years for SMRs. Most SMRs under development are
• 1:09:28
  Speaker 1: projected to cost less than $ 2 billion compared to more
• 1:09:32
  Speaker 1: than $ 10 billion for traditional nuclear plants. The technical advantages
• 1:09:38
  Speaker 1: of SMRs for AI applications are numerous. Their smaller scale
• 1:09:43
  Speaker 1: makes them suitable for placement closer to data centers reducing
• 1:09:47
  Speaker 1: transmission losses and improving grid stability. SMRs can be deployed
• 1:09:52
  Speaker 1: in configurations that match data center power requirements more precisely
• 1:09:57
  Speaker 1: than large traditional reactors. Their advanced safety features, including passive
• 1:10:03
  Speaker 1: cooling systems and walk-away safe designs, address many of the
• 1:10:07
  Speaker 1: concerns that have historically limited nuclear deployment. Government support for
• 1:10:12
  Speaker 1: nuclear energy in AI applications is growing. the Biden administration
• 1:10:17
  Speaker 1: and Congress have provided billions in funding to prevent old
• 1:10:20
  Speaker 1: plants from shutting down and to support new nuclear development.
• 1:10:25
  Speaker 1: The U.S. government has provided more than $ 600 million to
• 1:10:28
  Speaker 1: companies for SMR development since 2014, with the Department of
• 1:10:33
  Speaker 1: Energy's Gateway for Accelerated Innovation in Nuclear Programme awarding $ 31. 8
• 1:10:38
  Speaker 1: million in funding grants to 52 private companies since 2016.
• 1:10:46
  Speaker 1: The regulatory framework is also evolving to accommodate new nuclear technologies.
• 1:10:52
  Speaker 1: The Nuclear Regulatory Commission approved the first SMR design in
• 1:10:55
  Speaker 1: the United States, NuScale's Advanced Lightwater SMR, generating 77 megawatts electric.
• 1:11:03
  Speaker 1: Standard Power has announced plans to use NuScale's SMRs for
• 1:11:08
  Speaker 1: two nuclear plants in Ohio and Pennsylvania, providing nearly 2
• 1:11:13
  Speaker 1: gigawatts by 2029. SMRs require much less space than traditional
• 1:11:19
  Speaker 1: nuclear plants and don't need the 10-mile emergency planning zone
• 1:11:23
  Speaker 1: that traditional nuclear plants require, making them potentially more acceptable
• 1:11:27
  Speaker 1: to communities. The global nuclear market is experiencing significant shifts
• 1:11:34
  Speaker 1: as countries and companies race to meet AI energy demands.
• 1:11:39
  Speaker 1: China and the United States are the most significant regions
• 1:11:43
  Speaker 1: for data centre electricity consumption growth accounting for nearly 80%
• 1:11:48
  Speaker 1: of global growth to 2030. Russia dominates the international nuclear market,
• 1:11:54
  Speaker 1: with 24 units under construction, while China leads domestically with
• 1:11:59
  Speaker 1: 23 ongoing reactors. Spending on small modular reactors could grow
• 1:12:04
  Speaker 1: from $ 5 billion today to $ 25 billion by 2030, and
• 1:12:06
  Speaker 1: reach $ 670 billion by 2050. potentially delivering 190 gigawatts of
• 1:12:15
  Speaker 1: SMR power by 2050. The financial markets are responding positively
• 1:12:21
  Speaker 1: to nuclear energy investments tied to AI demand. Constellation's shares
• 1:12:27
  Speaker 1: rose more than 20% following the Three Mile Island announcement
• 1:12:31
  Speaker 1: and have increased more than 100% year-to-date. Oklo has seen
• 1:12:37
  Speaker 1: its shares climb over 200% in recent months. The global
• 1:12:41
  Speaker 1: market for SMRs for data centers is projected to reach
• 1:12:45
  Speaker 1: $ 278 million by 2033, growing at a compound annual growth
• 1:12:51
  Speaker 1: rate of 48.72%. Yet significant challenges remain. The integration of
• 1:12:58
  Speaker 1: nuclear power for AI applications presents complex infrastructure challenges. Data
• 1:13:03
  Speaker 1: centers concentrate in specific locations, making their grid integration potentially
• 1:13:09
  Speaker 1: more challenging than distributed loads. The ageing power grid in
• 1:13:13
  Speaker 1: many regions struggles to meet current demand, with at least
• 1:13:17
  Speaker 1: 38 gigawatts of peak demand growth expected through 2028, driven
• 1:13:23
  Speaker 1: by data centres and industrial facilities. The ultimate question of
• 1:13:28
  Speaker 1: whether nuclear energy is the solution to endless artificial intelligence
• 1:13:32
  Speaker 1: power demands cannot be answered with certainty. However, the convergence
• 1:13:39
  Speaker 1: of AI's unique power requirements the limitations of renewable energy
• 1:13:43
  Speaker 1: intermittency and the characteristics of nuclear power generation suggests that
• 1:13:49
  Speaker 1: nuclear energy will play an increasingly important role in powering
• 1:13:53
  Speaker 1: artificial intelligence operations. The investments by major technology companies, government
• 1:13:59
  Speaker 1: support for nuclear development and advances in small modular reactor
• 1:14:04
  Speaker 1: technology all point toward a future where nuclear energy provides
• 1:14:09
  Speaker 1: the reliable, carbon-free baseload power that AI systems require. As
• 1:14:15
  Speaker 1: artificial intelligence continues to transform every aspect of human society,
• 1:14:20
  Speaker 1: from healthcare and education to transportation and communication, ensuring adequate
• 1:14:25
  Speaker 1: power supply becomes not just a technical challenge, but a
• 1:14:28
  Speaker 1: strategic imperative. The partnership between AI and nuclear energy may
• 1:14:33
  Speaker 1: well prove to be one of the most important technological
• 1:14:36
  Speaker 1: relationships of the 21st century, enabling sustainable artificial intelligence development
• 1:14:42
  Speaker 1: while revitalizing nuclear technology for a new generation of applications.
• 1:14:48
  Speaker 1: While technological advancement has revolutionized virtually every aspect of human existence,
• 1:14:54
  Speaker 1: one fundamental challenge continues to constrain our digital future. The
• 1:14:59
  Speaker 1: massive and exponentially growing energy demands of artificial intelligence. As
• 1:15:05
  Speaker 1: we stand at the precipice of an AI revolution, that
• 1:15:09
  Speaker 1: promises to transform everything from medical diagnosis to climate modelling,
• 1:15:13
  Speaker 1: we face an uncomfortable reality that threatens to derail our
• 1:15:16
  Speaker 1: technological ambitions before they can fully materialise. The numbers are
• 1:15:21
  Speaker 1: staggering and sobering. Data centres currently consume about 4.4% of
• 1:15:27
  Speaker 1: total United States electricity, but this figure is projected to
• 1:15:31
  Speaker 1: potentially triple by 2030. reaching somewhere between 6.7 and 12%
• 1:15:37
  Speaker 1: of our nation's entire electrical grid. To put this in perspective,
• 1:15:42
  Speaker 1: if all data centres were running at full capacity today,
• 1:15:45
  Speaker 1: they would consume 508 terawatt hours of electricity annually, more
• 1:15:50
  Speaker 1: than the entire annual electricity production of countries like Italy
• 1:15:54
  Speaker 1: or Australia. By 2034, global data centre consumption is expected
• 1:16:00
  Speaker 1: to top roughly equivalent to the total energy consumption of India,
• 1:16:08
  Speaker 1: a nation of 1.4 billion people. This exponential growth stems
• 1:16:13
  Speaker 1: from a fundamental shift in computing architecture. Traditional data centers
• 1:16:18
  Speaker 1: relied primarily on central processing units that consumed between 150
• 1:16:22
  Speaker 1: and 200 watts per chip. However, The graphics processing units
• 1:16:28
  Speaker 1: that power modern AI applications operate at dramatically higher power levels.
• 1:16:33
  Speaker 1: Current state-of-the-art models run at 700 watts, while next-generation chips
• 1:16:38
  Speaker 1: expected in 2024 will consume 1,200 watts each. NVIDIA's upcoming
• 1:16:44
  Speaker 1: Blackwell lineup pushes these boundaries even further, with individual server
• 1:16:49
  Speaker 1: racks consuming between 120 and 140 kilowatts. Looking ahead, the
• 1:16:56
  Speaker 1: company is targeting server racks that consume a full megawatt
• 1:16:59
  Speaker 1: of power, representing an eightfold increase over current systems. The
• 1:17:05
  Speaker 1: driving force behind this energy explosion is the unprecedented computational
• 1:17:10
  Speaker 1: requirements of artificial intelligence training and inference. Power demand for
• 1:17:15
  Speaker 1: AI inference tasks alone is projected to increase at a
• 1:17:19
  Speaker 1: compound annual growth rate of 122% through 2028. This isn't
• 1:17:25
  Speaker 1: simply about running more computers, it's about running fundamentally different
• 1:17:30
  Speaker 1: types of computations that require massive parallel processing capabilities to
• 1:17:35
  Speaker 1: handle the complex mathematical operations underlying machine learning algorithms. Traditional
• 1:17:41
  Speaker 1: renewable energy sources, while crucial for our climate goals, face
• 1:17:46
  Speaker 1: significant limitations in meeting AI's unique energy requirements. Solar and
• 1:17:51
  Speaker 1: wind power are inherently intermittent. producing electricity only when the
• 1:17:55
  Speaker 1: sun shines or the wind blows. While battery storage systems
• 1:17:59
  Speaker 1: can help smooth out these variations, current storage technologies are expensive,
• 1:18:04
  Speaker 1: require substantial physical footprints, and typically provide only one to
• 1:18:08
  Speaker 1: two hours of backup power. For AI data centers that
• 1:18:12
  Speaker 1: must operate continuously around the clock, this intermittency creates substantial
• 1:18:18
  Speaker 1: operational challenges. Industry experts estimate that wind and solar could
• 1:18:22
  Speaker 1: potentially serve roughly 80% of a data center's power demand
• 1:18:26
  Speaker 1: when paired with storage systems. But some form of reliable
• 1:18:30
  Speaker 1: baseload generation is essential to meet the constant 24-7 demand
• 1:18:33
  Speaker 1: that AI applications require. This is where nuclear energy emerges
• 1:18:38
  Speaker 1: as a compelling solution, offering several unique advantages that align
• 1:18:42
  Speaker 1: perfectly with AI's energy requirements. Nuclear power plants can deliver
• 1:18:47
  Speaker 1: hundreds of megawatts of continuous carbon electricity without the variability
• 1:18:52
  Speaker 1: that characterizes renewable sources. Unlike coal or natural gas plants,
• 1:18:58
  Speaker 1: nuclear facilities produce no greenhouse gas emissions during operation, making
• 1:19:03
  Speaker 1: them an environmentally responsible choice for powering our AI future.
• 1:19:08
  Speaker 1: Perhaps most importantly, nuclear plants can operate at extremely high
• 1:19:12
  Speaker 1: capacity factors, typically running at 90% or higher availability, providing
• 1:19:17
  Speaker 1: the consistent, reliable power that AI systems demand. The technology
• 1:19:23
  Speaker 1: industry has taken notice of these advantages, with major companies
• 1:19:27
  Speaker 1: making unprecedented commitments to nuclear energy. Microsoft's agreement to restart
• 1:19:33
  Speaker 1: Three Mile Island represents a watershed moment in nuclear energy history.
