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