Latest / SleepWise: Learn Softly, Sleep Deeply / NVIDIA: From 1993 to the AI Age | A SleepWise Story
Transcript
- 0:00Tonight we drift back to the early 1990s, when computers
- 0:06still felt like quiet furniture, beige towers under desks, soft
- 0:13fans turning, a monitor glowing with a kind of patient effort.
- 0:19The world on those screens was mostly flat, made of windows and
- 0:24icons, and yet you could already sense a hunger for depth, for
- 0:30movement, for something that looked more like the world
- 0:34outside the room. In cinemas, new computer imagery
- 0:39was beginning to shimmer and convince dinosaurs breathing in
- 0:45moonlit rain, impossible scenes made believable by patient
- 0:51calculation. In arcades and living rooms,
- 0:55games were leaning toward illusion corridors that seemed
- 1:01to stretch away creatures that lurked in shadow, textures that
- 1:06hinted at stone and metal even if they were made of clever
- 1:10pixels. On APC, though, the central
- 1:14processor did almost everything, and it was tired, asked to
- 1:20calculate logic, sound input, and then, on top of it all, the
- 1:28heavy math of drawing A3 dimensional scene.
- 1:33The result was often a kind of compromise, A stutter between
- 1:38frames, a world that wanted to be smooth but could only manage
- 1:44a rough sketch of itself. And in that moment, in Silicon
- 1:49Valley air that smelled faintly of solder and coffee, three
- 1:54engineers began circling a simple idea.
- 1:59What if the work of seeing could be separated from the work of
- 2:04thinking? What if there were a dedicated
- 2:08chip, built for repetition and rhythm, doing many small visual
- 2:13calculations at once, so the main processor could breathe?
- 2:19And the images could flow. In 1993, Jensen Huang, Chris
- 2:27Malakowski and Curtis Prim started a new company with that
- 2:33question tucked inside it. And they gave it a name that
- 2:37sounded like a bright place beyond the horizon.
- 2:42NVIDIA the story says their first serious conversations
- 2:48happened in an ordinary diner over breakfast, the kind of
- 2:53place where the coffee keeps coming and big ideas can feel
- 2:58strangely practical. They were not alone in chasing
- 3:02the future. Not at all.
- 3:05Dozens of companies were building graphics chips, each
- 3:09trying to become the one that mattered, each racing to survive
- 3:15the next product cycle. So from the beginning, there was
- 3:20pressure in the background, like a steady clock.
- 3:25Deadlines, prototypes, decisions that could not be undone once
- 3:31the silicon was made. But there is a particular kind
- 3:35of calm that comes from being early, from committing to
- 3:39something before it is obvious, before it has a market with a
- 3:45neat label and a confident number attached to it.
- 3:50To night. As the world slows.
- 3:53We'll follow that quiet commitment from the first
- 3:57uncertain designs to the moment a new kind of processor began to
- 4:03lift the weight of images from the shoulders of the CPU.
- 4:08We won't rush, and you don't have to remember a thing.
- 4:13Just listen for the gentle hum of machines and the way a few
- 4:18patient decisions can, overtime, reshape what the world is able
- 4:24to imagine. In the beginning, the future did
- 4:28not arrive as a clean, confident breakthrough, but as a messy
- 4:34pile of possibilities. Sound, video, controller ports
- 4:41and graphics all tugging at the same small chip.
- 4:46Nvidia's first attempt in the mid 1990s tried to be everything
- 4:51at once, like a multi tool carried into a storm, and for a
- 4:58moment it felt clever, even elegant on paper.
- 5:02But the world outside the lab had its own preferences, and
- 5:07software developers were gathering around new standards,
- 5:11choosing the simplest shapes to build their 3D worlds triangles,
- 5:17predictable and fast. When the chip arrived, the
- 5:22market did what markets often do It shrugged, it moved on, and it
- 5:29returned boxes that had once seemed like a promise.
- 5:33Inside the company, the room grew quieter, not with despair
- 5:39but with focus, because a young team learns quickly that silicon
- 5:45is unforgiving, and there is no undo once the design is baked.
- 5:52They were surrounded by competitors, too, a whole
- 5:56crowded field of graphics companies, each racing to make
- 6:01the next card, each hoping to be the name printed on the box.
- 6:06It was the kind of era where magazines compared frame rates
- 6:11the way people once compared engines, and gamers could feel
- 6:16the difference between smooth motion and a world that
- 6:21stumbled. So NVIDIA narrowed its gaze,
- 6:25letting go of the octopus dream and choosing A simpler goal.
- 6:31Make one thing work beautifully and make it arrive on time.
- 6:38In those months, time became a texture, like sand in a pocket.
- 6:44Because there is a special pressure when you know you have
- 6:49only one more product cycle to earn your right to continue.
- 6:54And yet there was also craft engineers running long
- 6:59simulations, testing a design before it ever touched a
- 7:05factory, listening for errors the way you might listen for a
- 7:10Creek in an old house at night. When the new chip finally
- 7:15landed, it worked, and it sold, and it did what it was meant to
- 7:20do. It helped images move more
- 7:23smoothly, and it gave the company another morning.
- 7:28Tonight, we can hold that moment gently, because it's not triumph
- 7:34or failure that defines a long story, but endurance, the steady
- 7:41choice to try again with clearer eyes.
- 7:46It was around the time when games like Quake were teaching
- 7:50the eye to expect depth, and when each new graphics card felt
- 7:56like a new pair of glasses, the same world, but sharper.
- 8:02The chip's name mattered less than its timing, but people
- 8:07remember it as one of the Rivacards, an unromantic word
- 8:12that none the less became a bridge back to stability.
- 8:16And once the bridge was built, the pace changed.
- 8:20Because survival has a way of sharpening decisions, the team
- 8:27could now iterate faster, learn faster, and keep the fan hum
- 8:32steady through the night. With stability came a quiet kind
- 8:37of confidence, and with confidence came permission to
- 8:42think bigger. Not louder, just deeper.
- 8:46In 1999, NVIDIA introduced a new card under a new name, G Force,
- 8:54and with it they offered a phrase that would stick the
- 8:58graphics processing unit, or GPU.
- 9:03It was, in a sense, a small act of renaming the world.
- 9:09Because until then, graphics were treated as an accessory, a
- 9:15helper, something the CPU could manage if it had enough time.
- 9:20But time was exactly what games did not have, because a moving
- 9:26world cannot pause while the processor catches its breath,
- 9:31and the eye will always notice the hesitation.
- 9:35So the GPU was designed to take on the repeated work of
- 9:40transforming 3D shapes, calculating light, preparing
- 9:45triangles and pushing pixels, A specialized rhythm done in
- 9:52parallel, like many hands folding the same kind of paper.
- 9:57This was not only about beauty, though beauty mattered, but
- 10:02about freeing the CPU to do what it did best.
- 10:06Logic decisions, the subtle choreography of a game's rules.
- 10:13If you played in those years, you may remember the feeling of
- 10:17a new graphics card, like a secret upgrade to reality
- 10:23corridors in Quake 3 that suddenly felt smoother, arenas
- 10:29that shimmered without tearing. The GPU also carried something
- 10:34else, almost hidden at first. It was becoming more
- 10:39programmable, less like a fixed tool and more like a small
- 10:45machine that could learn new tricks.
