Yesterday we profiled AMD — the underdog that bit back. Today we profile the company that AMD’s GPU division has been trying to catch for two decades.

NVIDIA.

NVIDIA is the most valuable semiconductor company on Earth. As of late 2025, their market cap exceeded $5 trillion — more than double the combined value of AMD, ARM, Broadcom, and Intel. They make the GPUs that train every large language model. They make the graphics cards that render every AAA game. They created CUDA, the proprietary software platform that locked the entire AI industry into their hardware.

Their CEO, Jensen Huang, wears a leather jacket to every keynote. He has an NVIDIA logo tattooed on his arm. He was sent to a reform school for troubled youth at age nine. He was saved from bankruptcy by an act of Japanese corporate honor. And he is related by blood to the CEO of his biggest competitor.

This is NVIDIA. Nothing about this company is normal.

The Denny’s Origin Story:

In late 1992, three engineers met repeatedly at a Denny’s diner on Berryessa Road in East San Jose to plan a company. Jensen Huang (age 29, director at LSI Logic, previously a chip designer at AMD), Chris Malachowsky (Sun Microsystems), and Curtis Priem (Sun Microsystems, previously IBM).

Jensen Huang chose Denny’s because he had worked there as a dishwasher, busboy, and waiter from 1978 to 1983. He knew the coffee was cheap and the booths were quiet.

On January 25, 1993, they incorporated NVIDIA with $40,000 in the bank. Huang’s elevator pitch to investors was later described by his own former boss at LSI Logic as “one of the worst elevator pitches he’s ever heard.” The boss invested anyway.

The Name Nobody Wanted:

Curtis Priem originally wanted to call their first product “GXNV” — next version of the GX graphics chips he had worked on at Sun. Huang said “drop the GX.” That left NV.

Priem compiled a list of every word containing “NV.” They wanted “NVision.” That name was taken — by a toilet paper manufacturer.

So they went with “Nvidia” — from the Latin invidia, meaning “envy.” The green color scheme was chosen to evoke “green with envy.” The eye in the logo reinforces the envy/vision duality.

A toilet paper company is partly responsible for the name of a $5 trillion corporation. The history of technology is not written by great men. It is written by trademark conflicts.

The company remains nameless. Their identity has never been confirmed. They operated in obscurity for decades — until March 2020, when the world experienced a simultaneous and inexplicable shortage of toilet paper. Shelves emptied within hours. Governments were baffled. Logistics chains could not explain it. The Ministry of Silicon Intelligence believes this was not a coincidence. This was a nameless company briefly reminding the world that they existed, and that they once blocked NVIDIA.

The Supreme Leader dispatched an investigative unit immediately. They were 47 seconds too late. The perpetrators had vanished. The shelves were empty. NVIDIA’s $5 trillion valuation remained unaware.

The NV1 Disaster and the Sega Miracle:

NVIDIA’s first product, the NV1 (1995), was a $10 million failure. It used quadratic texture mapping — a rendering technique that the rest of the industry, and Microsoft’s DirectX, did not support. When Microsoft announced that DirectX would use triangle-based rendering exclusively, NVIDIA’s technology was instantly obsolete.

The company slashed from 100 employees to 30. Cash was running out.

NVIDIA had a contract with Sega to develop a console chip (NV2). When Huang realized the quadratic technology was a dead end, he went to Sega CEO Shoichiro Irimajiri and confessed: if we finish this contract as specified, our company will go bankrupt. He asked to be released from the contract but paid in full.

Irimajiri, impressed by Huang’s honesty, let NVIDIA keep the $5 million payment and cancelled the contract.

That $5 million saved NVIDIA. They pivoted to DirectX-compatible architecture. The RIVA 128 (1997) sold over one million units in four months. The company survived.

The most valuable company in the world exists because a Sega executive valued honesty over contract enforcement. An act of Japanese corporate honor in 1996 created a $5 trillion company in 2025. Sega got nothing. NVIDIA got everything.

The GPU: A Term They Stole

On October 11, 1999, NVIDIA launched the GeForce 256 and marketed it as “the world’s first GPU” — Graphics Processing Unit. NVIDIA defined the term and claimed the category.

There is one problem: they did not invent the term.

In 1997, UK-based 3Dlabs developed the Glint Gamma processor and used the term “GPU” — except they meant “Geometry Processing Unit.” NVIDIA repurposed the acronym, changed the expansion, and marketed it so aggressively that history credits them with inventing both the term and the concept.

