Building the Thinking Machine: Lessons from Jensen Huang and Nvidia’s Quiet Power

From CUDA to ChatGPT, what Nvidia’s story reveals about conviction, culture, and playing the long game in tech.

May 17, 2025

On February 21, 2024, Nvidia reported 30 billion in profit, and gross margins exceeding 70%. Its market cap rose by $277 billion in a single day—the largest single-day gain in Wall Street history.

But behind the financial headlines is a story of something rarer than scale: strategic patience. This isn’t a tale of hype cycles or vision-board bombast. It’s the story of Jensen Huang, a quiet operator who spent decades building the tools that would make artificial intelligence possible—long before the world cared.

Stephen Witt’s The Thinking Machine tells this story with narrative precision. It’s part biography, part business chronicle, part systems-level history of the AI age. Huang, who grew up as an isolated Taiwanese immigrant in rural Kentucky, didn’t start by trying to disrupt the world. He started by understanding the architecture of the present—and quietly designing the infrastructure for the future.

“Jensen didn’t try to predict the future,” Witt writes.
“He looked carefully at the capabilities of the circuits sitting in front of him… and only then allowed himself a step into the nebulous realm of vibes.”

For builders, this book isn’t just about Nvidia’s success. It’s about how to think clearly and build patiently in a world increasingly obsessed with speed and noise.

What this review covers

Rather than a chronological summary, this review highlights seven strategic ideas and one leadership profile that every builder should study. Each one draws from a pivotal decision, pattern, or philosophy embedded in Nvidia’s rise:

Why long-term platform bets (like CUDA) require conviction before validation
How power users at the fringe often signal future markets
What it means to build for resonance, not visibility
How to turn infrastructure into an unavoidable dependency
Why a founder’s habits inevitably shape the organization’s operating system
How innovation requires changing paradigms—not just accelerating within them
Why physical scaling (compute, energy) brings strategic and ethical consequences
And how Jensen Huang leads—not by volume or vision decks, but through presence, precision, and deep technical engagement

Each section unpacks a story or turning point from The Thinking Machine and distills the lesson for founders, product leaders, and anyone building under long-term uncertainty.

Let’s begin with the most radical and initially misunderstood move Nvidia ever made: the launch of CUDA in 2006, when no one—not investors, not developers—believed it was worth building.

Commit Early to Your Long Bet

In 2006, Nvidia made a bold—and seemingly irrational—move. Jensen Huang introduced CUDA, a software platform that let developers access the raw parallel-processing power of Nvidia’s GPUs. It wasn’t a minor product update. CUDA required developers to learn new mental models and rethink how computation worked.

Almost no one cared.

Wall Street analysts dismissed it. Developers ignored it. Even some Nvidia engineers weren’t sure why they were building it. “Not only did Wall Street not think CUDA was valuable,” one Nvidia employee recalled, “they thought it had negative value.”

But Huang wasn’t trying to respond to current demand—he was designing for a future he believed was inevitable. CUDA wasn’t for rendering prettier video games. It was a bet that one day, parallel compute would unlock new classes of problems: physics simulations, genomics, and maybe something more speculative—artificial intelligence.

That future arrived in 2012.

A relatively unknown graduate student, Alex Krizhevsky, used CUDA and a couple of retail Nvidia GPUs to train AlexNet, the neural network that would dominate the ImageNet competition and launch the deep learning revolution. It was a breakthrough moment not just for AI—but for CUDA itself. The market that didn’t exist six years earlier had suddenly become the most important thing in computing.

From that point forward, every AI company, academic lab, and hyperscaler became dependent on Nvidia’s stack. CUDA wasn’t just useful—it became unavoidable.

Lesson for builders:
If you're building for a paradigm shift, don’t wait for validation. A true platform often looks like a waste of time—right up until it powers the next revolution.

Think in Parallel—Literally and Strategically

In the early 2000s, most of the world still thought like Intel: faster chips meant better chips, and better chips meant higher single-thread performance. That logic defined the CPU era. But Jensen Huang saw something else. He bet that the future wouldn’t be about doing one thing faster—it would be about doing many things at once.

That conviction led Nvidia to embrace parallel processing at a time when almost no one outside of scientific computing cared. While Intel doubled down on clock speeds, Nvidia invested in massively parallel architectures, allowing GPUs to run thousands of operations simultaneously.

The idea wasn’t just technical. It was strategic.

At a time when CPUs were hitting physical and performance limits, GPUs opened up a new dimension. They weren’t just accelerators for rendering images. They were general-purpose engines for processing complex, non-linear workloads—from simulating proteins to training neural networks.

Early believers included developers like John Carmack, the legendary game designer behind Doom and Quake, who pushed the limits of real-time 3D rendering. He didn’t need theoretical arguments—he needed frame rates. Nvidia delivered.

Years later, that same architecture would prove critical for the rise of AI. Training large neural networks like AlexNet or GPT doesn’t just require speed. It requires computational parallelism at scale—exactly what Nvidia had been optimizing for all along.

“With CUDA, Nvidia had turned the GPU into a general-purpose computing platform,” Witt writes.
“What had once been a graphics chip was now a massively parallel supercomputer.”