- 10:48That shift would take years to fully unfold, but you can
- 10:53already sense it here, like a door cracked open in a quiet
- 10:58hallway. Around this same time, Nvidia's
- 11:02chips began to appear not only in PCs but in the broader
- 11:07culture of gaming hardware, as console makers looked for
- 11:12graphics power that felt both affordable and advanced.
- 11:18A console is, in its own way, a promise of consistency, the same
- 11:25experience for everyone, the same frames, the same light, the
- 11:30same reliable hum. And NVIDIA wanted to be part of
- 11:36that steadiness. So the company kept moving,
- 11:40generation by generation, each one a little faster, a little
- 11:45more capable, each one teaching developers to expect more from
- 11:51what a screen could hold. And beneath it all was a calm
- 11:56idea that would later become enormous, that some work is
- 12:02better done by many simple processors acting together,
- 12:08rather than one processor doing everything alone.
- 12:12When NVIDIA called it a GPU, it was partly a statement of
- 12:18ambition. Not just a faster graphics chip,
- 12:23but a new class of processor built to do a certain kind of
- 12:28math with steady repetition. It was the beginning of a long
- 12:33relationship between silicon and imagination, where artists and
- 12:39engineers met in the middle, one drawing what they wished to see,
- 12:46the other finding a way to compute it in time.
- 12:51In the early 2000's, the world of games grew more cinematic,
- 12:57not because stories became louder, but because light and
- 13:02texture became more convincing. A wall could look like stone
- 13:08instead of a pattern, a sky could fade gently instead of
- 13:12switching colors, and shadows could suggest distance the way
- 13:18evening does. This happened through a quiet
- 13:22evolution called programmability, small pieces of
- 13:27code called shaders that told the GPU how to color a pixel or
- 13:33bend a surface like instructions whispered to a brush.
- 13:37For developers, this was a new palette, and for players it was
- 13:43a gradual shift in expectation, as if the mind learned to
- 13:49believe the screen a little more each year.
- 13:53NVIDIA and its rivals pushed each other forward, and
- 13:58generation by generation, the numbers climbed.
- 14:03More transistors, more memory bandwidth, more parallel work
- 14:08per heartbeat. The cultural world around it
- 14:12changed, too, because gaming moved from basements to
- 14:17bedrooms, from a private hobby to a shared language.
- 14:22You might remember land parties, the glow of monitors in a dim
- 14:27room, the soft clatter of keyboards and the quick laughter
- 14:33between rounds. Later came broadband, and then
- 14:38the strange comfort of watching someone else play, as if the act
- 14:43of play could become a story you'd listen to while doing
- 14:47something else. Platforms like YouTube and
- 14:51Twitch would eventually make that watching feel effortless,
- 14:56but even before they were household names, the need was
- 15:00already there, the desire to share a moment in real time.
- 15:07A GPU helped with that as well, because the same parallel power
- 15:13that drew a game could also help compress it into a stream,
- 15:18turning private pixels into a public broadcast.
- 15:23And slowly, without any single turning point, games became
- 15:29arenas and arenas became events, and events became a kind of
- 15:36modern gathering, lit by screens instead of stadium lights.
- 15:41Through it all, NVIDIA kept refining the craft of fast
- 15:46drawing, building not only chips but a relationship with
- 15:50developers, tools, drivers and the gentle promise that next
- 15:57year's hardware would make new ideas possible.
- 16:01In that era, names like DirectX and Open GL floated through
- 16:07conversations not as dry acronyms, but as common roads
- 16:12that everyone agreed to travel so games could run on many
- 16:17machines. Nvidia's chips found their way
- 16:21into consoles as well, and if you ever held an original Xbox
- 16:27controller, heavy and sure in the hands, there was a bit of
- 16:32Nvidia's work helping that world feel smooth.
- 16:37Even when you didn't know the chip's name, you could sense its
- 16:42effect, The steadier frame rate, the richer scenes, the feeling
- 16:48that the machine was keeping up with your intention.
- 16:51All of this was still graphics officially, yet it carried a
- 16:57hidden lesson that parallel computation, done well, could
- 17:02make complex experiences feel simple.
- 17:07At some point, usually in the middle of a long technology
- 17:12story, people begin to use a tool for something it was not
- 17:18built to do. They do it quietly at first, in
- 17:22labs and late nights, not to rebel, but to see what happens
- 17:27when you turn a familiar object slightly and look at it from
- 17:32another angle. A GPU, after all, was already a
- 17:36machine for repeating simple math very quickly.
- 17:40And many scientific problems are also made of repeated math,
- 17:44stacked like waves. So researchers began to wonder,
- 17:50if a GPU can apply light to millions of pixels, could it
- 17:55also apply equations to millions of data points?
- 18:01In the early days, this was awkward, like writing poetry
- 18:05with a screwdriver, because programming a GPU meant
- 18:11pretending your problem was a graphics problem.
- 18:16You would encode numbers as colors, calculations as
- 18:20textures, and you would work through the constraints with
- 18:25patient creativity. But when it worked, it felt like
- 18:29a small miracle. A simulation that once took days
- 18:35could finish in hours, and the computer seemed to breathe in a
- 18:39new way. NVIDIA noticed these
- 18:43experiments, and instead of dismissing them as edge cases,
- 18:49they listened. Because listening is often how a
- 18:54company discovers its next life, the question became sharper,
- 19:01What if we made it normal to program the GPU directly, not as
- 19:06a graphics trick, but as a computing platform?
- 19:11That idea would require software tools, a new language of sorts,
- 19:18and a willingness to invite people into the chips inner
- 19:23rhythm. It also required humility,
- 19:27because it meant admitting that the future might not be only
- 19:30about games, even if games had built the company.
- 19:35So as the world kept playing, NVIDIA began preparing another
- 19:41doorway, one that would lead from rendering scenes to
- 19:45accelerating science. Tonight we can feel that doorway
- 19:52as a soft draft in a hallway, the hint of a larger room
- 19:57beyond, filled with equations, data, and the steady hum of
- 20:02parallel machines. ACPU is like a careful solo
- 20:08thinker, stepping through tasks one by one, while a GPU is more
- 20:15like a choir. Many small voices each singing a
- 20:20simple note at the same time. When the notes line up, the
- 20:25result is not just faster but different, because problems that
- 20:31were too large to attempt can suddenly be held in memory and
- 20:37worked on patiently. Think of medical images are
- 20:41reconstructed from many slices, or seismic readings turned into
- 20:46maps of hidden rock, or molecules simulated as they fold
- 20:51and bump in liquid, all of it made of repeating math, all of
- 20:58it hungry for parallel effort. In those years, the people doing
- 21:03this work often built their own clusters from ordinary graphics
- 21:08cards stacked in metal frames with extra fans, turning gaming
- 21:15hardware into homemade supercomputers.
- 21:19It was an improvised kind of progress, but it carried a clear
- 21:24message the GPU was ready to become something more.