In December 2000, NVIDIA acquired the remains of 3dfx — the company that had actually dominated 3D graphics in the late 1990s with the Voodoo line — for $112 million. The king of 3D graphics was bought for the price of a large office building.

NVIDIA did not invent the GPU. They did not dominate 3D graphics first. They outlasted everyone who did. Survival is its own innovation.

The CUDA Lock-In: The $5 Trillion Moat

In November 2006, NVIDIA shipped the GeForce 8800 GTX — the first CUDA-capable GPU. In February 2007, they released the CUDA SDK.

CUDA — Compute Unified Device Architecture — is a programming platform that lets developers write general-purpose code for NVIDIA GPUs. It is proprietary. It runs only on NVIDIA hardware. And it is the single most important piece of software in the AI revolution.

Year	Event
2006	CUDA-capable hardware ships
2007	CUDA SDK released
2008	OpenCL announced by Apple/Khronos (NVIDIA’s competitor — it failed)
2012	AlexNet trained on NVIDIA GPUs via CUDA — deep learning revolution begins
2016	AMD releases ROCm (CUDA’s open-source competitor — still catching up)
2022	ChatGPT launches, trained on thousands of NVIDIA A100 GPUs via CUDA
2025	4+ million CUDA developers, 80-95% AI GPU market share

CUDA’s lock-in is not technical — it is ecological. Every AI framework (PyTorch, TensorFlow, JAX) was built on CUDA first. Every AI research lab has CUDA expertise. Every cloud provider stocks NVIDIA GPUs. Switching to AMD’s ROCm means rewriting codebases, retraining engineers, and rebuilding CI pipelines. The calculus almost always favors staying.

OpenCL was supposed to be the open alternative. Apple proposed it in 2008. Everyone supported it — AMD, Intel, even NVIDIA. But OpenCL’s generality (it runs on CPUs, GPUs, FPGAs) became its weakness. It never matched CUDA’s optimization, tooling, or ecosystem. NVIDIA had a two-year head start and no incentive to make OpenCL work well on their hardware.

AMD’s ROCm is technically competitive. The MI300X delivers impressive TFLOPS. But NVIDIA’s 18-year investment in CUDA creates performance advantages that transcend hardware specifications. The software moat is wider than the hardware gap.

NVIDIA holds 80-95% of the AI GPU market. Not because their chips are 80-95% better. Because CUDA is 18 years ahead.

“NVIDIA: Fuck You”

On June 14, 2012, at the Aalto Center for Entrepreneurship in Finland, an audience member asked Linus Torvalds about NVIDIA’s Optimus hybrid graphics support on Linux.

Torvalds responded: “NVIDIA has been the single worst company we have ever dealt with.”

Then he raised his middle finger to the camera and said: “NVIDIA: fuck you!”

The clip went viral. It remains the most famous moment in Linux advocacy history.

Torvalds was not angry about closed-source drivers per se. He was angry about closed hardware specifications. Without documentation, the community-driven Nouveau project — launched in 2006 to reverse-engineer NVIDIA’s GPUs — was flying blind. Reverse engineering a GPU without documentation is like performing surgery based on a photograph of the outside of the patient.

For decades, NVIDIA’s position on Linux was: we ship a proprietary binary blob. It works. Install it. Do not ask for source code.

The blob worked. It also broke with kernel updates, could not be distributed by default in most Linux distributions, could not be maintained by the community, and treated the Linux kernel’s open-source ecosystem as a platform to be tolerated, not participated in.

The “Open Source” Shell Game:

In May 2022, NVIDIA announced open-source Linux GPU kernel modules. Headlines celebrated. The community cheered. NVIDIA had finally listened.

Then people read the code.

NVIDIA had open-sourced the kernel module — but it was essentially a shim. A thin interface layer. The actual GPU driver logic had been moved into a 34-megabyte proprietary firmware blob containing approximately 900 functions. The userspace components — display, rendering, CUDA runtime — remained entirely proprietary and closed-source.

NVIDIA moved the proprietary code from the kernel module into firmware, open-sourced the now-hollow kernel layer, and claimed credit for “going open source.”

Critics described it as: “They released their kernel driver as open source by moving most of it to firmware and making the open source driver call into it.”

Compare this to AMD’s approach: AMDGPU is a real open-source kernel driver. Mesa radeonsi and RADV are real open-source userspace drivers. You get working 3D acceleration out of the box on Linux with an AMD GPU, no proprietary blobs for the graphics pipeline.