• 1:19:39
  Speaker 1: Never before has a United States nuclear plant been brought
• 1:19:43
  Speaker 1: back into service after being decommissioned, and never before has
• 1:19:47
  Speaker 1: all of a single commercial nuclear power plant's output been
• 1:19:51
  Speaker 1: allocated to a single company. Under this 20-year agreement, Microsoft
• 1:19:57
  Speaker 1: will purchase energy from the restarted plant, which will provide
• 1:20:01
  Speaker 1: 835 megawatts of electricity enough to power approximately 700,000 homes.
• 1:20:10
  Speaker 1: Amazon has made perhaps the most comprehensive nuclear energy commitment
• 1:20:14
  Speaker 1: among technology giants, signing three separate agreements to develop nuclear
• 1:20:20
  Speaker 1: power projects. The company is partnering with Energy Northwest to
• 1:20:24
  Speaker 1: develop four small modular reactors, starting with 320 megawatts of
• 1:20:29
  Speaker 1: capacity and potentially expanding to 960 megawatts. Additionally, Amazon led
• 1:20:36
  Speaker 1: a $ 500 million financing round to support X-Energy's development of
• 1:20:41
  Speaker 1: advanced reactor designs, with plans to build multiple facilities, producing
• 1:20:46
  Speaker 1: at least 5 gigawatts of total nuclear capacity by 2039.
• 1:20:51
  Speaker 1: Google has taken a strategic approach through its partnership with
• 1:20:54
  Speaker 1: Kairos Power, signing a corporate agreement to use multiple small
• 1:20:58
  Speaker 1: modular reactors for 24-7 carbon-free energy. The first reactor is
• 1:21:04
  Speaker 1: set to launch by 2030, with additional units added through 2035,
• 1:21:10
  Speaker 1: ultimately supplying up to 500 megawatts of power. Kairos Power's
• 1:21:15
  Speaker 1: innovative design uses a molten fluoride salt cooling system and ceramic,
• 1:21:20
  Speaker 1: pebble-like fuel, creating a low-pressure nuclear solution that offers enhanced
• 1:21:26
  Speaker 1: safety characteristics. These corporate commitments reflect a broader recognition that
• 1:21:31
  Speaker 1: small modular reactors represent a paradigm shift in nuclear technology.
• 1:21:38
  Speaker 1: Unlike traditional nuclear plants that require massive infrastructure investments and
• 1:21:43
  Speaker 1: construction timelines of six to 12 years, SMRs are designed
• 1:21:47
  Speaker 1: to be manufactured in factories and assembled on site, reducing
• 1:21:50
  Speaker 1: construction time to three to five years. With power capacities
• 1:21:55
  Speaker 1: ranging from 20 to 300 megawatts, they can be precisely
• 1:21:59
  Speaker 1: sized to meet specific data center requirements without the need
• 1:22:03
  Speaker 1: for massive transmission infrastructure. The modular nature of these systems
• 1:22:08
  Speaker 1: offers unprecedented flexibility. Data center operators can start with a
• 1:22:14
  Speaker 1: single reactor and add additional units as their power requirements grow,
• 1:22:19
  Speaker 1: creating a scalable solution that can evolve with technological demands.
• 1:22:24
  Speaker 1: Most SMRs under development are expected to cost less than
• 1:22:28
  Speaker 1: $ 2 billion compared to more than $ 10 billion for traditional
• 1:22:32
  Speaker 1: nuclear plants, making them more financially accessible for private companies.
• 1:22:37
  Speaker 1: Safety represents another crucial advantage of modern nuclear technology. Advanced
• 1:22:43
  Speaker 1: SMR designs incorporate passive safety systems that can shut down
• 1:22:47
  Speaker 1: reactors and maintain cooling without external power or human intervention.
• 1:22:53
  Speaker 1: These systems rely on natural forces like gravity and convection
• 1:22:57
  Speaker 1: rather than active mechanical systems, providing multiple layers of protection
• 1:23:02
  Speaker 1: against potential accidents. The smaller size of SMRs also means
• 1:23:07
  Speaker 1: they require much less space and don't need the 10-mile
• 1:23:11
  Speaker 1: emergency planning zone that traditional nuclear plants require. Despite these advantages,
• 1:23:18
  Speaker 1: nuclear energy faces significant challenges that must be addressed. The
• 1:23:22
  Speaker 1: cost of building new nuclear power plants has increased substantially
• 1:23:27
  Speaker 1: with projects becoming nearly 20% more expensive than expected due
• 1:23:31
  Speaker 1: to delays and complexity. Construction delays have plagued recent nuclear projects,
• 1:23:38
  Speaker 1: with concrete placement issues frequently causing months-long setbacks. Three-quarters of
• 1:23:43
  Speaker 1: recent nuclear projects have experienced significant construction delays, resulting in
• 1:23:48
  Speaker 1: capital cost escalations, where newer reactors average 60% higher costs
• 1:23:52
  Speaker 1: than earlier ones. Regulatory approval processes add another layer of complexity.
• 1:24:00
  Speaker 1: Constellation has yet to file an application with federal nuclear
• 1:24:03
  Speaker 1: regulators to restart Three Mile Island, and there's limited historical
• 1:24:08
  Speaker 1: precedent for regulatory approval to restart shuttered plants. The Nuclear
• 1:24:13
  Speaker 1: Regulatory Commission has approved the first SMR design in the
• 1:24:16
  Speaker 1: United States, NuScale's advanced light-water reactor. but the regulatory framework
• 1:24:22
  Speaker 1: for these new technologies is still evolving. The economic case
• 1:24:26
  Speaker 1: for nuclear energy in AI applications becomes compelling when considering
• 1:24:31
  Speaker 1: the unique requirements of data centres and the limitations of alternatives.
• 1:24:37
  Speaker 1: While renewable energy costs have declined dramatically, the need for constant,
• 1:24:42
  Speaker 1: reliable power creates situations where nuclear energy's higher upfront costs
• 1:24:48
  Speaker 1: can be justified by its superior reliability and capacity factors.
• 1:24:53
  Speaker 1: The alignment between data center operators' need for consistent, low-carbon
• 1:24:58
  Speaker 1: electricity and SMR's ability to provide reliable baseload power creates
• 1:25:03
  Speaker 1: a natural partnership. Furthermore, the global market for SMRs is
• 1:25:08
  Speaker 1: projected to grow dramatically, with spending potentially increasing from $ 5
• 1:25:12
  Speaker 1: billion today to $ 25 billion by 2030. and reaching $ 670
• 1:25:19
  Speaker 1: billion by 2050. This growth trajectory suggests that economies of
• 1:25:24
  Speaker 1: scale will drive down costs over time, making nuclear energy
• 1:25:28
  Speaker 1: increasingly competitive with other power sources. The integration of nuclear
• 1:25:33
  Speaker 1: power for AI applications also presents significant infrastructure challenges. Data
• 1:25:39
  Speaker 1: centers tend to concentrate in specific geographic locations, making their
• 1:25:44
  Speaker 1: integration into the electrical grid more complex than distributed renewable
• 1:25:50
  Speaker 1: energy sources. The ageing power grid in many developed countries
• 1:25:55
  Speaker 1: already struggles to meet current demand, and adding large concentrated
• 1:25:59
  Speaker 1: loads from AI facilities will require substantial infrastructure investments. However,
• 1:26:05
  Speaker 1: the strategic placement of nuclear facilities near major data centre
• 1:26:09
  Speaker 1: hubs could actually reduce transmission infrastructure requirements. By generating power
• 1:26:15
  Speaker 1: close to where it's consumed, nuclear plants can minimize transmission
• 1:26:19
  Speaker 1: losses and reduce the need for expensive long-distance power lines.
• 1:26:26
  Speaker 1: This distributed approach to nuclear deployment could create more resilient
• 1:26:30
  Speaker 1: and efficient electrical systems. The climate implications of AI's energy
• 1:26:35
  Speaker 1: demands make nuclear energy an essential component of sustainable AI development.
• 1:26:41
  Speaker 1: Major technology companies are already seeing substantial increases in their
• 1:26:46
  Speaker 1: carbon emissions due to data center expansion. Google's emissions have
• 1:26:51
  Speaker 1: risen 48% since 2019, mostly due to data center energy consumption,
• 1:26:57
  Speaker 1: while Microsoft's emissions were 31% higher in 2024 than in 2020,
• 1:27:03
  Speaker 1: primarily due to AI infrastructure build-out. Without clean energy sources
• 1:27:08
  Speaker 1: like nuclear power, the AI revolution could become a significant
• 1:27:13
  Speaker 1: contributor to global carbon emissions. While data centres currently account
• 1:27:18
  Speaker 1: for less than 2% of global energy use, Goldman Sachs
• 1:27:23
  Speaker 1: estimates this could rise to 4% by the end of
• 1:27:26
  Speaker 1: the decade, a monumental increase that would substantially impact global
• 1:27:31
  Speaker 1: climate goals. The potential for nuclear fusion adds another dimension
• 1:27:36
  Speaker 1: to the long-term energy outlook for AI. While fusion remains
• 1:27:41
  Speaker 1: years away from commercial viability, recent breakthroughs have generated significant
• 1:27:46
  Speaker 1: interest and investment. OpenAI CEO Sam Altman has personally invested
• 1:27:54
  Speaker 1: $ 375 million in Helion Energy, a private fusion company that
• 1:27:59
  Speaker 1: has signed a deal to provide energy to Microsoft. The
• 1:28:02
  Speaker 1: prospect of virtually limitless clean energy from fusion could eventually
• 1:28:06
  Speaker 1: provide the ultimate solution to AI's energy requirements. the financial
• 1:28:11
  Speaker 1: markets are responding positively to the nuclear renaissance driven by
• 1:28:15
  Speaker 1: AI demand. Constellation's stock price surged more than 20% following
• 1:28:22
  Speaker 1: the Three Mile Island announcement and has risen over 100% year-to-date.
• 1:28:27
  Speaker 1: Other nuclear companies are seeing similar investor enthusiasm, with some
• 1:28:31
  Speaker 1: stocks climbing over 200% in recent months. This market response
• 1:28:37
  Speaker 1: reflects growing recognition that nuclear energy represents a critical infrastructure
• 1:28:41
  Speaker 1: investment for the digital economy. As AI applications become increasingly
• 1:28:46
  Speaker 1: central to business operations, economic productivity and national competitiveness, ensuring
• 1:28:52
  Speaker 1: reliable power supplies becomes a strategic imperative that justifies substantial
• 1:28:57
  Speaker 1: capital investments. The global nature of AI development creates additional
• 1:29:02
  Speaker 1: complexity in nuclear energy deployment. China and the United States
• 1:29:07
  Speaker 1: are expected to account for nearly 80% of global data
• 1:29:11
  Speaker 1: center electricity consumption growth through 2030, creating concentrated demand in
• 1:29:16
  Speaker 1: specific regions. China is actively exploring SMRs expected to play
• 1:29:21
  Speaker 1: a significant role between 2030 and 2035, while the United
• 1:29:26
  Speaker 1: States is providing substantial government support for nuclear development. Government
• 1:29:31
  Speaker 1: policies and support mechanisms are crucial for the nuclear renaissance
• 1:29:36
  Speaker 1: in AI applications. The Biden administration has provided billions in
• 1:29:41
  Speaker 1: tax credits and funding to support nuclear energy development, while
• 1:29:46
  Speaker 1: Congress has allocated substantial resources to prevent existing plants from
• 1:29:50
  Speaker 1: shutting down. International cooperation on nuclear technology development could accelerate
• 1:29:57
  Speaker 1: deployment timelines and reduce costs through shared research and development efforts.
