The Entire History of NVIDIA: From a Denny’s Booth to a $5 Trillion Empire — and the Four-Day Fall

A kid who washed dishes in a Denny’s ends up running the most valuable company on Earth — then loses it all in four days. That is the whole Nvidia story in one sentence, and it is all true. This guide tells it chapter by chapter — including what it means if you buy graphics cards or follow Canada’s bet on AI computing.

Watch the video version:

NVIDIA at a glance: the timeline

YearWhat happened
1993Huang, Malachowsky, and Priem found Nvidia at a Denny’s in San Jose
1995First chip, the NV1, flops — it bets on the wrong 3D standard
1997The RIVA 128 sells ~1 million units in four months and saves the company
1999IPO at $12 a share; GeForce 256 launches, marketed as the first GPU
2006CUDA launches — graphics chips become programmable for general computing
2012AlexNet, built at the University of Toronto on two Nvidia gaming cards, launches the deep-learning boom
2017Volta adds Tensor Cores, hardware built specifically for AI
2022–2026Hopper, Blackwell, and Vera Rubin turn Nvidia into a data-centre company
Oct 2025Nvidia becomes the first company worth $5 trillion
Nov 2026Huang resigns; Colette Kress becomes interim CEO; the board rules out his return

1. The Denny’s booth (1993)

Jensen Huang’s first job was at a Denny’s in Portland, Oregon — 15, washing dishes and clearing tables. Fifteen years later he walked into a different Denny’s, on Berryessa Road in San Jose, and sat down with engineers Chris Malachowsky and Curtis Priem to sketch out a company: a chip that could draw realistic 3D graphics on a PC. Almost nobody thought that was a business.

They incorporated it anyway, on April 5, 1993, with $40,000 in the bank; Sequoia and Sutter Hill later put in $20 million. The company was called Nvidia, from the Latin word for envy. Huang was 30, fresh from a senior chip-design job at LSI Logic, and would run it for 33 years. The booth is still there — Denny’s put a plaque on it in 2023.

2. The chip that nearly killed it (1995–1997)

The first chip nearly killed the company. The NV1, launched in May 1995, drew 3D scenes using four-sided shapes while the industry standardised on triangles and Microsoft’s new DirectX software. Nvidia had bet on the wrong standard, and the NV1 flopped. Worse, a deal to build a chip for Sega’s new console collapsed. Huang flew to Japan and told Sega’s president the chip was a dead end. Sega paid Nvidia $5 million anyway — money that bought time.

Huang cut the staff from about 100 to about 40 and bet everything left on one new chip built around triangles. The team tested it in a simulator before a single chip was manufactured, and Huang ordered it into production sight unseen. When the RIVA 128 shipped in August 1997, Nvidia had about one month of payroll left. It sold about a million units in four months. Huang turned the scare into a motto: the company is 30 days from going out of business.

3. Inventing the GPU (1999)

In January 1999, Nvidia went public, selling 3.5 million shares at $12 each and raising $42 million. Later that year it launched the GeForce 256, marketed as the world’s first graphics processing unit — the first GPU. It moved work computers did in software onto the chip, and made Nvidia the best-known name in PC gaming graphics.

Honesty note: the “first GPU” claim was Nvidia’s marketing, disputed by hardware historians — but the word GPU comes from that campaign, and it stuck. Gaming paid the bills for 20 years while Nvidia grew into a respected but unremarkable chip maker. That was about to change, because of a bet that looked like a distraction.

4. The bet Wall Street laughed at: CUDA (2006)

In November 2006, Nvidia unveiled CUDA: let programmers use the graphics chip — built for thousands of small calculations at once — for any kind of maths, not just pictures. The hardware was the $599 GeForce 8800 GTX; the inventor was Ian Buck, who had built an early version at Stanford. The tools shipped in 2007.

For years, CUDA looked like an expensive hobby: Nvidia poured money into hardware, software, and tools while the payoff stayed invisible, and Wall Street saw a gaming company funding science projects. Huang kept funding it.

What CUDA built was not just a chip but a moat: every researcher who learned Nvidia’s way, every software library written for it, and every course that taught it made the next buyer more likely to choose Nvidia too. Competitors could match the hardware; matching 15 years of accumulated software was the hard part. When the AI boom arrived, it arrived on CUDA because that was where the programmers already were.

5. The AI accident (2012–2017)

The someone who needed CUDA at enormous scale was in Toronto. In September 2012, three researchers at the University of Toronto — Alex Krizhevsky, Ilya Sutskever, and Geoffrey Hinton — entered a computer vision contest called ImageNet. Their neural network, AlexNet, trained on two Nvidia gaming cards, won by a landslide: a top-5 error rate of 15.3%, against 26.2% for the runner-up. The breakthrough that launched the modern AI boom was Canadian — and it ran on gaming hardware.

Within months, every major tech company was building deep learning teams on Nvidia hardware. In 2017, the Volta design added Tensor Cores, hardware built for AI maths, and the data centre business started its climb. Gaming was still the biggest business — not for long.

The Toronto connection runs deeper: Hinton, at the university since 1987, shared the 2024 Nobel Prize in Physics with John Hopfield for foundational work on neural networks, and co-founded Toronto’s Vector Institute, one of Canada’s national AI hubs.

6. The empire (2022–2026)

Then came the empire. Hopper arrived in 2022, Blackwell in 2024, and Vera Rubin in January 2026 — each generation sold not as a chip but as a whole rack of computing. The numbers became hard to believe:

Fiscal 2026 (year ended Jan 25, 2026)Amount
Total revenue$215.9 billion
Data centre revenue$193.7 billion (89.7%)
Gaming revenue$16.0 billion (7.4%)
Net income$120.1 billion

Gaming, the business that built the company, was now under a tenth of it. In October 2025, Nvidia became the first company worth $5 trillion, closing at $5.03 trillion; by spring 2026 it was worth around $5.5 trillion, and in early October the shares closed at a record $239.24, within reach of $6 trillion. Huang was the most powerful person in AI. Then, in four days in November, it ended.