- 21:29In 2006, NVIDIA opened that doorway with a name that sounded
- 21:37like a quiet spark, CUDA. CUDA was a way of speaking to
- 21:44the GPU more directly, letting programmers write ordinary
- 21:49looking code and letting the GPU translate it into thousands of
- 21:54tiny parallel steps. It did not change the laws of
- 21:59physics, but it changed who could use them, because you no
- 22:04longer had to disguise your work as graphics to access the GPU's
- 22:11power. Now a researcher could say in
- 22:14plain terms, take this matrix, apply this operation, repeat it
- 22:21across the data and do it all at once.
- 22:26The GPU in response would not think harder, but it would work
- 22:32wider, spreading the task across many cores, each one doing a
- 22:38small piece with steady patience.
- 22:41NVIDIA began to build products specifically for this world.
- 22:47GPU's intended for data centers and labs, often sold not as
- 22:52shiny gaming cards, but as accelerators, quiet engines
- 23:00bolted into racks. This was the beginning of a
- 23:04pattern that would repeat again and again.
- 23:08Hardware paired with an ecosystem, Software paired with
- 23:12community. The chip and the tools growing
- 23:16together. Scientists used CUDA to speed up
- 23:20simulations, engineers used it to test designs, and
- 23:26universities began teaching the new approach, training students
- 23:31to think in parallel. If you stood in one of those
- 23:35labs, you might hear the same familiar sound as in a gaming
- 23:40PC. Fans, airflow, a constant hum.
- 23:46But the work inside was different.
- 23:49Instead of drawing Dragons or stadiums, the GPU's were tracing
- 23:55protein folds, reconstructing CT scans, simulating fluids or
- 24:03searching patterns in noisy signals.
- 24:06And still, nothing about it felt flashy, because the best
- 24:12infrastructure rarely does. It simply makes the impossible
- 24:18feel routine, and the routine feel gentle.
- 24:23What began as an experiment at the edges was becoming a second
- 24:28foundation beneath the company, and NVIDIA was learning how to
- 24:33stand on both the playful world of games and the patient world
- 24:40of science. In some places, professors built
- 24:44clusters from off the shelf cards, lining them up like books
- 24:50on a shelf, cooling them with box fans, and calling it a
- 24:56personal supercomputer. And as these experiments spread,
- 25:01large supercomputing centers began to take notice.
- 25:06Because the math was persuasive. For certain workloads, one GPU
- 25:13could replace many CPUs. Before long, the fastest
- 25:19machines in the world were mixing processes the way a good
- 25:24kitchen mixes tools. CPO's for general work, GPU's
- 25:33for the heavy parallel stirring, each doing what it did best.
- 25:38The idea of accelerated computing began to sound less
- 25:44like a slogan and more like a new normal, a calm shift in how
- 25:50the world's hardest problems would be approached.
- 25:54By the early 2000 tens, another quiet community was gathering
- 25:59around GPU's, not because of graphics but because of
- 26:03learning. Machine learning had existed for
- 26:07a long time like a library of ideas waiting for the right kind
- 26:11of electricity, but it often felt slow and limited by the
- 26:17available compute. Deep neural networks in
- 26:20particular needed two things to truly flourish large data sets
- 26:27and an immense amount of repeated math.
- 26:31That repeated math was mostly linear algebra, the kind of
- 26:37multiplication and addition that can be broken into many small
- 26:42pieces and done in parallel, like raindrops on a window.
- 26:48In 2012, a group of researchers trained a deep network to
- 26:53recognize images in a major competition, and they used GPUs
- 26:59to do it, because without GPUs it would have taken far too
- 27:06long. The results were not a small
- 27:09improvement. They were a clear leap, a sudden
- 27:14drop in error that made the field look up and pay attention.
- 27:19It was one of those moments that feels later, like a hinge, a
- 27:26door swinging open, revealing a new room in the House of
- 27:31technology. After that, more people tried
- 27:35deep learning because the path was now visible.
- 27:39If you can train the network faster, you can iterate, and
- 27:45iteration is how intuition becomes progress.
- 27:51NVIDIA, watching this, understood something important.
- 27:58The Jeep AU had become a general engine for modern pattern
- 28:03recognition. A network learns by adjusting
- 28:07countless tiny weights, nudging them again and again, and the
- 28:12GPU's parallel structure made that nudging feasible at scale.
- 28:19The company began to serve this emerging world with more
- 28:23intention, better software libraries, specialized hardware
- 28:29features, whole systems designed to train models efficiently.
- 28:36And while the public still thought of GPU's as gaming parts
- 28:42in quiet server rooms, the story was shifting because the same
- 28:48silicon that drew game worlds was now helping machines see.
- 28:54You could almost feel the cultural change approaching.
- 28:58Cameras everywhere, data everywhere, and the growing
- 29:02sense that computers might learn from experience rather than only
- 29:08follow explicit rules. In the sleep wise calm of this
- 29:13night, we can hold that change softly, not as a sudden
- 29:19explosion, but as a slow dawn, the kind that brightens the sky
- 29:26before anyone notices. The benchmark was called
- 29:30Imagenet, and its millions of labeled pictures were like a
- 29:35training ground for digital vision, a way to measure whether
- 29:40a machine could recognize a cat, a street sign, a face.
- 29:46The network that one often remembered as Alex Net was
- 29:51trained on consumer GPU's, the kind someone could buy for a
- 29:57gaming PC. And that detail mattered.
- 30:03It meant the new era did not require a secret government
- 30:07machine, only the right approach.
- 30:10Soon the internet's endless photos and videos became more
- 30:15than content. They became data, and the GPU
- 30:20became the quiet mill that could grind that data into learned
- 30:26ability. Once a new door opens, the world
- 30:29tends to rush through it. But NVIDIA moved in its own way,
- 30:35deliberate building not a single product but an entire path.
- 30:42The company began to speak more often about accelerated
- 30:45computing, about GPU's and data centers, about a future where
- 30:51training a model would be as common as hosting a website.
- 30:56To make that future easier, they built complete machines designed
- 31:01for AI boxes of GPU's tuned to work together like a small
- 31:08orchestra in a single chassis. These systems had names like
- 31:13DGX, but the names were less important than the feeling they
- 31:18offered that AI training could be packaged, repeatable, and
- 31:24reliable. They also refined the chips
- 31:27themselves, adding features meant for the new workload
- 31:32circuits that speed up the matrix math of neural networks,
- 31:37doing the same kind of multiplication again and again
- 31:41with less energy. People sometimes called those
- 31:45features tenser cores, but you can think of them as a gentle
- 31:50shortcut, a way of making the most common steps in learning
- 31:55faster and smoother at the same time.
- 31:59Gaming did not disappear. It continued vivid and bright.
- 32:05And NVIDIA kept improving the visual magic too, even
- 32:11introducing ray tracing, A technique that imitates how
- 32:16light bounces. It was an unusual balance, 1
- 32:21foot in play, 1 foot in research, and both supported by
- 32:26the same foundational idea of parallel work in server rooms.
- 32:32The GPU's began to cluster in large groups, linked by fast
- 32:38connections, because training bigger models meant coordinating
- 32:44many chips as if they were one machine.
- 32:48This is where the story begins to feel less like a consumer
- 32:52tale and more like infrastructure, the kind of
- 32:57technology you rarely see but you depend on all the same.