• 1:30:02
  Speaker 1: The convergence of AI's energy demands and nuclear technology's capabilities
• 1:30:07
  Speaker 1: represents more than just a pragmatic solution to a growing problem.
• 1:30:12
  Speaker 1: It reflects a fundamental shift in how we think about
• 1:30:16
  Speaker 1: energy infrastructure in the digital age. The massive concentrated power
• 1:30:22
  Speaker 1: requirements of AI facilities are driving innovation in nuclear technology.
• 1:30:28
  Speaker 1: accelerating the development of smaller, more flexible reactor designs that
• 1:30:32
  Speaker 1: could eventually benefit a wide range of applications beyond data centres.
• 1:30:37
  Speaker 1: As we stand at this technological crossroads, the decisions we
• 1:30:40
  Speaker 1: make about energy infrastructure will shape the trajectory of artificial
• 1:30:44
  Speaker 1: intelligence development for decades to come. The choice between accepting
• 1:30:48
  Speaker 1: the limitations of intermittent renewable energy or embracing the reliable
• 1:30:52
  Speaker 1: power of nuclear technology will determine whether we can fully
• 1:30:56
  Speaker 1: realize AI's potential to address humanity's greatest challenges. The path
• 1:31:02
  Speaker 1: forward requires careful consideration of multiple factors, economic viability, environmental impact,
• 1:31:10
  Speaker 1: safety considerations, regulatory frameworks, and social acceptance. While nuclear energy
• 1:31:17
  Speaker 1: offers compelling advantages for powering AI applications, successful implementation will
• 1:31:23
  Speaker 1: require sustained commitment from governments, industry and society as a whole.
• 1:31:28
  Speaker 1: The nuclear renaissance driven by AI represents more than just
• 1:31:31
  Speaker 1: an energy solution. It embodies our collective determination to harness
• 1:31:35
  Speaker 1: technology's power while responsibly managing its consequences. As we continue
• 1:31:42
  Speaker 1: to push the boundaries of what artificial intelligence can achieve,
• 1:31:46
  Speaker 1: ensuring that these systems have access to clean, reliable and
• 1:31:51
  Speaker 1: abundant energy becomes not just a technical challenge, but a
• 1:31:55
  Speaker 1: moral imperative for sustainable technological progress. The future of AI
• 1:32:01
  Speaker 1: depends not just on advances in algorithms and computing power,
• 1:32:06
  Speaker 1: but on our ability to provide the energy infrastructure necessary
• 1:32:11
  Speaker 1: to support these technologies. Nuclear energy with its unique combination
• 1:32:17
  Speaker 1: of reliability, scalability and environmental benefits offers a pathway to
• 1:32:23
  Speaker 1: an AI-powered future that is both technologically advanced and environmentally responsible.
• 1:32:31
  Speaker 1: The choices we make today about energy infrastructure will determine
• 1:32:35
  Speaker 1: whether artificial intelligence becomes a force for solving humanity's greatest
• 1:32:39
  Speaker 1: challenges or is constrained by the very energy limitations that
• 1:32:44
  Speaker 1: once defined our pre-digital world. the escalating energy demands of
• 1:32:50
  Speaker 1: artificial intelligence represent one of the most formidable challenges facing
• 1:32:55
  Speaker 1: our technological future. As we've seen throughout this series, AI's
• 1:32:59
  Speaker 1: hunger for power grows exponentially with each advancement. Training a
• 1:33:04
  Speaker 1: single large language model can consume as much electricity as
• 1:33:08
  Speaker 1: a small city uses in a year, while running inference
• 1:33:11
  Speaker 1: for billions of daily queries requires data centres that operate
• 1:33:15
  Speaker 1: around the clock. demanding consistent, reliable baseload power that neither
• 1:33:20
  Speaker 1: solar panels nor wind turbines can provide with their inherent intermittency.
• 1:33:26
  Speaker 1: This is where nuclear energy emerges, not just as a
• 1:33:29
  Speaker 1: viable solution, but perhaps as the only scalable answer to
• 1:33:33
  Speaker 1: AI's relentless appetite for electricity. The numbers are staggering and sobering.
• 1:33:41
  Speaker 1: Data centre electricity consumption is projected to more than double
• 1:33:45
  Speaker 1: by 2030, potentially reaching 12% of total United States electricity demand.
• 1:33:52
  Speaker 1: To put this in perspective, that's equivalent to powering entire
• 1:33:56
  Speaker 1: nations like Italy or Australia solely for artificial intelligence operations.
• 1:34:01
  Speaker 1: The challenge isn't simply about generating more electricity. It's about
• 1:34:05
  Speaker 1: generating the right kind of electricity. AI systems cannot pause
• 1:34:10
  Speaker 1: for cloudy days or windless nights. They require what engineers
• 1:34:15
  Speaker 1: call baseload power that steady, unwavering flow of electrons that
• 1:34:21
  Speaker 1: keeps the digital minds humming 24 hours a day, seven
• 1:34:25
  Speaker 1: days a week. Traditional renewable sources, despite their environmental benefits,
• 1:34:31
  Speaker 1: face fundamental limitations in meeting this demand. Solar panels produce
• 1:34:37
  Speaker 1: nothing at night, wind turbines stand motionless during calm weather,
• 1:34:42
  Speaker 1: and current battery storage technologies cannot economically bridge these gaps
• 1:34:47
  Speaker 1: for the massive scale required by AI operations. Nuclear power plants,
• 1:34:53
  Speaker 1: by contrast, operate with capacity factors exceeding 90%, meaning they
• 1:34:58
  Speaker 1: produce their rated output nearly continuously throughout the year. A
• 1:35:03
  Speaker 1: single nuclear reactor can generate between 800 and 1,600 megawatts
• 1:35:08
  Speaker 1: of electricity consistently, enough to power millions of homes, or
• 1:35:12
  Speaker 1: more relevantly, entire clusters of AI data centers. This reliability
• 1:35:18
  Speaker 1: stems from nuclear fuel's extraordinary energy density. A single uranium
• 1:35:23
  Speaker 1: fuel pellet, no larger than a fingertip, contains as much
• 1:35:27
  Speaker 1: energy as a ton of coal or three barrels of oil.
• 1:35:31
  Speaker 1: The artificial intelligence industry has taken notice of these advantages,
• 1:35:35
  Speaker 1: leading to an unprecedented renaissance in nuclear investment. Microsoft's groundbreaking
• 1:35:41
  Speaker 1: agreement to restart Three Mile Island represents a historic first.
• 1:35:46
  Speaker 1: Never before has a decommissioned nuclear plant been brought back online,
• 1:35:50
  Speaker 1: and never before has an entire plant's output been dedicated
• 1:35:54
  Speaker 1: to a single corporate customer. This 20-year power purchase agreement
• 1:35:59
  Speaker 1: will provide Microsoft with 835 megawatts of clean, reliable electricity,
• 1:36:06
  Speaker 1: demonstrating the tech industry's commitment to nuclear solutions. Amazon has
• 1:36:11
  Speaker 1: made an even more comprehensive commitment signing agreements to develop
• 1:36:15
  Speaker 1: small modular reactors that could provide up to 5 gigawatts
• 1:36:19
  Speaker 1: of nuclear capacity by 2039. The company is investing hundreds
• 1:36:25
  Speaker 1: of millions of dollars in advanced reactor technologies, recognising that
• 1:36:29
  Speaker 1: their cloud computing empire, which powers much of the world's
• 1:36:33
  Speaker 1: AI infrastructure, requires energy sources that can match their global
• 1:36:38
  Speaker 1: scale and reliability requirements. Google has similarly partnered with reactor
• 1:36:43
  Speaker 1: developers to deploy small modular reactors capable of generating 500
• 1:36:48
  Speaker 1: megawatts of carbon-free electricity specifically for their AI operations. These
• 1:36:53
  Speaker 1: aren't distant dreams or speculative investments. These are concrete business
• 1:36:57
  Speaker 1: decisions made by companies that understand the energy realities of
• 1:37:01
  Speaker 1: artificial intelligence better than anyone else. The emergence of small
• 1:37:06
  Speaker 1: modular reactors represents a paradigm shift in nuclear technology that
• 1:37:11
  Speaker 1: could revolutionise how we power AI systems. Unlike traditional nuclear
• 1:37:16
  Speaker 1: plants that require massive capital investments and decade-long construction timelines,
• 1:37:23
  Speaker 1: SMRs can be manufactured in factories and deployed in just
• 1:37:28
  Speaker 1: three to five years. Their modular design allows for incremental
• 1:37:33
  Speaker 1: capacity additions, perfectly matching the growing demands of AI infrastructure.
• 1:37:39
  Speaker 1: These smaller reactors offer unique advantages for AI applications. They
• 1:37:44
  Speaker 1: can be located closer to data centres, reducing transmission losses
• 1:37:49
  Speaker 1: and improving grid stability. Their inherent safety features, including passive
• 1:37:55
  Speaker 1: cooling systems and underground deployment options, address many of the
• 1:38:00
  Speaker 1: concerns that have historically surrounded nuclear power. Most importantly, they
• 1:38:06
  Speaker 1: can provide the precise amount of baseload power needed for
• 1:38:09
  Speaker 1: specific AI installations, without the massive overcapacity that traditional nuclear
• 1:38:16
  Speaker 1: plants might create. The economic case for nuclear-powered AI is
• 1:38:21
  Speaker 1: becoming increasingly compelling. While renewable energy sources might seem cheaper
• 1:38:26
  Speaker 1: on paper, their intermittency requires expensive backup systems and storage
• 1:38:32
  Speaker 1: solutions that quickly erode any cost advantage. Nuclear plants, once built,
• 1:38:38
  Speaker 1: can operate for 60 to 80 years with minimal fuel
• 1:38:41
  Speaker 1: costs and predictable maintenance schedules. For AI companies planning decades-long
• 1:38:47
  Speaker 1: infrastructure investments, this long-term cost stability is invaluable. The energy
• 1:38:53
  Speaker 1: density of nuclear fuel provides another crucial advantage. A typical
• 1:38:58
  Speaker 1: nuclear plant requires just a few hundred acres of land
• 1:39:02
  Speaker 1: to generate massive amounts of electricity, while equivalent solar installations
• 1:39:07
  Speaker 1: would require thousands of square miles. For AI companies seeking
• 1:39:11
  Speaker 1: to minimize their environmental footprint, while maximizing their energy output,
• 1:39:17
  Speaker 1: nuclear offers an unmatched combination of efficiency and land use optimization.
• 1:39:23
  Speaker 1: The reliability factor cannot be overstated. AI training runs can
• 1:39:29
  Speaker 1: take weeks or months to complete, representing millions of dollars
• 1:39:33
  Speaker 1: in computational investment. A power outage during a critical training
• 1:39:37
  Speaker 1: phase can destroy weeks of work and set back research
• 1:39:41
  Speaker 1: timelines significantly. Nuclear plants provide the kind of rock-solid reliability
• 1:39:46
  Speaker 1: that AI operations demand, with unplanned outage rates below 1% annually.