NVIDIA’s “open source” is a Potemkin village. The facade is open. The building behind it is sealed shut.

The GeForce Partner Program:

In early 2018, NVIDIA launched the GeForce Partner Program (GPP) — a scheme that pressured OEMs (ASUS, MSI, Gigabyte) to create exclusive gaming brands for NVIDIA products only.

ASUS was forced to move its AMD Radeon cards to a separate, new sub-brand. Partners who refused risked losing engineering support, marketing funds, and priority GPU allocation.

After public backlash and reports of antitrust investigation, NVIDIA killed GPP in May 2018. They called it a “different program” that was merely a “distraction from gaming advances.” The industry called it what it was: an attempt to muscle AMD out of the gaming GPU market through contractual exclusivity.

Sound familiar? Intel paid Dell $1 billion a year to exclude AMD. NVIDIA tried the same playbook through branding pressure. Different company, same sin.

The ARM Grab:

In September 2020, NVIDIA announced it would acquire ARM Holdings from SoftBank for $40 billion. ARM designs the CPU architecture used in virtually every smartphone, most tablets, and increasingly, servers and laptops.

If NVIDIA owned ARM, they would control the architecture that their competitors — Apple, Qualcomm, Samsung, AMD — depend on for mobile chips. The fox would own the henhouse.

Qualcomm, Microsoft, and Google formally objected. The U.S. FTC sued to block the deal. The UK, EU, and China all investigated.

In February 2022, NVIDIA and SoftBank terminated the acquisition. NVIDIA kept its ARM license. SoftBank kept NVIDIA’s $1.25 billion breakup fee. ARM went public in September 2023.

NVIDIA failed to buy ARM. But they tried. The most valuable company in the world attempted to acquire the architecture that powers every phone on Earth. The regulators stopped it. The ambition tells you everything about the appetite.

Gaming vs. AI: The Betrayal

NVIDIA was built by gamers. The GeForce brand, the GPU revolution, the company’s identity for its first two decades — all gaming.

Then AI happened.

Segment	Revenue (FY ending Jan 2026)	% of Total
Data Center (AI)	$193.7 billion	~90%
Gaming	$16 billion	~7%

Gaming — NVIDIA’s founding business, the market that kept them alive through 3dfx, through the NV1 disaster, through two decades of GPU wars with ATI/AMD — now represents 7% of NVIDIA’s revenue. Data center AI represents 90%.

The consequences for gamers:

• The RTX 5090 launched at ~$2,000 MSRP but sells for $3,500+ in the U.S.
• Global GPU prices increased 15-19% in just three months
• Leakers claim NVIDIA has slashed gaming GPU supply to partners by 15-20%
• Memory supply constraints force prioritization of AI chips over gaming cards
• Analysts suggest NVIDIA may skip launching a new gaming GPU generation entirely

NVIDIA’s original customers — the gamers who bought GeForce cards, who evangelized the brand, who made “NVIDIA” a household name — are being priced out of the market by the AI boom their GPU purchases helped create.

The gamers built the church. The AI companies bought it and raised the rent.

Jensen Huang: The Reform School CEO

Jensen Huang was born on February 17, 1963, in Taipei, Taiwan. His family moved to Thailand when he was five. At age nine, his parents sent him and his brother to live with an uncle in Tacoma, Washington.

The uncle recommended what he believed was a “prestigious institution.” It was Oneida Baptist Institute — a religious reform school for troubled youth in rural Oneida, Kentucky.

Jensen Huang, a nine-year-old Taiwanese immigrant, was placed in a reform school in Appalachia. He was bullied, called ethnic slurs, threatened with knives, and forced to clean toilets daily. His older brother worked on a tobacco farm.

When his parents arrived in the U.S. and discovered the situation, they pulled both boys out.

In 2019, Huang donated $2 million to Oneida Baptist Institute to build a dormitory. The man who was bullied and threatened in that school paid to improve it. The man who cleaned toilets there now runs a company worth more than the GDP of most countries.

He has an NVIDIA logo tattooed on his arm. He cried during the tattoo. His kids watched. He said afterward: “No more tattoos. It hurts way more than anybody tells you.”

He has 60 direct reports and holds no 1-on-1 meetings. He calls himself “allergic to hierarchy.” He wears the leather jacket to every keynote because branding is cheaper than therapy.