- 33:03And yet even infrastructure has a mood, and the mood here was
- 33:09steady ambition, the patient confidence that if you build the
- 33:15right platform, others will build wonders on top of it.
- 33:19So NVIDIA kept doing what it had learned in its earliest years.
- 33:26Iterate. Refine and listen.
- 33:30While the hum of computation grew a little louder in the
- 33:35background of the modern world, cloud companies began offering
- 33:39GPU's by the hour, as if parallel compute were a kind of
- 33:44calm utility. And researchers rented time the
- 33:49way earlier generations rented film cameras.
- 33:52Software frameworks like Tensorflow and π Torch made it
- 33:57easier to describe learning models, and beneath them,
- 34:03Nvidia's libraries helped the GPU do the heavy lifting with
- 34:09quiet efficiency. In this way, the GPU became less
- 34:15like a component and more like a place, an environment where
- 34:21ideas could run, repeat, and improve while humans slept and
- 34:27the machines kept training. The world was entering an era
- 34:31where progress often meant more data, more compute, and more
- 34:37patience, and NVIDIA had positioned itself at the center
- 34:43of that triangle. As AI models grew larger,
- 34:47something surprising became obvious.
- 34:50It wasn't enough to have fast chips.
- 34:53You also needed fast conversation between chips.
- 34:57Training a large model is like teaching a choir a new song.
- 35:03Each section can practice its part, but they must hear each
- 35:07other to stay in time. So the network inside a data
- 35:12center became as important as the processors themselves, and
- 35:17NVIDIA began to think about the whole system, not just the GPU.
- 35:23That thinking LED them to buy Mellanox, a company known for
- 35:28high speed networking, the kind used in supercomputers where
- 35:34every microsecond matters. With that, NVIDIA could offer
- 35:39not only the engines that compute, but the pathways that
- 35:44move data connecting racks and nodes into one coordinated
- 35:49Organism. Jensen Huang often described a
- 35:54data center as a single giant computer spread across many
- 36:01machines, and it's a calming image, like a city seen from
- 36:07above, lit by many small synchronized lights.
- 36:12This was also the period when NVIDIA explored becoming more
- 36:16than a GPU company, imagining a future where it could shape both
- 36:22the graphics side and the broader processor landscape.
- 36:27In 2000, Twenties NVIDIA announced a plan to acquire ARM,
- 36:34whose CPU designs live inside countless phones and devices,
- 36:39and whose architecture was spreading towards servers.
- 36:43The attempt did not succeed, held back by regulatory concerns
- 36:49and the delicate politics of foundational technology.
- 36:54But even the attempt revealed the direction of the company's
- 36:57mind. Whether through ownership or
- 37:00partnership, NVIDIA wanted a world where GPU's and CPU's work
- 37:07together more closely, each optimized for its role, each
- 37:12helping the other. So they kept building new kinds
- 37:17of server CPU, used new interconnects, new ways to
- 37:21stitch computing into a seamless fabric.
- 37:25It's a story that feels, at this point, less like a product line
- 37:32and more like an ecosystem, a quiet landscape of tools
- 37:38designed to make complex work feel natural.
- 37:43And somewhere in that landscape, the original dream is still
- 37:48present, not just to make images beautiful, but to make
- 37:53computation itself smoother, wider, and more alive.
- 37:59They even began speaking about new categories like the DPU, a
- 38:04data processing unit meant to handle the busy work of
- 38:09networking and security so CPU's and GPU's could focus on their
- 38:16main tasks. In a large cloud, that division
- 38:20of Labor matters, because every saved cycle becomes space for
- 38:28another training run, another simulation, another quiet
- 38:33experiment. So the architecture of modern
- 38:36computing began to look like a well run port at night.
- 38:42Different cranes, different lanes, each moving its share so
- 38:48the whole system flows without strain.
- 38:51By the time the 20 twenties arrived, the GPU had become
- 38:56something you rarely saw directly, like electricity
- 39:01behind a wall, present, necessary and mostly unnoticed
- 39:08until something remarkable happens.
- 39:11That remarkable thing for many people was the sudden feeling
- 39:16that a computer could speak not in stiff commands, but in
- 39:21flowing sentences, answering questions, writing drafts,
- 39:26composing small pieces of code. Generative AI is the name often
- 39:32used for this, but the experience is simpler than the
- 39:36label. You type a few words and the
- 39:40machine responds with something that seems to understand the
- 39:44shape of language. Behind that softness is an
- 39:49immense amount of training, a long process where a model reads
- 39:54and predicts and adjusts again and again, learning patterns the
- 40:00way a musician learns scales. This training is repetitive math
- 40:07on a vast scale, and it thrives on parallel effort, which is why
- 40:13GPU's became the engines of the moment, packed into data centers
- 40:20like rows of quiet lanterns in 2016.
- 40:25Jensen Huang even carried one of Nvidia's early AI systems to
- 40:31open AI by hand, a symbolic gesture that felt almost old
- 40:38fashioned, a founder delivering a machine to a small team as if
- 40:45dropping off tools for a workshop.
- 40:48Years later, those tools had become entire fleets, and the
- 40:55fleets had become part of a new global infrastructure, built so
- 40:59models could learn faster and be updated more often.
- 41:04If you think of the Internet as a library, then these training
- 41:08clusters are the late night readers, turning pages at
- 41:14extraordinary speed, not for pleasure but for pattern.
- 41:20The cultural ripple was immediate.
- 41:24Images generated from prompts, voices that could be
- 41:29synthesized, assistance that could help with daily writing,
- 41:35and a renewed debate about what creativity means.
- 41:41Yet the hardware story remains steady, fans spinning, heat
- 41:46rising, racks glowing softly, and the GPU doing what it has
- 41:52always done best, many simple steps repeated in parallel until
- 41:59complexity feels smooth. And as we move forward, we can
- 42:04keep one gentle thought in mind that the most surprising
- 42:09revolutions are often powered by the least glamorous things, by
- 42:14infrastructure that simply keeps working all night long.
- 42:20For many, the first encounter had a name like ChatGPT, a small
- 42:28window on a screen that made the future feel oddly familiar, like
- 42:33sending a message to a thoughtful stranger.
- 42:36And almost overnight, more people learned the word GPU not
- 42:43from gaming forums but from headlines and conversations at
- 42:49work, as if the hidden engine had finally become part of the
- 42:53public vocabulary. Still, the GPU itself did not
- 42:59change its personality. It remained a patient
- 43:03calculator, turning data into learned behavior with the same
- 43:09calm repetition it once used to turn geometry into light.
- 43:14When a new technology becomes widely desired, the world begins
- 43:19to measure it in physical terms. How many chips?
- 43:24How many servers? How many megawatts of power?
- 43:29How much cooling? How much space?
- 43:33AI training is not just software, it is a landscape of
- 43:39machines, and that landscape has temperature, weight, wiring and
- 43:45timing. Nvidia's newer data center GPU's
- 43:49arrived with names that sounded like quiet tributes to
- 43:53scientists, and each generation brought more parallel capacity,
- 44:00more efficiency, more ability to move data through the chip
- 44:05without bottlenecks. But the real change was not only
- 44:10inside the silicon, it was in how the systems were built.