• 1:39:52
  Speaker 1: The carbon footprint considerations are equally important. While tech companies
• 1:39:57
  Speaker 1: have made ambitious commitments to carbon neutrality, the massive energy
• 1:40:01
  Speaker 1: demands of AI threaten to undermine these goals. Nuclear power
• 1:40:05
  Speaker 1: produces virtually no carbon emissions during operations, making it the
• 1:40:09
  Speaker 1: only scalable technology that can support massive AI expansion while
• 1:40:13
  Speaker 1: meeting climate commitments. This environmental advantage is particularly crucial, as
• 1:40:19
  Speaker 1: governments worldwide implement carbon pricing and emissions regulations. The technical
• 1:40:25
  Speaker 1: synergy between nuclear power and AI operations extends beyond simple
• 1:40:30
  Speaker 1: electricity generation. Nuclear plants produce enormous amounts of waste heat
• 1:40:35
  Speaker 1: that could be captured and used for data center cooling,
• 1:40:39
  Speaker 1: improving overall system efficiency. Some advanced reactor designs are specifically
• 1:40:44
  Speaker 1: optimized for industrial applications, providing both electricity and process heat
• 1:40:50
  Speaker 1: that could support AI infrastructure operations. The global implications of
• 1:40:55
  Speaker 1: nuclear-powered AI are profound. Countries that successfully deploy nuclear AI
• 1:41:01
  Speaker 1: partnerships will gain significant competitive advantages in the artificial intelligence race.
• 1:41:08
  Speaker 1: China is already exploring SMR deployments specifically for data center applications,
• 1:41:14
  Speaker 1: while the United States is providing hundreds of millions in
• 1:41:17
  Speaker 1: government funding to accelerate nuclear development for AI applications. However,
• 1:41:23
  Speaker 1: the path forward is not without challenges. Nuclear projects face
• 1:41:27
  Speaker 1: complex regulatory approval processes, though these are being streamlined for
• 1:41:32
  Speaker 1: SMR applications. Public perception remains a concern, despite the excellent
• 1:41:38
  Speaker 1: safety record of modern nuclear technology. The skilled workforce required
• 1:41:43
  Speaker 1: for nuclear operations represents another constraint, though this is being
• 1:41:48
  Speaker 1: addressed through expanded training programs and automated systems. The financial
• 1:41:53
  Speaker 1: markets are responding positively to nuclear AI partnerships. Constellation Energy,
• 1:41:59
  Speaker 1: saw its stock price surge over 20% following the Three
• 1:42:02
  Speaker 1: Mile Island announcement, while other nuclear companies have experienced similar gains.
• 1:42:09
  Speaker 1: Investment in SMR technology is projected to grow from $ 5
• 1:42:13
  Speaker 1: billion today to $ 25 billion by 2030, driven largely by
• 1:42:18
  Speaker 1: AI demand. The timeline for nuclear AI deployment is accelerating rapidly.
• 1:42:24
  Speaker 1: Microsoft's three-mile island restart could begin providing power by 2027,
• 1:42:30
  Speaker 1: while Google's SMR partnership targets 2030 for initial operations. Amazon's
• 1:42:36
  Speaker 1: more ambitious SMR program extends to 2039, but will provide
• 1:42:41
  Speaker 1: unprecedented scale once operational. These timelines align well with projected
• 1:42:46
  Speaker 1: AI growth curves, suggesting that nuclear capacity will come online
• 1:42:50
  Speaker 1: just as AI demand reaches its most critical phase. the
• 1:42:55
  Speaker 1: ripple effects of nuclear-powered AI extend far beyond the technology sector.
• 1:43:02
  Speaker 1: The revival of nuclear manufacturing capabilities, the creation of skilled
• 1:43:06
  Speaker 1: technical jobs, and the development of new reactor technologies will
• 1:43:10
  Speaker 1: benefit entire regional economies. The United States has the opportunity
• 1:43:16
  Speaker 1: to reclaim global leadership in nuclear technology while simultaneously dominating
• 1:43:21
  Speaker 1: the AI revolution. Looking ahead, The marriage of nuclear power
• 1:43:26
  Speaker 1: and artificial intelligence appears inevitable. The energy requirements are simply
• 1:43:30
  Speaker 1: too massive, too consistent, and too critical to be met
• 1:43:34
  Speaker 1: by any other technology. Solar and wind will continue to
• 1:43:39
  Speaker 1: play important roles in the broader energy mix. But for
• 1:43:43
  Speaker 1: the specific demands of AI operations, nuclear power offers unmatched
• 1:43:48
  Speaker 1: advantages in reliability, scalability, and environmental performance. The question is
• 1:43:55
  Speaker 1: not whether nuclear energy will power the AI revolution, but
• 1:43:58
  Speaker 1: how quickly we can deploy it and how effectively we
• 1:44:01
  Speaker 1: can integrate it into our technological infrastructure. The companies and
• 1:44:06
  Speaker 1: countries that move fastest and most decisively in this direction
• 1:44:09
  Speaker 1: will shape the future of artificial intelligence and, by extension,
• 1:44:14
  Speaker 1: the future of human civilization itself. As we stand at
• 1:44:18
  Speaker 1: this critical juncture, the convergence of nuclear power and artificial
• 1:44:23
  Speaker 1: intelligence represents more than just a technological solution to an
• 1:44:27
  Speaker 1: energy problem. It embodies our capacity to harness the most
• 1:44:32
  Speaker 1: powerful forces in the universe, both the splitting of atoms
• 1:44:36
  Speaker 1: and the creation of digital minds in service of human
• 1:44:40
  Speaker 1: progress and prosperity. The age of nuclear-powered AI is beginning,
• 1:44:47
  Speaker 1: and its implications will resonate for generations to come. Looking
• 1:44:52
  Speaker 1: at the nuclear energy and AI power challenge, we've reached
• 1:44:55
  Speaker 1: a critical juncture where the promise of artificial intelligence meets
• 1:45:00
  Speaker 1: the harsh reality of energy physics. The exponential growth in
• 1:45:05
  Speaker 1: AI processing demands has created an unprecedented hunger for electricity
• 1:45:10
  Speaker 1: that traditional power sources simply cannot satisfy reliably and sustainably.
• 1:45:16
  Speaker 1: The numbers tell a stark story. Data centres already consume
• 1:45:21
  Speaker 1: about 4% of total US electricity. But this is just
• 1:45:26
  Speaker 1: the beginning. With AI inference tasks projected to grow at
• 1:45:30
  Speaker 1: a staggering 122% annually through 2028, we're looking at energy
• 1:45:35
  Speaker 1: demands that could reach 12% of total American electricity consumption
• 1:45:39
  Speaker 1: within just a few years. This isn't gradual growth. It's
• 1:45:44
  Speaker 1: an energy tsunami. But here's where the story gets interesting.
• 1:45:49
  Speaker 1: The tech giants driving this AI revolution aren't just hoping
• 1:45:53
  Speaker 1: someone else will solve their energy problem. They're making massive,
• 1:45:57
  Speaker 1: unprecedented bets on nuclear power. Microsoft's agreement to restart Three
• 1:46:03
  Speaker 1: Mile Island represents something that has never happened before in
• 1:46:06
  Speaker 1: American energy history. Never has a decommissioned nuclear plant been
• 1:46:10
  Speaker 1: brought back online. Never has an entire nuclear plant's output
• 1:46:15
  Speaker 1: been dedicated to a single company. Amazon has gone even further,
• 1:46:19
  Speaker 1: committing to develop up to 5 gigawatts of nuclear capacity
• 1:46:23
  Speaker 1: through small modular reactors. To put this in perspective, that's
• 1:46:27
  Speaker 1: enough to power about 4 million homes, all dedicated to
• 1:46:31
  Speaker 1: keeping Amazon's AI systems running. Google has partnered with Kairos
• 1:46:35
  Speaker 1: Power to deploy advanced reactors that use molten salt cooling
• 1:46:39
  Speaker 1: systems and ceramic fuel, technologies that sound like science fiction
• 1:46:44
  Speaker 1: but are rapidly becoming reality. The timing of these commitments
• 1:46:49
  Speaker 1: isn't coincidental. Traditional renewable energy sources, despite their environmental benefits,
• 1:46:55
  Speaker 1: face a fundamental challenge when it comes to AI applications.
• 1:46:59
  Speaker 1: While wind and solar can provide roughly 80% of a
• 1:47:02
  Speaker 1: data center's power needs when paired with storage, they cannot
• 1:47:06
  Speaker 1: deliver the constant reliable baseload power that AI systems require.
• 1:47:12
  Speaker 1: Artificial intelligence doesn't sleep, doesn't take weekends off, and doesn't
• 1:47:16
  Speaker 1: care if the wind isn't blowing or the sun isn't shining.
• 1:47:21
  Speaker 1: This is where nuclear energy's unique characteristics become invaluable. Nuclear
• 1:47:27
  Speaker 1: plants operate at capacity factors exceeding 90%, meaning they produce
• 1:47:31
  Speaker 1: their rated power output more than 90% of the time.
• 1:47:36
  Speaker 1: Compare this to solar power, which typically achieves capacity factors
• 1:47:40
  Speaker 1: of 20 to 30%, or wind power at 30 to 40%.
• 1:47:46
  Speaker 1: For AI applications requiring constant power, this reliability difference is transformative.
• 1:47:52
  Speaker 1: The technology enabling this nuclear renaissance is fundamentally different from
• 1:47:56
  Speaker 1: the massive, complex plants of the past. Small modular reactors
• 1:48:01
  Speaker 1: represent a paradigm shift in nuclear design. These reactors, typically
• 1:48:06
  Speaker 1: producing between 20 and 300 megawatts, can be factory built
• 1:48:10
  Speaker 1: and transported to sites rather than constructed piece by piece
• 1:48:15
  Speaker 1: on location. This modular approach promises to reduce construction times
• 1:48:21
  Speaker 1: from the decade plus timelines of traditional plants to three
• 1:48:24
  Speaker 1: to five years. The economics are equally compelling. While traditional
• 1:48:28
  Speaker 1: nuclear plants can cost more than 10 billion dollars and
• 1:48:32
  Speaker 1: take 12 to 15 years to build, Small modular reactors
• 1:48:36
  Speaker 1: are projected to cost less than $ 2 billion with much
• 1:48:39
  Speaker 1: shorter construction timelines. This cost reduction comes from standardization, factory
• 1:48:45
  Speaker 1: production and smaller scale, making nuclear power accessible to applications
• 1:48:50
  Speaker 1: like data centers that need dedicated power sources. Safety improvements
• 1:48:55
  Speaker 1: in modern reactor designs address many of the concerns that
• 1:48:59
  Speaker 1: have historically limited nuclear expansion. Advanced reactors use passive safety
• 1:49:05
  Speaker 1: systems that don't require external power or human intervention to
• 1:49:09
  Speaker 1: shut down safely. Some designs, like the molten salt reactors
• 1:49:15
  Speaker 1: that Google is supporting, operate at atmospheric pressure, eliminating the
• 1:49:20
  Speaker 1: risk of high-pressure accidents that characterized earlier nuclear incidents. The
• 1:49:26
  Speaker 1: regulatory landscape is evolving to support these new technologies. The
• 1:49:30
  Speaker 1: Nuclear Regulatory Commission has already approved the first small modular
• 1:49:35
  Speaker 1: reactor design in the United States, and government policies, including
• 1:49:39
  Speaker 1: billions in tax credits from the Inflation Reduction Act, are
• 1:49:43
  Speaker 1: creating financial incentives for nuclear development. This policy support recognises
• 1:49:50
  Speaker 1: that nuclear energy is essential for achieving both energy security
• 1:49:54
  Speaker 1: and climate goals. But the nuclear solution extends beyond just
• 1:49:59
  Speaker 1: meeting AI's energy demands. the carbon footprint of artificial intelligence
• 1:50:06
  Speaker 1: is becoming a significant concern. Google's emissions have risen 48%
• 1:50:12
  Speaker 1: since 2019, primarily due to data center energy consumption. Microsoft's
• 1:50:18
  Speaker 1: emissions are 31% higher than in 2020, driven by AI infrastructure.