Lesson for builders:
Sometimes the answer isn’t to move faster—it’s to move differently. Innovation often comes not from improving within the current paradigm, but from shifting to a new computational logic entirely.

Serve the Extremes First

Before Nvidia became the engine of the AI revolution, it found traction in an unexpected place: professional gaming.

In the early 2000s, the esports scene was just emerging, and players like Johnathan “Fatal1ty” Wendel were redefining competitive gaming. Victory didn’t come from strategy alone—it came from milliseconds. Higher frame rates meant faster reactions, and faster reactions meant survival.

One of Fatal1ty’s key weapons? Nvidia’s TNT2 graphics card, which gave him a crucial edge over opponents. It wasn’t built for esports. But in the hands of extreme users, it became a competitive differentiator.

This wasn’t a one-off. Again and again, Nvidia’s earliest breakthroughs came not from mass-market adoption, but from fringe users pushing the hardware to its limits—gamers, modders, graphics researchers, and eventually deep learning scientists.

Perhaps the most famous example: Alex Krizhevsky, a grad student at the University of Toronto who used consumer-grade Nvidia GPUs to train AlexNet under his desk. He didn’t have access to a supercomputer. He had CUDA and a couple of GeForce cards—and that was enough to redefine AI.

“Some graduate student would skip the grant circus, buy an Nvidia GPU with his housing stipend, and usher in a revolution,” Witt writes.

These extreme users weren’t Nvidia’s official target market—but they saw the product’s true potential before the rest of the world did.

And when the world finally caught on, Nvidia was already the default.

Lesson for builders:
The most visionary users aren’t always in the boardroom. They’re in the dorm room, the gaming den, the research lab. If you want to build the future, serve the fringe first—because the mainstream often follows right behind.

Cultivate Resonance, Not Hype

Jensen Huang doesn’t pitch Mars colonies. He doesn’t drop product demos on livestreams. He rarely speaks in slogans. And yet, he may be the most consequential founder of his generation.

Why? Because he listens more than he speaks.

Stephen Witt introduces a powerful idea in the book—Huang’s instinct for “resonance”. While other tech leaders chase attention and headlines, Huang tunes into something quieter: the edge of technical possibility, where developers and researchers begin whispering about a new breakthrough. He’s not trying to predict the future through grand vision. He’s trying to feel it through subtle signals.

This shows up most clearly in Nvidia’s early recognition of the transformer architecture. When models like GPT and BERT began gaining traction in the research community, Nvidia didn’t wait to be told what to do. Huang had his compiler team rebuild their entire software stack around transformers—not because customers asked, but because the future was already shifting underneath them.

“He could feel when something was about to change,” Witt writes.
“His gift was not storytelling, but timing. He listened until the signal became undeniable—and then he moved decisively.”

This pattern repeats throughout Nvidia’s history. CUDA wasn’t launched with fanfare; it was launched because Huang believed someone would need it. That someone turned out to be the entire AI industry.

Lesson for builders:
Don’t mistake hype for validation. And don’t assume silence means nothing is happening. The real inflection points often arrive as whispers, not announcements. Pay attention to what resonates quietly at the edges—that’s where the next wave begins.

Create Dependencies, Not Just Products

Most companies sell features. Some sell ecosystems. Nvidia sells inevitability.

By the time the AI boom arrived, startups and researchers didn’t choose Nvidia—they were already dependent on it. Not just because of speed or performance, but because CUDA had become the foundation of their codebases, workflows, and research infrastructure.

This didn’t happen by accident.

CUDA, introduced in 2006, was a painful product. It forced developers to think in parallel threads, write strange new syntax, and re-architect their computational logic. Adoption was slow. At times, even Nvidia employees weren’t sure it would last.

But Huang stuck with it.

“The performance gains from CUDA had to be so great, and so obvious, that customers would voluntarily build whole new academic disciplines around the platform,” Witt writes.

That’s exactly what happened.

Once AlexNet proved CUDA’s value in 2012, the platform became sticky—not just because it worked, but because it became too embedded to replace. New AI models, academic papers, internal tools, startup infrastructure—all of it began assuming CUDA as the baseline. Switching away was no longer just inconvenient—it was prohibitive.

This is how platform moats are truly built—not by locking users in, but by becoming too valuable to leave.

Lesson for builders:
Don’t just aim to delight your users. Aim to enable them so deeply that they reorganize their workflows around your product. That’s when you stop being a tool—and start becoming infrastructure.

Scale Your Mind Through the Company

As Nvidia’s valuation crossed $1 trillion, you might expect Jensen Huang to step back—delegate more, focus on high-level strategy, maybe appoint a successor.

Instead, he doubled down.

Even at the peak of Nvidia’s power, Huang maintained 55 direct reports. He still wakes at 4 a.m. to tackle problems personally. He still spends hours obsessing over chip designs, compiler behavior, and developer tooling. His fingerprints are on everything—not out of ego, but out of conviction.

Witt paints a picture of a leader who didn’t just build a product—he built a company in his own image. From Nvidia’s spaceship-like campus to its precise internal culture, the organization reflects Huang’s brain: focused, exacting, relentless.