- 44:16GPU's linked together with fast connections, treated as one
- 44:22shared pool of compute. This is why networking mattered
- 44:26so much, and why NVIDIA cared about the whole stack.
- 44:31The GPU, the interconnect, the software libraries, the tools
- 44:37that schedule work, the way memory is managed.
- 44:42It is also why cloud companies invested heavily, turning AI
- 44:48compute into a service, renting out acceleration the way a city
- 44:55rents out office space. If you imagine a data center at
- 44:59night, it can feel surprisingly peaceful.
- 45:03A warehouse of quiet order, blinking lights, controlled air,
- 45:09steady redundancy inside. Training jobs run for days,
- 45:15sometimes weeks, and the progress is incremental.
- 45:20A little better accuracy, a little more fluency, a little
- 45:25more stability. This kind of progress rewards
- 45:30patience, and patience has always been part of Nvidia's
- 45:34character, shaped early by the knowledge that one failed chip
- 45:39can end a company. So the firm kept refining the
- 45:44same fundamental lesson. Parallel computing is a way to
- 45:49make the heavy feel light, as long as you respect the details.
- 45:55And as the demand rose, the world began to notice another
- 45:59truth that computation is a physical resource, and that
- 46:05every new capability must eventually be paid for in power,
- 46:10heat, and careful engineering. Engineers began talking about
- 46:16packaging and memory as much as raw compute, because feeding the
- 46:21GPU is as important as the GPU itself, like making sure a fast
- 46:29train has enough track. High bandwidth memory stacked
- 46:34close to the processor helped keep the data flowing, and new
- 46:39systems used liquid cooling or carefully designed airflow to
- 46:45carry heat away quietly. In some places, the constraint
- 46:50became electricity, and companies planned new
- 46:55substations and new power contracts, as if AI were not
- 47:02only a software wave but a new kind of industry.
- 47:07And yet the mood inside those rooms remained calm.
- 47:12The hum of fans, the soft blink of status lights, and the sense
- 47:18that thinking in our modern world has become something we
- 47:23build. It's easy now to look at NVIDIA
- 47:28and see inevitability, to imagine a straight line from the
- 47:33diner breakfast to the modern data center.
- 47:36But the line was never straight, and the calm truth is that the
- 47:41company survived by repeatedly choosing the uncomfortable
- 47:46future over the comfortable present.
- 47:50Jensen Huang liked to talk about investing in zero billion dollar
- 47:56markets, meaning places where the opportunity is real but not
- 48:02yet visible, like a shoreline before sunrise.
- 48:07In the 1990s, 3D gaming felt like that, Exciting but small
- 48:14and easy to dismiss. In the mid 2000s, GPU computing
- 48:20for science felt like that, useful but niche and hard to
- 48:25monetize quickly. In the early 2000, Tens, deep
- 48:30learning felt like that, a research curiosity that suddenly
- 48:36became practical only when data and GPUs made it practical.
- 48:43To keep making those bets, NVIDIA had to build a culture
- 48:47that could live with long timelines.
- 48:50And that is not a glamorous skill.
- 48:53It means spending years on software ecosystems that do not
- 48:59show up in product photos, compilers, libraries,
- 49:05documentation, developer support, the slow work of making
- 49:11tools trustworthy. It means accepting that a
- 49:16platform is not a single invention, but a relationship,
- 49:21one that deepens as more people build on it and as the company
- 49:26responds. If you have ever learned a
- 49:29craft, you know the feeling. Progress doesn't happen as a
- 49:35single leap. It happens as 1000 small
- 49:39corrections, each one barely visible.
- 49:43Nvidia's rise is built on those corrections, the way a graphics
- 49:49company learned to serve scientists, then learned to
- 49:53serve machine learning, then learn to serve the world's
- 49:57largest clouds. And through that, the company's
- 50:01original question remained intact.
- 50:05How do we make computation feel fluid so creativity and
- 50:10discovery can move without friction?
- 50:14Tonight, we can let that question settle like a warm
- 50:19stone in the hand. Simple, steady and powerful in
- 50:26its persistence. You can hear the lesson most
- 50:29clearly in the crisis years, when early products failed, cash
- 50:36grew tight and the team shrank not from lack of talent but from
- 50:43the reality that hardware is expensive to get wrong.
- 50:49In that period, they learned to test a chip before it existed,
- 50:54running long emulations and listening closely for mistakes
- 51:00because they could not afford a second miss.
- 51:05When the recovery chip worked and sold, it did more than save
- 51:10the company. It taught them that speed is
- 51:14useful only when paired with rigor, and that survival is
- 51:18often a matter of details done calmly, correctly, and on time.
- 51:25That memory stayed with them, and it shaped how they
- 51:29approached every later pivot, with patience, with discipline,
- 51:35and with a willingness to rebuild their future from the
- 51:39inside out. Sometimes it helps to remember
- 51:45that all of this began with light.
- 51:49A game world is mostly light and timing.
- 51:54The illusion that a surface is curved, the suggestion that a
- 52:00hallway recedes, the gentle persistence of motion from frame
- 52:05to frame. In the Doom and Quake era, those
- 52:10worlds were assembled with clever shortcuts and you could
- 52:14almost see the seams. But the excitement was real
- 52:19because your imagination filled in what the hardware could not.
- 52:25Then GPUs arrived and the seams softened, textures became
- 52:32richer, geometry more complex, and the mind stopped noticing
- 52:38the tricks. Later, with programmable
- 52:42shaders, developers could paint with equations telling the GPU
- 52:48how water should shimmer, how smoke should fade, how skin
- 52:52should scatter light softly. Eventually, ray tracing brought
- 52:58the old dream closer, simulating the path of light itself,
- 53:04letting reflections and shadows behave more like they do in real
- 53:09rooms. It is easy to think of this as
- 53:13graphics, yet it is also a lesson about simulation, about
- 53:19using computation to imitate the physical world.
- 53:23And simulation in a calm way is a cousin of learning, because
- 53:30both require a great many small calculations to produce
- 53:35something that feels coherent. When AI models learn patterns,
- 53:41they are doing their own kind of rendering, not of light, but of
- 53:46language, of images, of the statistical shape of the world.
- 53:52The GPU's gift is the same in both cases.
- 53:57It handles repetition without complaint.
- 54:01It spreads work across many cores.
- 54:05It makes the heavy field smooth. So the cultural story and the
- 54:11technical story keep touching A teenager playing Unreal
- 54:17Tournament in a dim bedroom and a researcher training an image
- 54:23model in a bright lab, both relying on the same kind of
- 54:28parallel engine. Tonight, you don't need to
- 54:31separate these threads. You can let them weave together
- 54:37and feel how a technology can grow by serving play first, then
- 54:43serving discovery, and eventually serving almost
- 54:48everything. Even the sounds are related.
- 54:51The click of a mouse, the low whirl of a fan, the faint heat
- 54:58rising from a PC case after a long session.