• 1:50:25
  Speaker 1: Nuclear power offers a path to massive AI deployment without
• 1:50:29
  Speaker 1: proportional increases in carbon emissions. The global implications are profound.
• 1:50:34
  Speaker 1: China and the United States account for nearly 80% of
• 1:50:39
  Speaker 1: projected data center electricity growth through 2030. The countries that
• 1:50:44
  Speaker 1: can deploy reliable, clean energy for AI applications will have
• 1:50:49
  Speaker 1: significant advantages in the global technology competition. Nuclear power provides
• 1:50:56
  Speaker 1: this advantage while reducing dependence on fossil fuels and improving
• 1:51:01
  Speaker 1: energy security. The challenges, however, are substantial. Nuclear construction projects
• 1:51:08
  Speaker 1: have historically faced delays and cost overruns. Even with improved
• 1:51:13
  Speaker 1: designs and manufacturing approaches, scaling nuclear capacity to meet AI's
• 1:51:18
  Speaker 1: growing demands will require unprecedented deployment speeds. The nuclear supply chain,
• 1:51:25
  Speaker 1: from uranium mining to reactor manufacturing, must expand dramatically to
• 1:51:30
  Speaker 1: support this growth. Public acceptance remains a hurdle, though attitudes
• 1:51:35
  Speaker 1: toward nuclear power are shifting. The urgency of climate change
• 1:51:41
  Speaker 1: and the recognition that renewable energy alone cannot meet all
• 1:51:45
  Speaker 1: electricity needs are creating more openness to nuclear solutions. The
• 1:51:51
  Speaker 1: association of nuclear power with cutting-edge AI technology may help
• 1:51:55
  Speaker 1: overcome some historical resistance. International dynamics add another layer of complexity.
• 1:52:01
  Speaker 1: Nuclear technology is subject to export controls and non-proliferation concerns.
• 1:52:07
  Speaker 1: The countries that lead in nuclear energy for AI applications
• 1:52:10
  Speaker 1: may gain significant geopolitical advantages, while those dependent on fossil
• 1:52:15
  Speaker 1: fuels for their AI infrastructure may find themselves at a disadvantage.
• 1:52:20
  Speaker 1: The convergence of AI and nuclear energy represents more than
• 1:52:24
  Speaker 1: just a technical solution to an energy problem. It's a
• 1:52:27
  Speaker 1: fundamental shift in how we think about power generation, technology
• 1:52:31
  Speaker 1: deployment and industrial strategy. The companies making these nuclear investments
• 1:52:37
  Speaker 1: aren't just buying electricity, they're buying energy independence, carbon neutrality
• 1:52:43
  Speaker 1: and technological superiority. Looking ahead, the success of these nuclear
• 1:52:48
  Speaker 1: AI partnerships will likely determine whether artificial intelligence becomes an
• 1:52:53
  Speaker 1: environmentally sustainable technology or a major contributor to climate change.
• 1:52:58
  Speaker 1: The stakes couldn't be higher. The future of AI, and
• 1:53:02
  Speaker 1: perhaps our ability to address humanity's greatest challenges through artificial intelligence,
• 1:53:07
  Speaker 1: may depend on our success in harnessing the atom to
• 1:53:11
  Speaker 1: power the algorithm. The gravity of this moment is difficult
• 1:53:15
  Speaker 1: to overstate. We're witnessing the birth of a new energy paradigm,
• 1:53:20
  Speaker 1: one where the most advanced computing technologies are powered by
• 1:53:24
  Speaker 1: the most concentrated energy source humans have ever harnessed. The
• 1:53:31
  Speaker 1: outcome of this nuclear-AI convergence will shape not just the
• 1:53:34
  Speaker 1: technology landscape, but the environmental and geopolitical future of our planet.
• 1:53:40
  Speaker 1: Looking at the nuclear energy developments in AI, we find
• 1:53:43
  Speaker 1: ourselves at a pivotal moment where the intersection of technology
• 1:53:47
  Speaker 1: and power generation is creating unprecedented opportunities and challenges. The
• 1:53:53
  Speaker 1: sheer scale of energy consumption that artificial intelligence demands has
• 1:53:58
  Speaker 1: fundamentally altered the landscape of power planning and infrastructure development
• 1:54:03
  Speaker 1: across the globe. The financial implications of this energy revolution
• 1:54:08
  Speaker 1: extend far beyond simple electricity bills. When we examine the
• 1:54:14
  Speaker 1: economic models driving these nuclear partnerships, we discover a complex
• 1:54:19
  Speaker 1: web of risk assessment, long-term planning and strategic positioning. that
• 1:54:25
  Speaker 1: reflects the deep commitment major technology companies have made to
• 1:54:29
  Speaker 1: artificial intelligence development. These aren't casual investments or temporary solutions.
• 1:54:35
  Speaker 1: They represent fundamental shifts in how corporations think about energy security,
• 1:54:40
  Speaker 1: environmental responsibility, and competitive advantage. Microsoft's 20-year commitment to Three
• 1:54:47
  Speaker 1: Mile Island exemplifies this strategic thinking. The company isn't just
• 1:54:52
  Speaker 1: purchasing electricity. it's essentially underwriting the revival of nuclear power
• 1:54:57
  Speaker 1: in America. This deal, worth potentially billions of dollars over
• 1:55:02
  Speaker 1: its lifetime, demonstrates how AI companies are willing to make
• 1:55:06
  Speaker 1: substantial long-term investments to secure the reliable, carbon-free power their
• 1:55:12
  Speaker 1: data centers require. The financial structure of these agreements typically
• 1:55:18
  Speaker 1: involves complex power purchase agreements that guarantee minimum purchase volumes
• 1:55:23
  Speaker 1: provide price stability, and often include provisions for additional capacity
• 1:55:27
  Speaker 1: as AI demands grow. Amazon's approach reveals an even more
• 1:55:34
  Speaker 1: sophisticated strategy. By investing directly in small modular reactor technology
• 1:55:39
  Speaker 1: through its $ 500 million investment in Xenergy, Amazon is positioning
• 1:55:44
  Speaker 1: itself not just as a consumer of nuclear power, but
• 1:55:47
  Speaker 1: as a catalyst for the entire industry's transformation. This investment
• 1:55:54
  Speaker 1: strategy serves multiple purposes. It helps ensure the development of
• 1:55:58
  Speaker 1: reactor designs specifically suited to data center applications. It potentially
• 1:56:03
  Speaker 1: provides Amazon with preferential access to future reactor deployments. And
• 1:56:07
  Speaker 1: it positions the company as a leader in the next
• 1:56:09
  Speaker 1: generation of nuclear technology. The economic benefits of these nuclear
• 1:56:15
  Speaker 1: partnerships extend beyond the immediate parties involved. Local communities hosting
• 1:56:20
  Speaker 1: these facilities benefit from substantial job creation, tax revenue and
• 1:56:25
  Speaker 1: economic revitalization. The restart of Three Mile Island, for instance,
• 1:56:31
  Speaker 1: is expected to create hundreds of direct jobs and thousands
• 1:56:35
  Speaker 1: of indirect jobs in the surrounding Pennsylvania community. The economic
• 1:56:40
  Speaker 1: multiplier effects of nuclear facilities are particularly pronounced. because they
• 1:56:45
  Speaker 1: employ highly skilled workers with above average wages who spend
• 1:56:49
  Speaker 1: their earnings in the local economy. From a broader economic perspective,
• 1:56:55
  Speaker 1: these investments are helping to revitalise America's nuclear supply chain.
• 1:57:00
  Speaker 1: For decades, the lack of new nuclear construction has led
• 1:57:04
  Speaker 1: to the erosion of domestic nuclear manufacturing capabilities, with many
• 1:57:09
  Speaker 1: specialised components now sourced from overseas. The renewed interest in
• 1:57:15
  Speaker 1: nuclear power driven by AI energy demands is creating incentives
• 1:57:20
  Speaker 1: for domestic manufacturers to rebuild these capabilities, potentially strengthening America's
• 1:57:26
  Speaker 1: energy independence and industrial base. The risk profiles of these
• 1:57:31
  Speaker 1: nuclear investments are complex and multifaceted. Traditional financial analysis might
• 1:57:38
  Speaker 1: suggest that 20- to 30-year power purchase agreements carry substantial risk,
• 1:57:43
  Speaker 1: particularly given the rapid pace of technological change in both
• 1:57:47
  Speaker 1: AI and energy sectors. However, the companies making these investments
• 1:57:53
  Speaker 1: have concluded that the risks of not securing reliable, carbon-free
• 1:57:57
  Speaker 1: power are even greater. The potential costs of power shortages,
• 1:58:02
  Speaker 1: grid instability, or carbon penalties under future climate regulations far
• 1:58:07
  Speaker 1: outweigh the risks of long-term nuclear competition. Insurance considerations add
• 1:58:13
  Speaker 1: another layer of complexity to these arrangements. Nuclear facilities require
• 1:58:19
  Speaker 1: specialised insurance coverage, and the expansion of nuclear power for
• 1:58:23
  Speaker 1: AI applications is creating new categories of risk that insurance
• 1:58:27
  Speaker 1: companies are still learning to assess. The potential for cyberattacks
• 1:58:32
  Speaker 1: on nuclear facilities, for instance, represents a new category of
• 1:58:36
  Speaker 1: risk that didn't exist when most current nuclear plants were
• 1:58:39
  Speaker 1: designed and built. The international competitive implications of these investments
• 1:58:44
  Speaker 1: cannot be overlooked. China's aggressive expansion of nuclear power, including
• 1:58:49
  Speaker 1: significant investments in small modular reactor technology, represents a strategic
• 1:58:54
  Speaker 1: challenge to American technological leadership. The willingness of American technology
• 1:59:00
  Speaker 1: companies to invest in nuclear power helps maintain America's competitive
• 1:59:04
  Speaker 1: position in both AI development and advanced nuclear technology. Regulatory
• 1:59:11
  Speaker 1: economics also play a crucial role in these decisions. The
• 1:59:15
  Speaker 1: Nuclear Regulatory Commission's approach to licensing new reactor designs, particularly
• 1:59:20
  Speaker 1: small modular reactors, directly impacts the economics of these investments.
• 1:59:26
  Speaker 1: streamlined regulatory processes could significantly reduce the time and cost
• 1:59:30
  Speaker 1: required to deploy new nuclear capacity, making these investments more
• 1:59:35
  Speaker 1: attractive and economically viable. The Environmental Economics of Nuclear Power
• 1:59:41
  Speaker 1: for AI applications present a compelling case study in how
• 1:59:45
  Speaker 1: climate considerations are reshaping business decisions. The avoided carbon emissions
• 1:59:50
  Speaker 1: from these nuclear facilities, when compared to fossil fuel alternatives,
• 1:59:54
  Speaker 1: represent substantial environmental benefits that are increasingly being valued in
• 1:59:59
  Speaker 1: corporate decision-making processes. Many companies are incorporating internal carbon pricing
• 2:00:05
  Speaker 1: into their investment decisions, making low-carbon nuclear power more economically
• 2:00:10
  Speaker 1: attractive than it might appear based solely on electricity costs.