“The company was his thinking machine,” Witt writes.
“It didn’t just run on GPUs—it ran on Jensen.”

That’s not always scalable. But it worked because Huang built systems that translated his priorities into action—through culture, hiring, technical rigor, and architectural alignment. Nvidia scaled not by diluting his influence, but by amplifying it.

Lesson for builders:
You’re not just scaling a product—you’re scaling your thinking. Your priorities, habits, and decision frameworks will become embedded in your team, your culture, your platform. Be intentional about what you transmit—because your company will scale what you model.

Industrialize Thought—But Own the Consequences

Nvidia no longer just sells chips—it industrializes cognition. Its GPUs power everything from protein folding and LLMs to image generators and autonomous vehicles. What began as a graphics company is now the foundation of AI infrastructure.

But with that power comes unease.

Witt details how even AI’s godfathers—Yoshua Bengio, Geoffrey Hinton, Ilya Sutskever—have become alarmed by the systems Nvidia enables. After seeing the pace of GPT-like models, Bengio confessed to feeling what he called “The Fear.”

Huang, for his part, denies Nvidia is building anything like a bomb. But the tension remains: his hardware makes near-infinite inference possible. And that raises questions not just about capability—but control.

“Oppenheimer built a bomb. We’re not doing that,” Huang told Witt. But the subtext is clear: this might be even bigger.

Lesson for builders:
If you’re building foundational technology, you can’t outsource the ethical conversation. You don’t need to solve everything—but you do need to own what your product makes possible, and engage with the consequences—before they’re decided for you.

Precision, Pressure, Presence: How Jensen Huang Leads

Jensen Huang isn’t a product evangelist. He’s not a charismatic keynote showman or a hands-off visionary. He’s something else: a founder-operator with a craftsman’s mindset and an engineer’s attention to edge cases—who happens to be running a trillion-dollar company.

He’s involved. Deeply.

Even as Nvidia became the world’s most valuable semiconductor firm, Huang retained an unusually flat reporting structure: 55 direct reports, no clear successor, and an operational rhythm that starts at 4 a.m. He remains hands-on not because he distrusts delegation, but because he knows exactly what matters—and refuses to compromise on it.

Inside Nvidia, he’s revered—but not because he’s loud or magnetic. His influence comes from presence. He shows up to hard meetings. He gives direct, highly technical feedback. He expects you to have done the thinking—because he definitely has.

“The company was his thinking machine,” Witt writes. “It didn’t just run on GPUs—it ran on Jensen.”

Huang cultivates pressure, but not toxicity. Colleagues describe him as demanding, clear, and completely committed. He brings the same intensity to a customer pitch as he does to a chip review meeting. And when the stakes are existential—as they were after the NV1 failure or Xbox collapse—he doesn’t outsource the solution. He gets in the room.

His personality is also reflected in Nvidia’s culture: technically rigorous, relentlessly forward-facing, intolerant of shortcuts, and surprisingly quiet. Internally, the company has little appetite for hype or vision slides. They talk latency, architecture, power efficiency, compiler flags. They build. And that ethos starts at the top.

Lesson for builders:
Your presence is your product. The values you model—what you tolerate, where you show up, what you obsess over—will quietly define your company’s culture. You don’t need to be loud to lead. But you do need to be precise.

Final Thoughts: Building in the Age of Thinking Machines

Nvidia didn’t arrive at the center of the AI revolution by accident.

It got there because Jensen Huang built quietly, patiently, and obsessively—for decades. While others chased trends, he committed to infrastructure. While others launched products, he cultivated platforms. While others gave TED talks, he redesigned compilers.

That’s what makes The Thinking Machine such a valuable read. It’s not a hype story. It’s a story about doing hard things for a long time, and trusting that if you’ve built the right foundation, the future will find you.

For tech builders, the lessons here aren’t abstract—they’re concrete, applicable, and necessary in an era where substance often hides beneath noise.

Key Takeaways for Builders

Bet early—if your conviction is clear. CUDA looked like a waste of time for years. Then it became the foundation of AI.
Don’t chase the center. Serve the edge. Nvidia’s first believers were gamers and grad students, not enterprise IT.
Great leaders don't predict the future—they listen for it. Huang followed resonance, not vision decks.
Build infrastructure, not just products. Moats form when people need what you’ve built to function.
Your company is a reflection of your cognitive style. Jensen’s precision, rigor, and technical depth shaped Nvidia’s culture as much as any strategy did.
If you industrialize intelligence, own the consequences. You don’t need to solve AI ethics alone—but you can’t sit it out.

The Thinking Machine isn’t just a biography. It’s a strategic lens on how to build lasting advantage in deep tech. Not by chasing attention. Not by moving fast and breaking things.

But by thinking deeply, building patiently—and playing the long game better than anyone else.

Careless People: A Cautionary Tale of Power, Greed, and Lost Idealism, by Sarah Wynn-Williams 2025

Xiaomi: Startup Thinking, by Lei Jun 2022

Gambling Man: The Secret Story of the World's Greatest Disruptor, Masayoshi Son, by Lionel Barber 2025

Read to Build

Discussion about this post