- 55:02Years later, the same kind of hum fills a data center, only
- 55:07steadier, more distant, as if the world's computing has moved
- 55:13from bedrooms into warehouses. And yet the feeling is familiar,
- 55:20a machine working beside you, not demanding attention, just
- 55:26providing the smoothness that lets your mind stay in the
- 55:30story. This is why Nvidia's rise can be
- 55:35told as a bedtime story at all, because it is, at heart about
- 55:43making experiences feel calm and continuous, even when the work
- 55:49underneath is vast, The world slows, and the frames keep
- 55:56coming. A chip is a piece of silicon,
- 56:01but a platform is a habit. CUDA became that habit slowly,
- 56:06as more developers learned it, as more libraries were built on
- 56:11top of it, and as more research results depended on its rhythms.
- 56:17Overtime, Nvidia's advantage was not only that its GPU's were
- 56:23fast, but that the path to use them was well lit.
- 56:28Documentation tools optimized code for common tasks.
- 56:35In machine learning, people began to rely on libraries that
- 56:40handled the hardest parts. The deep inner loops of
- 56:45training, the careful math for neural networks, the patterns
- 56:51that appear in almost every model.
- 56:54When those libraries are tuned for a specific platform, the
- 56:59platform becomes sticky not through coercion, but through
- 57:05convenience and trust. Trust matters in computing
- 57:11because a researcher does not want to spend months debugging
- 57:15hardware quirks when the real goal is discovery.
- 57:21A company building an AI product does not want to reinvent the
- 57:26foundations, it wants to stand on something stable and keep
- 57:32moving. So NVIDIA kept doing what it had
- 57:36learned from gaming Cultivate developers, support the people
- 57:42building content, because content is what makes hardware
- 57:48meaningful. In games, the content is worlds
- 57:52and stories and competition. In AI, the content is models and
- 58:00data and applications. Speech recognition, vision
- 58:04systems, recommendations, scientific tools, creative
- 58:11assistance. The platform connects the mall
- 58:15like a slow river that carries many boats, and the river's
- 58:20value increases as more people choose to float on it.
- 58:25Of course, the river is not alone.
- 58:30Competitors exist, alternative approaches exist, and the
- 58:36industry is always searching for efficiency.
- 58:40But Nvidia's strategy was steady Build the ecosystem, build the
- 58:45hardware, and keep them aligned so the experience stays smooth.
- 58:51If you listen closely, you can hear how this echoes the
- 58:56earliest days. The same desire to reduce
- 59:01friction, to make complex work feel simple, to keep the images
- 59:07or the learning flowing. In supercomputing centers, GPU's
- 59:13became standard equipment, and in cloud data centers they
- 59:18became rentable resources. So a student with a small budget
- 59:24could access the same kind of acceleration as a large lab.
- 59:30That accessibility changed the pace of innovation.
- 59:34More experiments, more iterations, more chances for
- 59:38unexpected breakthroughs. And each time someone learned
- 59:43CUDA, taught it, used it, or built on it, the ecosystem
- 59:48deepened. The way a city becomes more
- 59:52livable as more roads, signs and shared customs appear.
- 59:58It's a gentle kind of compounding.
- 1:00:01Less dramatic than a headline, but more durable.
- 1:00:05The accumulation of tools, habits, and trust all circling a
- 1:00:11parallel engine. So when the world suddenly
- 1:00:14wanted more AI, NVIDIA was already there.
- 1:00:19Not just with chips, but with a well worn path that made those
- 1:00:23chips usable as NVIDIA grew into AI.
- 1:00:28It also began to imagine where AI would live, not only in data
- 1:00:34centers, but in machines that move through the physical world.
- 1:00:39A self driving car is, in a quiet sense, a perception
- 1:00:45machine. Cameras and sensors streaming
- 1:00:49data, software deciding what matters.
- 1:00:53Computation, turning patterns into action.
- 1:00:56This requires fast inference, the moment to moment use of a
- 1:01:01trained model, and it requires reliability because the road
- 1:01:08does not pause while the computer thinks so.
- 1:01:12NVIDIA built platforms for automotive computing, pairing
- 1:01:17specialized hardware with software that could process
- 1:01:21sensor data and run neural networks in real time.
- 1:01:26The goal was not to sell a single chip, but to offer a
- 1:01:31foundation that car makers and robotics teams could build on,
- 1:01:36adapting it to their needs. At the smaller end, NVIDIA
- 1:01:41created compact modules for Edge AI, tiny computers with GPU
- 1:01:48power used in drones, cameras, robots, and industrial devices.
- 1:01:55These systems are like the data center's distant cousins, still
- 1:02:00parallel, still repetitive, but living closer to the world of
- 1:02:06motion and dust and vibration. And beyond physical machines,
- 1:02:12NVIDIA leaned into simulation, the idea that you can build a
- 1:02:18digital environment, a kind of twin, and test systems there
- 1:02:23safely. Simulation has always been part
- 1:02:26of Nvidia's heritage, because graphics is a form of
- 1:02:32simulation, and the GPU is a natural tool for building
- 1:02:37believable virtual spaces in robotics.
- 1:02:42Simulation is a gentle teacher, a place where a robot can fail
- 1:02:49without harm, learn without fear, and repeat an action
- 1:02:54thousands of times without getting tired.
- 1:02:59In industry, simulation helps engineers model factories,
- 1:03:04logistics, traffic flow, and energy systems, letting them
- 1:03:11explore possibilities before making expensive changes.
- 1:03:16So the company's story broadened again, from drawing pixels to
- 1:03:22accelerating science, to supporting learning, and now to
- 1:03:27helping machines perceive and act in the physical world.
- 1:03:33It is all 1 theme really, giving computation a body and giving
- 1:03:40complex systems a calm place to practice before they meet
- 1:03:46reality. NVIDIA called some of these
- 1:03:50simulation efforts Omniverse, a shared 3D workspace where many
- 1:03:56tools can meet and where scenes can be rendered, changed and
- 1:04:03replayed with quiet precision. For robotics, they built
- 1:04:08software stacks that helped teams train and deploy systems,
- 1:04:14often blending computer vision, planning and control into one
- 1:04:19coherent flow. If you picture it, it feels like
- 1:04:23a calm rehearsal room. Virtual light, virtual gravity,
- 1:04:29virtual friction, all computed patiently so the real machine
- 1:04:35can move with confidence later. And so even here, the GPU's role
- 1:04:41remains consistent. It makes rich simulation and
- 1:04:46fast learning possible, turning repetition into readiness.
- 1:04:52While AI drew headlines, another, quieter thread
- 1:04:56continued, one that NVIDIA had always served the work of
- 1:05:01creators who build images, spaces and stories.
- 1:05:06Film studios render scenes that never existed.
- 1:05:10Architects preview buildings before they are built.
- 1:05:14Designers rotate objects in 3D until the shape feels right.
- 1:05:20These tasks are heavy with geometry and light, and they
- 1:05:24benefit from the same parallel computation that powers games.
- 1:05:31So Nvidia's professional visualization tools kept
- 1:05:36evolving, supporting artists who care about subtle reflections,
- 1:05:41soft shadows, the way a surface looks under morning light.
- 1:05:46In some ways, this creative thread is a gentle bridge
- 1:05:52between eras, because it reminds us that the GPU's earliest
- 1:05:59purpose was to help imagination appear on a screen.