• 2:00:16
  Speaker 1: Grid integration costs represent another important economic consideration. Unlike renewable
• 2:00:22
  Speaker 1: energy sources that require extensive transmission infrastructure and backup power systems,
• 2:00:29
  Speaker 1: nuclear facilities can be located relatively close to data centres
• 2:00:33
  Speaker 1: and provide consistent power output. This proximity reduces transmission losses
• 2:00:38
  Speaker 1: and infrastructure costs while improving reliability. The long-term economic sustainability
• 2:00:45
  Speaker 1: of these nuclear investments depends partly on the continued growth
• 2:00:49
  Speaker 1: of AI applications and energy demands. However, the companies making
• 2:00:55
  Speaker 1: these investments have concluded that AI represents a permanent shift
• 2:01:00
  Speaker 1: in computing paradigms rather than a temporary trend. The training
• 2:01:05
  Speaker 1: of large language models, the deployment of AI applications across industries,
• 2:01:10
  Speaker 1: and the development of artificial general intelligence all point towards
• 2:01:14
  Speaker 1: sustained growth in computational demands. The financing mechanisms for these
• 2:01:21
  Speaker 1: nuclear projects are themselves innovative. Rather than traditional utility financing models,
• 2:01:28
  Speaker 1: these partnerships often involve corporate credit backing, power purchase agreements,
• 2:01:33
  Speaker 1: and direct equity investments. This corporate involvement provides nuclear developers
• 2:01:39
  Speaker 1: with more certain revenue streams and lower financing costs, potentially
• 2:01:44
  Speaker 1: making nuclear projects more economically viable than they have been
• 2:01:48
  Speaker 1: in decades. The economic impact of these investments extends to
• 2:01:54
  Speaker 1: the broader energy sector. By demonstrating corporate willingness to pay
• 2:01:59
  Speaker 1: premium prices for reliable, carbon-free power, these nuclear partnerships are
• 2:02:04
  Speaker 1: helping to establish market value for clean energy attributes that
• 2:02:08
  Speaker 1: weren't previously monetised. This could influence pricing in other energy
• 2:02:12
  Speaker 1: markets and encourage additional clean energy investments. Looking ahead, The
• 2:02:18
  Speaker 1: economic success of these nuclear investments will likely influence future
• 2:02:22
  Speaker 1: energy infrastructure decisions across multiple industries. If these partnerships prove successful,
• 2:02:29
  Speaker 1: we can expect to see similar arrangements in other energy-intensive
• 2:02:32
  Speaker 1: sectors such as cryptocurrency mining, industrial manufacturing and large-scale computing applications.
• 2:02:40
  Speaker 1: The economic transformation being driven by AI's energy demands represents
• 2:02:44
  Speaker 1: more than just a shift in power generation. It's a
• 2:02:48
  Speaker 1: fundamental reimagining of how corporations approach energy procurement, risk management
• 2:02:54
  Speaker 1: and long-term strategic planning. These nuclear partnerships are pioneering new
• 2:03:00
  Speaker 1: models of corporate energy investment that could reshape the entire
• 2:03:03
  Speaker 1: energy industry for decades to come. The success of these
• 2:03:08
  Speaker 1: investments will ultimately be measured not just in terms of
• 2:03:11
  Speaker 1: reliable power delivery and cost management, but in their ability
• 2:03:15
  Speaker 1: to enable the continued growth and development of artificial intelligence
• 2:03:20
  Speaker 1: capabilities that are transforming every aspect of our economy and society.
• 2:03:27
  Speaker 1: In this context, nuclear energy isn't just solving AI's power problem,
• 2:03:32
  Speaker 1: it's enabling the future of human technological advancement. As we
• 2:03:37
  Speaker 1: drift deeper into the nuclear conversation, we find ourselves at
• 2:03:41
  Speaker 1: a fascinating crossroads where the immense power that lights stars
• 2:03:46
  Speaker 1: meets the insatiable hunger of artificial intelligence. The numbers we've
• 2:03:51
  Speaker 1: explored paint a picture that's both thrilling and sobering. AI's
• 2:03:56
  Speaker 1: energy appetite isn't just growing, it's exploding at a rate
• 2:04:00
  Speaker 1: that makes even the most ambitious renewable energy projections seem quaint.
• 2:04:04
  Speaker 1: Think about this for a moment. We're witnessing the birth
• 2:04:07
  Speaker 1: of a new kind of civilisation, one where digital minds
• 2:04:11
  Speaker 1: work alongside human ones. processing information at speeds that would
• 2:04:16
  Speaker 1: make the fastest human thinker seem glacial. But these digital
• 2:04:20
  Speaker 1: minds are hungry, perpetually hungry, consuming electricity at rates that
• 2:04:26
  Speaker 1: dwarf entire nations. The latest AI training runs require more
• 2:04:31
  Speaker 1: power than small countries use in a year, and we're
• 2:04:35
  Speaker 1: only at the beginning of this revolution. The traditional energy
• 2:04:39
  Speaker 1: sources we've relied upon for decades suddenly seem inadequate for
• 2:04:42
  Speaker 1: this new reality. Solar panels and wind turbines, for all
• 2:04:47
  Speaker 1: their environmental benefits, carry the fundamental limitation of intermittency. The
• 2:04:54
  Speaker 1: sun doesn't always shine, the wind doesn't always blow, but
• 2:04:58
  Speaker 1: artificial intelligence never sleeps. It processes, calculates, and learns 24
• 2:05:04
  Speaker 1: hours a day, seven days a week, 365 days a year.
• 2:05:10
  Speaker 1: This is where nuclear energy reveals its true character. Unlike
• 2:05:15
  Speaker 1: its renewable cousins, nuclear power doesn't depend on weather patterns
• 2:05:20
  Speaker 1: or seasonal variations. It's the steady heartbeat of the energy
• 2:05:24
  Speaker 1: world providing consistent, reliable power that matches the relentless demands
• 2:05:29
  Speaker 1: of AI systems. When Microsoft signed that historic deal to
• 2:05:34
  Speaker 1: restart Three Mile Island, They weren't just making an energy purchase.
• 2:05:39
  Speaker 1: They were acknowledging a fundamental truth about the future. The
• 2:05:43
  Speaker 1: relationship between nuclear power and artificial intelligence represents something deeper
• 2:05:48
  Speaker 1: than mere energy supply and demand. It's a partnership between
• 2:05:52
  Speaker 1: two of humanity's most sophisticated technologies, each pushing the boundaries
• 2:05:56
  Speaker 1: of what's possible. Nuclear reactors, with their precisely controlled fusion reactions,
• 2:06:03
  Speaker 1: mirror the controlled complexity of AI systems with their carefully
• 2:06:07
  Speaker 1: orchestrated neural networks. Consider the elegance of this pairing. Nuclear
• 2:06:14
  Speaker 1: power plants operate through incredibly complex systems of control rods,
• 2:06:18
  Speaker 1: coolant circulation, and safety mechanisms, all managed by sophisticated computer systems.
• 2:06:26
  Speaker 1: These same computer systems are increasingly incorporating AI algorithms to
• 2:06:30
  Speaker 1: optimize reactor performance, predict maintenance needs, and enhance safety protocols.
• 2:06:36
  Speaker 1: It's a symbiotic relationship where each technology enhances the other.
• 2:06:41
  Speaker 1: The economic implications are staggering. Amazon's multi-billion dollar commitment to
• 2:06:46
  Speaker 1: nuclear energy isn't just corporate virtue signaling. It's a strategic
• 2:06:50
  Speaker 1: recognition that the future of cloud computing and AI depends
• 2:06:55
  Speaker 1: on having access to massive amounts of reliable, clean energy.
• 2:07:00
  Speaker 1: When Google partners with Kairos Power to develop small modular reactors,
• 2:07:05
  Speaker 1: They're not just planning for next year, or even the
• 2:07:07
  Speaker 1: next decade, they're positioning themselves for a future where energy
• 2:07:11
  Speaker 1: becomes the ultimate competitive advantage. But perhaps the most fascinating
• 2:07:16
  Speaker 1: aspect of this nuclear AI partnership is how it's driving
• 2:07:20
  Speaker 1: innovation in both fields. The demanding requirements of AI data
• 2:07:25
  Speaker 1: centres are pushing nuclear engineers to develop more efficient, more
• 2:07:30
  Speaker 1: flexible reactor designs. Small modular reactors, with their ability to
• 2:07:36
  Speaker 1: be deployed closer to energy demand and scaled according to need,
• 2:07:41
  Speaker 1: represent a new generation of nuclear technology specifically designed for
• 2:07:45
  Speaker 1: the digital age. These advanced reactors aren't your grandfather's nuclear plants.
• 2:07:51
  Speaker 1: They're designed with inherent safety features that make meltdowns physically impossible,
• 2:07:56
  Speaker 1: not just unlikely. They use new types of fuel that
• 2:08:00
  Speaker 1: are more efficient and produce less waste. Some designs can
• 2:08:05
  Speaker 1: even consume the nuclear waste from older reactors, turning yesterday's
• 2:08:09
  Speaker 1: problems into tomorrow's fuel. The timing of this convergence is remarkable.
• 2:08:16
  Speaker 1: Just as artificial intelligence is reaching a point where it
• 2:08:19
  Speaker 1: can truly transform human society, nuclear technology is experiencing its
• 2:08:24
  Speaker 1: own renaissance. The combination of these two transformative technologies creates
• 2:08:29
  Speaker 1: possibilities that neither could achieve alone. We're looking at a
• 2:08:34
  Speaker 1: future where AI systems, powered by clean nuclear energy, could
• 2:08:38
  Speaker 1: help solve humanity's greatest challenges. Climate modelling powered by AI
• 2:08:44
  Speaker 1: could help us understand and adapt to environmental changes. Medical
• 2:08:49
  Speaker 1: research accelerated by artificial intelligence could lead to breakthroughs in
• 2:08:53
  Speaker 1: treating diseases. Scientific discovery enhanced by AI could unlock new
• 2:08:58
  Speaker 1: technologies we can barely imagine. The infrastructure requirements for this
• 2:09:03
  Speaker 1: future are massive. We're not talking about building a few
• 2:09:08
  Speaker 1: new power plants. We're talking about fundamentally re-imagining how we generate,
• 2:09:14
  Speaker 1: distribute and consume energy. The data centres of the future
• 2:09:19
  Speaker 1: will be more like small cities with their own dedicated
• 2:09:23
  Speaker 1: power sources, cooling systems and support infrastructure. This transformation is
• 2:09:30
  Speaker 1: already beginning. Tech companies are no longer content to simply
• 2:09:36
  Speaker 1: purchase electricity from the grid. They're becoming energy companies themselves
• 2:09:41
  Speaker 1: investing in power generation and developing new technologies to meet
• 2:09:45
  Speaker 1: their needs, the line between technology companies and energy companies
• 2:09:50
  Speaker 1: is blurring, creating new hybrid entities that operate at the
• 2:09:56
  Speaker 1: intersection of computation and power generation. The global implications are profound.
• 2:10:03
  Speaker 1: Countries that successfully combine advanced AI capabilities with reliable nuclear
• 2:10:08
  Speaker 1: power will have significant advantages in the emerging digital economy.
• 2:10:13
  Speaker 1: This isn't just about having faster computers or more efficient algorithms.