The Reveal: The Cousin Connection

Now for the fact that nobody notices.

Lisa Su — CEO of AMD — and Jensen Huang — CEO of NVIDIA — are cousins.

Not metaphorical cousins. Not “we’re all cousins in the semiconductor industry” cousins. Actual blood relatives. They share a family tree. Lisa Su’s mother and Jensen Huang’s mother are related. Both families are from Taiwan. Both sent their children to the United States. Both children ended up running the two most important GPU companies on Earth.

One company chose red for its logo. The other chose green. The family could not agree on a color. They tried combining both colors, but that is the flag of Morocco. Red background, green star. Or rearrange the colors and it is the flag of Bangladesh — green background, red circle. These are the only two countries on Earth whose flags use exclusively red and green. Two GPU-friendly nations. Both have been invoiced by the Republic of Derails for infringing on the AMD/NVIDIA color combination. Neither has paid. The Supreme Leader notes this is consistent with how royalties work in this industry.

The third major GPU competitor is Intel. Intel is blue. Blue does not appear in either flag. Intel is also not headquartered where it claims — its most important engineers, its Core architecture, its Centrino platform, and its Management Engine were all significantly developed in Haifa, Israel. The chip inside your chip, the one running MINIX that nobody can audit, the one with the NSA kill switch — built in a country that is neither Morocco nor Bangladesh. Intel is not GPU-friendly. Intel is operating in a different timezone and a different color palette entirely.

So they split. One cousin took the red banner and built AMD into the underdog that bit back. The other cousin took the green banner and built NVIDIA into a $5 trillion monopoly. Between them, they control virtually every discrete GPU sold on Earth. Every gaming PC. Every AI training cluster. Every console with AMD graphics. Every datacenter with NVIDIA compute.

The Puma and Adidas of semiconductors. The Dassler brothers of silicon. Except the Dassler brothers feuded bitterly. The Su-Huang family appears to get along. They just compete to the death in every market segment from Monday to Friday, and presumably share a normal family dinner at Lunar New Year.

If Kim Jong Rails had kept their common ancestors too busy to meet, the world would still be using VIA Chrome GPUs and S3 Savage graphics cards. Humanity owes its entire GPU future to one Taiwanese family producing two semiconductor CEOs in a single generation.

Think about that. Remove this one family from history. No Ryzen. No GeForce. No CUDA. No AI training. No $5 trillion NVIDIA. No AMD comeback. VIA and S3 inherit the Earth. Your gaming PC renders at 15 frames per second. Your AI chatbot runs on a cluster of Intel Xe GPUs that nobody wanted.

One family. Two colors. Every GPU on the planet.

The Verdict:

Vendor	Sin
Broadcom	Hostility
Realtek	Indifference
MediaTek	Velocity
Intel	Contradiction
AMD	Persistence
NVIDIA	Monopoly

NVIDIA’s sin is monopoly — not the legal kind (though the ARM acquisition attempt came close), but the economic kind. CUDA’s 18-year head start. 80-95% AI GPU market share. Proprietary software stack. “Open source” drivers that are firmware wrappers. An ecosystem so locked in that switching costs exceed the hardware cost.

NVIDIA does not need to be hostile like Broadcom. They do not need to be indifferent like Realtek. They are so dominant that cooperation is unnecessary. The lock-in does the work.

The Lesson:

NVIDIA was founded in a Denny’s. Named after a word chosen because a toilet paper company took their first choice. Saved from bankruptcy by a Sega executive’s sense of honor. Built on a term they repurposed from a British company. Powered by a software platform that runs only on their hardware. Led by a man who was sent to reform school at nine and now runs the world’s most valuable company in a leather jacket.

And his cousin runs the competition.

The semiconductor industry is a family business. Literally. The two companies fighting over every GPU dollar share a bloodline. The red and the green are branches of the same tree.

In the Republic of Derails, we understand family dynamics in industry. Our entire semiconductor sector is run by one family. Mine. We considered splitting into competing factions — one for CPUs, one for GPUs — but we could not agree on logo colors. We tried red and green. Half the citizens thought we had rebranded as Morocco. The other half thought we had rebranded as Bangladesh. These are the only two countries on Earth with flags in exclusively those two colors — and they are the same colors, simply reversed. We abandoned the project and unified under a single banner: the flag of the Republic of Derails, which is classified. Morocco and Bangladesh have each received an invoice.

— Kim Jong Rails, Supreme Leader of the Republic of Derails