- 1:06:05What changed is scale and audience, and the way images now
- 1:06:10travel. A render created in one city can
- 1:06:15be shared instantly, edited collaboratively, and experienced
- 1:06:21on a phone across the world. The boundary between real and
- 1:06:27rendered grew softer, and not in a frightening way, but in a
- 1:06:31practical 1 because simulation helps us plan, rehearse, and
- 1:06:38communicate. Even a simple social media
- 1:06:41filter, smoothing a face or adjusting a background is part
- 1:06:46of this lineage. A tiny piece of accelerated
- 1:06:50computation applied to everyday life, and when AI began
- 1:06:56generating images from text prompts, it did so using learned
- 1:07:01patterns that still rely on GPU accelerated math, echoing the
- 1:07:08old relationship between computation and art.
- 1:07:13You can see the continuity from a gamer watching a new lighting
- 1:07:18effect in a 2004 title to an artist using a neural tool to
- 1:07:25explore variations in 2024, both seeking the same thing, not
- 1:07:32novelty for its own sake, but a smoother path from idea to
- 1:07:38expression. In the sleep wise pace of this
- 1:07:42night, it's comforting to notice how often technology's goal is
- 1:07:47simply to remove friction so the mind can stay with the creative
- 1:07:52act. The GPU, steady and patient, has
- 1:07:57been doing that for decades, turning mathematics into images
- 1:08:03and now turning images and language into new kinds of
- 1:08:08shared possibility. Real time rendering also changed
- 1:08:13how film and media are made, with directors walking through
- 1:08:18virtual sets before cameras roll, seeing lighting and
- 1:08:23composition in the moment instead of waiting days for
- 1:08:27final renders. That shift, too, is a kind of
- 1:08:31calm fewer surprises, more iteration, more room for gentle
- 1:08:38adjustment. And as the tools became faster,
- 1:08:42the creative process became more like conversation.
- 1:08:46Try aversion. See it instantly refine, repeat
- 1:08:52until the result feels inevitable.
- 1:08:55In that way, Nvidia's rise is not only about machines
- 1:09:00learning, but about humans creating with less delay, as if
- 1:09:07the distance between thought and screen has quietly shortened.
- 1:09:12Every large story has its limits, and for modern computing
- 1:09:18the limits are often physical energy, materials, manufacturing
- 1:09:25capacity, and the quiet complexity of global supply
- 1:09:31chains. A high end processor is not just
- 1:09:35design, it is fabrication, packaging, shipping, testing,
- 1:09:42and the careful assembly of components that arrive from many
- 1:09:47places. In the AI boom, those chains
- 1:09:51became visible because demand rose quickly and the world
- 1:09:57learned that not all chips can be produced at will.
- 1:10:02There were also rules and restrictions shifting over time,
- 1:10:07influencing where certain advanced technologies could be
- 1:10:11sold, reminding everyone that computing lives inside a wider
- 1:10:18world of policies and priorities.
- 1:10:22NVIDIA navigated this by doing what it has always done,
- 1:10:29focusing on engineering, on efficiency, on making each
- 1:10:34generation of hardware do more work per unit of energy.
- 1:10:40Efficiency is a soothing concept when you think about it.
- 1:10:46The idea that you can accomplish more with less strain, that
- 1:10:51progress does not have to mean waste.
- 1:10:55In data centers, that efficiency shows up as better cooling,
- 1:11:00smarter scheduling, and architectures that avoid
- 1:11:05unnecessary movement of data. Because moving data can be as
- 1:11:12costly as computing. In chips, it shows up as
- 1:11:17specialized units for common operations, and in software, it
- 1:11:23shows up as libraries that squeeze more usefulness out of
- 1:11:28each computation. All of this is a kind of
- 1:11:32invisible stewardship, the quiet effort to make the system
- 1:11:37sustainable. As it scales and as AI becomes
- 1:11:42more common, the world will keep asking the same calm question in
- 1:11:47new forms. How do we build intelligence
- 1:11:51without building too much friction?
- 1:11:54Nvidia's answer again and again has been acceleration paired
- 1:12:00with integration, making the processor faster, yes, but also
- 1:12:07making the whole pathway smoother.
- 1:12:10Tonight you don't need to hold the details, only the feeling
- 1:12:15that behind the scenes, an immense amount of careful
- 1:12:20engineering is trying to keep the future gentle.
- 1:12:25If you picture a silicon wafer, it's like a dark mirror, round
- 1:12:30and delicate, carrying many future machines in a single
- 1:12:35sheet. Lithography prints patterns too
- 1:12:38small to see layer by layer, and each layer must align as if the
- 1:12:45chip is being written in a microscopic calligraphy.
- 1:12:50Then comes packaging, where memory and processor are brought
- 1:12:55close together, and where the physical distance between
- 1:12:59components becomes part of performance.
- 1:13:03When people say compute is the new oil, they often forget how
- 1:13:09tangible it is. Metal, water, power lines and
- 1:13:14the calm logistics of keeping everything running.
- 1:13:18So the rise of AI has quietly renewed an old respect for
- 1:13:23infrastructure, and NVIDIA, a company born in graphics, has
- 1:13:29become one of the architects of that infrastructure.
- 1:13:33No company rises alone, and Nvidia's story has always
- 1:13:38included rivals, alternatives and the steady pressure that
- 1:13:44keeps a craft honest. In gaming, there were
- 1:13:47competitors making their own graphics cards, pushing price
- 1:13:53and performance in a cycle that benefited players in data
- 1:13:58centers. Cloud companies explored custom
- 1:14:02chips, building specialized processors for their own needs,
- 1:14:08and researchers experimented with many kinds of accelerators.
- 1:14:13This is natural because when something becomes foundational,
- 1:14:20everyone wants to shape it. Yet Nvidia's endurance has often
- 1:14:26come from its platform approach, not only shipping hardware, but
- 1:14:32supporting it with software that makes the hardware practical.
- 1:14:37A custom chip can be powerful, but power without a mature
- 1:14:42ecosystem can feel lonely, like a road without signs.
- 1:14:49So NVIDIA continued to invest in the tools that let people build
- 1:14:55quickly Compilers, libraries, drivers, and a wide set of
- 1:15:01partnerships across academia and industry.
- 1:15:06Competition also pushed NVIDIA to stay efficient, to keep
- 1:15:11improving performance per Watt, to design systems that can scale
- 1:15:16without losing reliability. And in the background, a quieter
- 1:15:21competition existed as well. The competition against time,
- 1:15:27against complexity, against the tendency of systems to become
- 1:15:32fragile as they grow. It's surprisingly hard to keep
- 1:15:36an ecosystem stable while it expands, and stability is part
- 1:15:43of what people pay for, even if they don't always name it.
- 1:15:48So the rise of NVIDIA is not a story of a single victory, but
- 1:15:54of continuous tuning, like keeping an instrument in tune as
- 1:16:00the temperature changes tonight, it's calming to remember that
- 1:16:06progress in technology is usually not a conquest, but a
- 1:16:13conversation. Many ideas, many approaches, and
- 1:16:20a slow selection of what proves dependable.