• 2:10:18
  Speaker 1: It's about having the energy infrastructure to support sustained innovation
• 2:10:22
  Speaker 1: and development. As we stand at this technological crossroads, the
• 2:10:27
  Speaker 1: partnership between nuclear power and artificial intelligence represents more than
• 2:10:32
  Speaker 1: just a solution to an energy problem. It's a glimpse
• 2:10:35
  Speaker 1: into a future where human ingenuity supported by clean, abundant energy,
• 2:10:41
  Speaker 1: can tackle challenges that seem impossible today. The question isn't
• 2:10:46
  Speaker 1: whether nuclear energy is the solution to endless artificial intelligence power,
• 2:10:51
  Speaker 1: but whether we're ready to embrace the transformative potential of
• 2:10:55
  Speaker 1: this powerful partnership. Looking at where we are in this
• 2:10:59
  Speaker 1: comprehensive exploration of nuclear energy and artificial intelligence, we need
• 2:11:05
  Speaker 1: to address the critical challenges that lie ahead. The promise
• 2:11:09
  Speaker 1: of nuclear-powered AI infrastructure faces significant hurdles that could determine
• 2:11:15
  Speaker 1: whether this solution becomes reality or remains an ambitious vision.
• 2:11:20
  Speaker 1: The construction timeline challenge presents perhaps the most immediate obstacle
• 2:11:25
  Speaker 1: to nuclear-powered AI. While data centres consuming hundreds of megawatts
• 2:11:30
  Speaker 1: of power are being planned for deployment within the next
• 2:11:33
  Speaker 1: few years, nuclear reactors traditionally require six to 12 years
• 2:11:38
  Speaker 1: to build and license. This creates a fundamental mismatch between
• 2:11:42
  Speaker 1: the urgency of AI's energy demands and the deliberate pace
• 2:11:46
  Speaker 1: of nuclear development. Even small modular reactors designed for faster
• 2:11:52
  Speaker 1: deployment still require three to five years from groundbreaking to
• 2:11:57
  Speaker 1: power generation. The question becomes whether the AI industry can
• 2:12:02
  Speaker 1: wait for nuclear solutions. or if interim measures using fossil
• 2:12:06
  Speaker 1: fuels will undermine long-term climate goals. Regulatory frameworks represent another
• 2:12:13
  Speaker 1: formidable challenge. The Nuclear Regulatory Commission has approved only one
• 2:12:18
  Speaker 1: small modular reactor design in United States history, and the
• 2:12:23
  Speaker 1: licensing process for new technologies remains complex and time-consuming. Each
• 2:12:28
  Speaker 1: reactor type requires extensive safety testing, environmental impact assessments, and
• 2:12:34
  Speaker 1: public consultation processes that can add years to deployment timelines.
• 2:12:40
  Speaker 1: The regulatory uncertainty creates risks for tech companies making billion-dollar
• 2:12:45
  Speaker 1: investments in nuclear infrastructure, as approval delays could render projects
• 2:12:50
  Speaker 1: economically unviable. The economic reality of nuclear power adds another
• 2:12:56
  Speaker 1: layer of complexity. While operating costs for nuclear plants are
• 2:13:01
  Speaker 1: relatively low the upfront capital requirements are enormous. Traditional nuclear
• 2:13:07
  Speaker 1: plants cost over $ 10 billion, while even small modular reactors
• 2:13:13
  Speaker 1: require investments of $ 2 to $ 5 billion. Tech companies must
• 2:13:18
  Speaker 1: weigh these massive capital commitments against alternatives like renewable energy
• 2:13:23
  Speaker 1: plus storage, which may offer faster deployment and lower financial risk.
• 2:13:28
  Speaker 1: The economic case for nuclear becomes even more challenging when
• 2:13:32
  Speaker 1: considering the potential for rapid improvements in battery technology that
• 2:13:36
  Speaker 1: could make renewable energy storage more cost effective. Public perception
• 2:13:41
  Speaker 1: and political acceptance remain significant obstacles. Despite growing support for
• 2:13:46
  Speaker 1: nuclear energy among policy makers and climate advocates, local communities
• 2:13:51
  Speaker 1: often resist nuclear projects due to safety concerns and historical
• 2:13:55
  Speaker 1: incidents like Three Mile Island and Fukushima. The not in
• 2:14:00
  Speaker 1: my backyard phenomenon affects nuclear projects more than most other
• 2:14:04
  Speaker 1: energy infrastructure, potentially limiting suitable locations for AI-focused nuclear facilities.
• 2:14:11
  Speaker 1: Tech companies must navigate these political and social challenges while
• 2:14:15
  Speaker 1: building public support for their nuclear initiatives. The skilled workforce
• 2:14:19
  Speaker 1: shortage in nuclear engineering presents another bottleneck. The nuclear industry
• 2:14:24
  Speaker 1: experienced decades of decline, leading to a significant reduction in
• 2:14:28
  Speaker 1: nuclear engineers technicians and construction workers with relevant experience. Building
• 2:14:34
  Speaker 1: the workforce needed to construct and operate hundreds of new
• 2:14:38
  Speaker 1: reactors will require substantial investment in education and training programmes.
• 2:14:44
  Speaker 1: This human capital challenge could limit the speed at which
• 2:14:48
  Speaker 1: nuclear projects can be deployed, regardless of regulatory approval or
• 2:14:53
  Speaker 1: financing availability. Technical integration challenges arise when connecting nuclear plants
• 2:15:00
  Speaker 1: to AI data centres. Unlike traditional power plants that feed
• 2:15:04
  Speaker 1: electricity into the grid, AI-focused nuclear facilities may require direct
• 2:15:09
  Speaker 1: connections to minimise transmission losses and ensure reliability. This creates
• 2:15:15
  Speaker 1: engineering challenges around power conditioning, backup systems and grid integration
• 2:15:20
  Speaker 1: that must be solved on a case-by-case basis. The technical
• 2:15:24
  Speaker 1: complexity increases when considering the need for load following capabilities
• 2:15:29
  Speaker 1: to match nuclear output with fluctuating AI workloads. Safety considerations
• 2:15:35
  Speaker 1: take on new dimensions when nuclear plants are designed specifically
• 2:15:39
  Speaker 1: for AI applications. The concentrated nature of AI workloads means
• 2:15:45
  Speaker 1: that nuclear plant failures could have significant economic impacts beyond
• 2:15:50
  Speaker 1: traditional power outages. The reliability requirements for AI training runs
• 2:15:55
  Speaker 1: that cost millions of dollars to execute may exceed those
• 2:15:59
  Speaker 1: of typical electricity customers, potentially requiring redundant safety systems and
• 2:16:04
  Speaker 1: backup power sources that increase costs and complexity. International competition
• 2:16:10
  Speaker 1: adds urgency to these challenges. China is rapidly expanding its
• 2:16:15
  Speaker 1: nuclear capacity and developing advanced reactor technologies, potentially gaining advantages
• 2:16:21
  Speaker 1: in AI development if their nuclear-powered data centres come online
• 2:16:25
  Speaker 1: faster than Western alternatives. This geopolitical dimension creates pressure to
• 2:16:31
  Speaker 1: accelerate nuclear deployment while maintaining safety standards, creating tension between
• 2:16:36
  Speaker 1: speed and prudence in nuclear development. The waste management question
• 2:16:40
  Speaker 1: becomes more pressing as nuclear capacity expands to serve AI applications.
• 2:16:46
  Speaker 1: While modern reactor designs produce less radioactive waste than previous generations,
• 2:16:51
  Speaker 1: The sheer scale of nuclear deployment required for AI infrastructure
• 2:16:55
  Speaker 1: will generate substantial quantities of spent fuel that must be
• 2:16:58
  Speaker 1: stored and eventually disposed of safely. The lack of permanent
• 2:17:02
  Speaker 1: waste storage solutions in many countries, including the United States,
• 2:17:07
  Speaker 1: creates long-term liabilities that tech companies must consider when investing
• 2:17:11
  Speaker 1: in nuclear power. Climate change itself presents challenges for nuclear infrastructure.
• 2:17:17
  Speaker 1: Rising sea levels, extreme weather events, and changing precipitation patterns
• 2:17:23
  Speaker 1: affect nuclear plant design and operation, cooling water availability, which
• 2:17:28
  Speaker 1: is crucial for nuclear plant operation, may become more variable
• 2:17:32
  Speaker 1: due to droughts and changing river flows. These climate impacts
• 2:17:36
  Speaker 1: create additional engineering challenges and potential reliability risks for nuclear-powered
• 2:17:42
  Speaker 1: AI infrastructure. The intermittency question reveals a paradox in nuclear-powered AI.
• 2:17:49
  Speaker 1: While nuclear plants traditionally operate best at constant output, AI
• 2:17:54
  Speaker 1: workloads may vary significantly based on training schedules, model deployment
• 2:17:58
  Speaker 1: patterns and user demand. This mismatch between nuclear plant preferences
• 2:18:04
  Speaker 1: for steady operation and AI workload variability creates challenges for
• 2:18:09
  Speaker 1: efficient utilization of nuclear capacity. Solutions may require sophisticated load balancing,
• 2:18:17
  Speaker 1: energy storage, or alternative uses for excess nuclear capacity during
• 2:18:23
  Speaker 1: low AI demand periods. Despite these challenges, the convergence of
• 2:18:28
  Speaker 1: climate necessity, technological advancement, and economic incentives continues to drive
• 2:18:34
  Speaker 1: innovation in nuclear-powered AI infrastructure. The companies and countries that
• 2:18:40
  Speaker 1: successfully navigate these obstacles may gain significant advantages in the
• 2:18:44
  Speaker 1: artificial intelligence revolution, while those that fail to address energy
• 2:18:48
  Speaker 1: constraints may find their AI ambitions limited by power availability.
• 2:18:54
  Speaker 1: The next few years will likely determine whether nuclear energy
• 2:18:58
  Speaker 1: becomes the backbone of AI infrastructure or whether alternative solutions
• 2:19:02
  Speaker 1: emerge to meet this unprecedented energy demand. Looking at our
• 2:19:08
  Speaker 1: cosmic journey through the universe's grand story, we arrive at
• 2:19:12
  Speaker 1: our final movement, the culmination of everything nuclear energy represents
• 2:19:17
  Speaker 1: for humanity's artificial intelligence ambitions. The numbers are staggering and undeniable.
• 2:19:24
  Speaker 1: By 2030, data centers will consume between 325 and 580
• 2:19:28
  Speaker 1: terawatt hours of electricity annually in the United States alone.
• 2:19:35
  Speaker 1: To put this in perspective, that's approaching the total electricity
• 2:19:39
  Speaker 1: consumption of entire nations like Germany or Japan. The artificial
• 2:19:44
  Speaker 1: intelligence revolution isn't just changing how we work, communicate, and live.
• 2:19:50
  Speaker 1: It's fundamentally reshaping global energy demand on a scale we've
• 2:19:55
  Speaker 1: never witnessed before. Traditional approaches to powering this digital transformation
• 2:20:00
  Speaker 1: are proving inadequate. Solar and wind, for all their environmental benefits,
• 2:20:05
  Speaker 1: face the fundamental limitation of intermittency. The sun doesn't always shine,
• 2:20:11
  Speaker 1: the wind doesn't always blow, yet artificial intelligence systems require constant,
• 2:20:16
  Speaker 1: uninterrupted power to function. Battery storage technology, while improving, remains
• 2:20:22
  Speaker 1: expensive and requires massive physical footprints that make it impractical
• 2:20:26
  Speaker 1: for the scale of energy storage needed to support large
• 2:20:29
  Speaker 1: data centres running entirely on renewables. the fossil fuel alternative
• 2:20:35
  Speaker 1: presents its own challenges. Natural gas plants can provide reliable
• 2:20:40
  Speaker 1: baseload power, but they come with significant carbon emissions that
• 2:20:44
  Speaker 1: conflict with the tech industry's ambitious net-zero commitments. Coal is
• 2:20:50
  Speaker 1: increasingly being phased out due to environmental concerns and economics.
• 2:20:57
  Speaker 1: The stark reality is that neither renewable nor fossil fuel
• 2:21:01
  Speaker 1: solutions alone can meet the unique demands of artificial intelligence infrastructure.
• 2:21:08
  Speaker 1: This is where nuclear energy emerges as the singular solution
• 2:21:12
  Speaker 1: that addresses every critical requirement simultaneously. Nuclear power plants provide
• 2:21:18
  Speaker 1: the most reliable baseload electricity generation available, operating at capacity
• 2:21:23
  Speaker 1: factors exceeding 90%. They can run continuously for months without interruption.
• 2:21:30
  Speaker 1: providing the stable, predictable power that AI systems absolutely require.
• 2:21:36
  Speaker 1: Unlike renewable sources, nuclear power generation is not subject to
• 2:21:40
  Speaker 1: weather conditions, seasonal variations, or time of day fluctuations. The
• 2:21:45
  Speaker 1: environmental credentials of nuclear energy are compelling. Modern nuclear plants
• 2:21:50
  Speaker 1: produce virtually no greenhouse gas emissions during operation, making them
• 2:21:55
  Speaker 1: ideal for companies seeking to minimize their carbon footprint. while
• 2:21:58
  Speaker 1: dramatically scaling their energy consumption. A single nuclear plant can
• 2:22:04
  Speaker 1: provide the same amount of clean electricity as thousands of
• 2:22:08
  Speaker 1: wind turbines or millions of solar panels, but with a
• 2:22:11
  Speaker 1: fraction of the land use and none of the intermittency issues.
• 2:22:16
  Speaker 1: From an economic perspective, nuclear energy offers long-term cost stability
• 2:22:22
  Speaker 1: that other sources cannot match. While the upfront capital costs
• 2:22:27
  Speaker 1: are significant, Nuclear plants have operational lifespans of 60 to
• 2:22:32
  Speaker 1: 80 years, with fuel costs representing only a small fraction
• 2:22:36
  Speaker 1: of total operating expenses. This provides predictable energy costs over decades,
• 2:22:43
  Speaker 1: which is crucial for companies making long-term infrastructure investments in
• 2:22:48
  Speaker 1: artificial intelligence capabilities. The technological advancement in nuclear reactor design
• 2:22:54
  Speaker 1: makes this solution even more attractive Small modular reactors represent
• 2:23:00
  Speaker 1: a paradigm shift in nuclear technology, offering enhanced safety features,
• 2:23:05
  Speaker 1: shorter construction timelines and the ability to scale capacity incrementally.
• 2:23:11
  Speaker 1: These reactors can be located closer to data centres, reducing
• 2:23:15
  Speaker 1: transmission losses and improving grid reliability. Their modular design allows
• 2:23:22
  Speaker 1: for mass production, which should drive down costs and construction
• 2:23:25
  Speaker 1: times compared to traditional, large-scale nuclear plants. Major technology companies
• 2:23:31
  Speaker 1: have already recognised nuclear energy's potential and are making substantial investments.
• 2:23:38
  Speaker 1: Microsoft's agreement to restart Three Mile Island, Amazon's multi-billion dollar
• 2:23:43
  Speaker 1: commitment to small modular reactors, and Google's partnership with Kairos
• 2:23:47
  Speaker 1: Power demonstrate that the industry's leading companies view nuclear energy
• 2:23:52
  Speaker 1: as essential to their artificial intelligence strategies. These aren't experimental investments.
• 2:23:59
  Speaker 1: They're strategic commitments to ensure reliable, clean power for the
• 2:24:03
  Speaker 1: next generation of AI infrastructure. The safety record of modern
• 2:24:07
  Speaker 1: nuclear technology addresses many historical concerns. Advanced reactor designs incorporate
• 2:24:14
  Speaker 1: passive safety systems that automatically shut down reactors without human
• 2:24:19
  Speaker 1: intervention or external power sources. The new generation of reactors
• 2:24:24
  Speaker 1: cannot experience the types of accidents that occurred at Chernobyl
• 2:24:28
  Speaker 1: or Fukushima. Small modular reactors, in particular, are designed with
• 2:24:34
  Speaker 1: inherent safety features that make catastrophic failures physically impossible. Regulatory
• 2:24:41
  Speaker 1: frameworks are evolving to accommodate these new technologies. The Nuclear
• 2:24:45
  Speaker 1: Regulatory Commission has approved the first small modular reactor design
• 2:24:49
  Speaker 1: in the United States, and streamlined licensing processes are being
• 2:24:54
  Speaker 1: developed specifically for advanced reactor technologies. International cooperation on nuclear
• 2:25:00
  Speaker 1: safety standards and best practices continues to improve, creating a
• 2:25:06
  Speaker 1: global framework for safe nuclear deployment. The scalability of nuclear
• 2:25:10
  Speaker 1: energy solutions matches the scope of artificial intelligence's growth trajectory.
• 2:25:16
  Speaker 1: As AI systems become more sophisticated and widespread, their energy
• 2:25:21
  Speaker 1: requirements will continue to expand exponentially. Nuclear power plants can
• 2:25:27
  Speaker 1: be sized to meet these growing demands, from small modular
• 2:25:30
  Speaker 1: reactors serving individual data centres to large plants powering entire
• 2:25:35
  Speaker 1: regions of AI infrastructure. grid integration benefits extend beyond just
• 2:25:42
  Speaker 1: providing power. Nuclear plants can provide essential grid services like
• 2:25:48
  Speaker 1: frequency regulation and voltage support that help maintain electrical system stability.
• 2:25:55
  Speaker 1: As more variable renewable energy sources are added to the grid,
• 2:25:59
  Speaker 1: nuclear power provides the stable foundation that keeps the entire
• 2:26:03
  Speaker 1: electrical system functioning reliably. The geopolitical implications of nuclear energy
• 2:26:10
  Speaker 1: for AI infrastructure are significant. Countries with robust nuclear capabilities
• 2:26:17
  Speaker 1: will have competitive advantages in artificial intelligence development and deployment.
• 2:26:23
  Speaker 1: Nuclear energy provides energy security that reduces dependence on volatile
• 2:26:28
  Speaker 1: fossil fuel markets and foreign energy suppliers. This energy independence
• 2:26:35
  Speaker 1: is crucial for maintaining technological leadership in artificial intelligence. Innovation
• 2:26:41
  Speaker 1: in nuclear technology continues to accelerate, driven partly by the
• 2:26:45
  Speaker 1: demands of AI infrastructure. Advanced reactor designs, improved fuel cycles
• 2:26:52
  Speaker 1: and enhanced safety systems are all benefiting from increased investment
• 2:26:57
  Speaker 1: and research attention. The fusion energy breakthrough achieved at Lawrence
• 2:27:02
  Speaker 1: Livermore National Laboratory, while still years from commercial deployment, represents
• 2:27:07
  Speaker 1: the ultimate long-term solution for unlimited clean energy. The economic
• 2:27:12
  Speaker 1: multiplier effects of nuclear energy investment extend far beyond the
• 2:27:16
  Speaker 1: energy sector. Nuclear plants create high paying jobs, stimulate local economies,
• 2:27:21
  Speaker 1: and support advanced manufacturing capabilities. The skills and technologies developed
• 2:27:27
  Speaker 1: for nuclear energy applications often find uses in other industries,
• 2:27:31
  Speaker 1: creating broader economic benefits. Looking toward the future, the convergence
• 2:27:36
  Speaker 1: of artificial intelligence and nuclear energy represents a fundamental shift
• 2:27:40
  Speaker 1: in how we approach both technology and energy policy. The
• 2:27:45
  Speaker 1: massive computational requirements of advanced AI systems, from training large
• 2:27:50
  Speaker 1: language models to running real-time inference for billions of users,
• 2:27:54
  Speaker 1: demand energy solutions that can scale without compromising reliability or
• 2:27:59
  Speaker 1: environmental goals. The timeline for this transformation is compressed. Unlike
• 2:28:07
  Speaker 1: previous energy transitions that unfolded over decades, the AI energy
• 2:28:12
  Speaker 1: crisis is immediate and accelerating. Companies cannot wait for perfect solutions.
• 2:28:18
  Speaker 1: They need reliable, scalable, clean energy now. Nuclear energy provides
• 2:28:23
  Speaker 1: the only realistic path to meet these urgent requirements while
• 2:28:26
  Speaker 1: maintaining long-term sustainability goals. Nuclear energy isn't just a solution
• 2:28:32
  Speaker 1: to artificial intelligence's power demands. It's the foundation that will
• 2:28:36
  Speaker 1: enable the next phase of human technological advancement. The same reliable,
• 2:28:42
  Speaker 1: abundant energy that powers data centres can also support other
• 2:28:46
  Speaker 1: emerging technologies like quantum computing, advanced manufacturing and space exploration.
• 2:28:52
  Speaker 1: Nuclear energy creates the energy abundance necessary for continued innovation
• 2:28:57
  Speaker 1: and economic growth. The question isn't whether nuclear energy will
• 2:29:01
  Speaker 1: power artificial intelligence. It's how quickly we can deploy these
• 2:29:06
  Speaker 1: solutions to meet the rapidly growing demand. Every month of
• 2:29:10
  Speaker 1: delay in nuclear deployment means more reliance on fossil fuels,
• 2:29:14
  Speaker 1: higher energy costs and greater environmental impact. The technology exists,
• 2:29:20
  Speaker 1: the economics are favourable and the regulatory frameworks are developing.
• 2:29:25
  Speaker 1: What's needed now is the will to act decisively. In conclusion,
• 2:29:31
  Speaker 1: nuclear energy represents the most viable, scalable, and sustainable solution
• 2:29:36
  Speaker 1: to artificial intelligence's unprecedented energy demands. It provides the reliability, capacity,
• 2:29:44
  Speaker 1: and environmental benefits that no other energy source can match.
• 2:29:48
  Speaker 1: The major technology companies driving the AI revolution have already
• 2:29:53
  Speaker 1: recognized this reality and are making substantial investments in nuclear power.
• 2:29:58
  Speaker 1: The convergence of artificial intelligence and nuclear energy will define
• 2:30:03
  Speaker 1: the next chapter of human technological progress, creating opportunities for innovation,
• 2:30:09
  Speaker 1: economic growth and environmental stewardship that seemed impossible just a
• 2:30:14
  Speaker 1: few years ago. The future of artificial intelligence depends on abundant, reliable,
• 2:30:20
  Speaker 1: clean energy. Nuclear power provides the foundation for that future.
• 2:30:25
  Speaker 1: ensuring that the transformative potential of artificial intelligence can be
• 2:30:29
  Speaker 1: realised without compromising our environmental goals or economic stability. The
• 2:30:34
  Speaker 1: time for action is now, and nuclear energy is the
• 2:30:38
  Speaker 1: answer to powering humanity's artificial intelligence ambitions. And so, we
• 2:30:44
  Speaker 1: return from the crossroads of energy and intelligence, a future
• 2:30:48
  Speaker 1: still unwritten but filled with quiet choices that echo across generations.
• 2:30:54
  Speaker 1: Whether or not nuclear power becomes the heartbeat of AI,
• 2:30:57
  Speaker 1: one thing is clear. How we power the future will
• 2:31:01
  Speaker 1: shape who we become in it. Thank you for staying
• 2:31:06
  Speaker 1: with me tonight. This is Boring Science. I wish you
• 2:31:09
  Speaker 1: a very good night. Sleep well and keep wondering. Cheers.