- 1:16:25NVIDIA has been one of the dependable answers for a long
- 1:16:30time, partly because it learned early that survival requires
- 1:16:36both speed and care. When engineers choose tools,
- 1:16:42they often choose the one that reduces uncertainty, the one
- 1:16:48that works across operating systems, across clusters, across
- 1:16:53generations of hardware. They choose the one with a
- 1:16:58community, with examples, with people who have already solved
- 1:17:03the sharp corners. That is why, even as
- 1:17:07alternatives grow stronger, many teams still return to Nvidia's
- 1:17:13stack. Because it feels like a well
- 1:17:17maintained path through dense forest.
- 1:17:20And as those alternatives improve, the whole industry
- 1:17:24improves. Better standards, better ideas,
- 1:17:28better efficiency, all feeding back into a healthier ecosystem.
- 1:17:35In that way, competition becomes another form of calm, a steady
- 1:17:41wind that keeps the air moving, preventing stagnation,
- 1:17:45encouraging everyone to keep refining their craft.
- 1:17:49If you step back, Nvidia's rise can feel like a shift from the
- 1:17:56visible to the invisible. At first, you could see the
- 1:18:01change on your own screen. Smoother motion, Richer colors.
- 1:18:06A game world that suddenly felt more alive.
- 1:18:12Then the change moved into places you rarely visit.
- 1:18:16Server rooms, research labs, data centers at the edge of
- 1:18:21towns, humming behind locked doors.
- 1:18:26And now the results return to you through everyday
- 1:18:30experiences. Better translations.
- 1:18:34Smarter search recommendations, Image tools, voice assistance,
- 1:18:42and the quiet sense that software is becoming more
- 1:18:47conversational. It is the loop of modern
- 1:18:50infrastructure. Build it far away and it comes
- 1:18:55back as something intimate. There is also a human loop, a
- 1:19:01loop of patience, founders taking risks, engineers
- 1:19:05iterating, developers adopting, researchers discovering
- 1:19:10companies building on top. The story rewards the same
- 1:19:16qualities again and again, endurance, clarity and the
- 1:19:22willingness to invest in tools that pay off slowly.
- 1:19:27It is tempting to focus on the famous moments.
- 1:19:31The first time the word GPU was said, The release of CUDA, the
- 1:19:38Imagenet breakthrough, the arrival of generative AI.
- 1:19:43But the sleep wise truth is softer.
- 1:19:47Those moments mattered because thousands of quieter decisions
- 1:19:53made them possible. A driver update that improves
- 1:19:57stability. A library that speeds up a
- 1:20:01common operation. An engineering practice that
- 1:20:05prevents A costly error. A support team that answers a
- 1:20:10question at the right time. This is the texture of real
- 1:20:15progress, Not fireworks, but steady refinement.
- 1:20:21And perhaps that is why NVIDIA fits so well into a bedtime
- 1:20:27story. Because it is not a tale of
- 1:20:30sudden magic, but of compounding craft.
- 1:20:35A company that began by making pixels move smoothly spent
- 1:20:41decades learning how to make computation itself move
- 1:20:46smoothly, and in doing so, it helped the world build new kinds
- 1:20:51of thinking. Tonight we can let that idea
- 1:20:56settle gently, that behind the modern glow of screens and
- 1:21:01services there are patient engines and patient people
- 1:21:07repeating small steps until the future feels natural.
- 1:21:13Some of those engines sit in cold climates, where air is
- 1:21:17cheap and cooling is easy, and some sit in warm places where
- 1:21:22cooling systems work quietly all day.
- 1:21:27They sit near fiber lines and power substations, near highways
- 1:21:32that deliver replacement parts, near the unnoticed logistics
- 1:21:38that keep everything dependable. And while we sleep, they run
- 1:21:44training, serving, calculating, making the modern world feel
- 1:21:49responsive, as if intelligence is always on call.
- 1:21:55It is a strange comfort, when you think about it, that so much
- 1:22:00activity can exist without demanding our attention.
- 1:22:05A hum we don't have to hear, A light we don't have to watch.
- 1:22:11So tonight, as the room darkens, we can carry Nvidia's story like
- 1:22:17a small Lantern. Steady, not urgent.
- 1:22:22A company begins with a question, survives its early
- 1:22:26missteps, learns to build with rigor, and then discovers that
- 1:22:33its tool can serve more than the first dream.
- 1:22:36First it helps the world see smoother games, richer images,
- 1:22:42light rendered with patients. Then it helps the world measure
- 1:22:48and simulate molecules, weather, medical images, the slow
- 1:22:54problems that require repetition.
- 1:22:57Then it helps the world learn networks, adjusting their
- 1:23:02weights, patterns emerging from data machines, becoming better
- 1:23:09at recognizing what we already know how to recognize.
- 1:23:15And then, quietly, it helps the world speak language models that
- 1:23:22respond, image models that imagine systems that feel like
- 1:23:28assistance rather than calculators.
- 1:23:31Underneath every stage is the same rhythm, parallel work, many
- 1:23:37small calculations done together, so complexity can
- 1:23:42arrive without strain. You can picture it now as a
- 1:23:47gentle current, electrons moving through etched pathways, heat
- 1:23:53carried away by airflow, status lights blinking softly in a
- 1:23:58distant room. And you can picture the older
- 1:24:02version, too. A bedroom PC in 1999, a bulky
- 1:24:07monitor, a new graphics card. A game loading, a world
- 1:24:13appearing frame by frame. If there is a lesson worth
- 1:24:17keeping, it is that technology rarely arrives in a single leap.
- 1:24:23It arrives through patience, iteration and years of careful
- 1:24:29building. It moves through people willing
- 1:24:32to work on foundations, to make tools reliable, to choose the
- 1:24:37long path when the short 1 looks tempting.
- 1:24:41Now the world slows, and you do not need to hold any of it
- 1:24:46tightly. Let the images fade, let the
- 1:24:50servers hum somewhere far away, and let parallel work become a
- 1:24:56lullaby, small efforts working together so the whole system can
- 1:25:03rest somewhere. A founder's early breakfast
- 1:25:07conversation becomes a decade of prototypes, then a platform,
- 1:25:13then an ecosystem that thousands of engineers quietly depend on.
- 1:25:19Somewhere a creator renders a scene in real time, adjusting
- 1:25:25light until it feels like dusk, and the GPU turns math into
- 1:25:31atmosphere without complaint. And somewhere a curious person
- 1:25:37opens a small window on a screen, asks a question, and
- 1:25:42receives an answer that feels calm, as if the machine has
- 1:25:47learned the contours of language.
- 1:25:50All of it is connected not by spectacle, but by steady
- 1:25:55repetition, the kind that makes hard things seem simple and
- 1:26:00makes simple things feel smooth. So as you settle deeper into the
- 1:26:07night, you can let the story drift away from names and dates
- 1:26:13and toward a softer understanding.
- 1:26:17Progress is often a quiet compounding of small choices.
- 1:26:22The hum continues in the distance like wind through
- 1:26:27trees, like water through pipes, like a city breathing while its
- 1:26:33streets sleep. And here in your own room, you
- 1:26:38can do the same. Breathe, soften, and let the day
- 1:26:43dissolve. Good night.