7. The fall: four days in November 2026

On November 8, Huang told employees he had been unfaithful to his wife. The same day, the board opened an independent investigation, run by the law firm Wachtell Lipton, after reports Huang had been targeted by a foreign intelligence operation. On November 11, the board accepted his resignation, announced the next morning with no reason given. CFO Colette Kress became interim chief executive, keeping her finance role, and Harvey Jones, a director since 1993, became the board’s first-ever chairman.

The shares fell 13% when trading opened on November 12, then recovered to close down about 4%. By Friday, the stock closed at $162.85 — roughly 32% below its early October record, about $1.8 trillion of market value erased.

The New York Times and the Financial Times reported that American intelligence agencies suspect Alice Huang — no relation — a former Megaspeed executive, was recruited by China’s Ministry of State Security and targeted Jensen Huang. That is an allegation from unnamed officials: Beijing denies it, Reuters could not verify it, no charges have been brought, and the investigation has made no findings public.

But Megaspeed sits at the centre of the fight over Nvidia’s most advanced chips. The Information reported that at least $7.5 billion of Nvidia chips reached China through Megaspeed and a Malaysian subsidiary over two years; Megaspeed says its business is legitimate and lawful, and denies wrongdoing. (AFP put the figure at upwards of $4 billion; neither number is settled fact.) Nvidia cut Megaspeed off in June 2025, the Justice Department is running a criminal inquiry, and four government departments wanted Megaspeed on America’s entity list — but the Commerce Department, which runs the list, kept delaying. A petition calling for Huang’s return passed 190,000 signatures, and investors Brad Gerstner and Gavin Baker urged the board to keep a return open. Even the Denny’s plaque came down after the resignation; days later, it went back up.

The wider picture matters more than any one executive. Reuters mapped more than $3 billion in Chinese AI computing projects built around restricted Nvidia processors, and a forensic report from Unemployment Labs said six of China’s newest AI models carry the numerical fingerprint of Blackwell chips — evidence from the models’ own weights that export controls leak. China’s substitutes are not ready: SMIC makes mostly mature-node chips, and ByteDance’s in-house chip effort lost its lead after its first chip fell short. We covered the China angle in our special weekend brief and standalone video.

8. What happens next — and what it means if you buy chips

The first real test came on November 17, when Nvidia reported quarterly results under Kress. The business was fine: revenue of $115.8 billion, up 103% from a year earlier, with $107.6 billion from data centres. Kress forecast $130 billion for the next quarter — above expectations — assuming no data centre revenue from China at all, and confirmed Nvidia has stopped shipping its licensed Hopper chips to China after selling $2.4 billion of them last quarter. New Vera Rubin systems delivered nearly $20 billion in their first production quarter, the fastest ramp in company history. The shares rose after hours.

That morning, Huang had posted that he would be honoured to return if the board wanted him. Opening the earnings call, chairman Harvey Jones answered: the board had unanimously accepted the resignation and did not intend to reinstate him. The CEO search is open; Kress leads the company in the meantime.

Three questions remain. First, what do the investigations find? The board’s inquiry is open, and the Wall Street Journal reported the SEC has opened a civil probe into whether Nvidia’s disclosures misled investors. Huang is due before House committees on December 1, the Senate Banking Committee has asked him to appear on December 3, and Commerce Secretary Howard Lutnick testifies on December 10. Second, export controls: Congress is pushing to tighten them, with the AI Overwatch Act aimed at the annual defence bill. Third, can China fill the gap? Huawei’s Ascend chips are produced in the low hundreds of thousands, and its executives say domestic demand will not be met until 2030. Huang himself once said Nvidia’s share of the Chinese market had fallen from 95% to zero.

For buyers, the read-through is straightforward. If you build PCs, nothing here changes what is on shelves: gaming is a small slice of Nvidia’s business, and production plans did not change with the CEO. If you run AI workloads or buy cloud computing by the hour, the story is supply, not leadership — demand still outruns supply, and the export-control fight decides who gets served first.

The Canadian angle

Canada sits on both sides of this story. The AI boom started, in part, in a Toronto lab — and Canada is spending public money to keep its computing. Budget 2024 committed C$2 billion over five years to the Canadian Sovereign AI Compute Strategy: up to $700 million for commercial AI data centres, up to $1 billion for public supercomputing, and up to $300 million for an AI Compute Access Fund. In April 2026, roughly $890 million more over seven years was added for a large Canadian-owned AI supercomputer. Nvidia has named Canada among the countries driving its sovereign AI business.

Here is the catch: Canada does not fabricate leading-edge AI chips at home. Nvidia designs its chips but does not manufacture them — TSMC in Taiwan fabricates the advanced GPUs. So Canadian labs and data-centre builders shop in the same constrained global market as everyone else, competing for the same TSMC capacity and Nvidia systems. The sovereign compute strategy is, in effect, Canada buying its way to the front of that queue with public money. The number to watch is not the share price but whether the domestic compute actually gets built.

For a parallel story of a tech empire meeting reality, see our guide to the rise and fall of the Cybertruck.

Frequently asked questions

When was Nvidia founded?

Nvidia was incorporated on April 5, 1993, by Jensen Huang, Chris Malachowsky, and Curtis Priem, after the three sketched it out at a Denny’s on Berryessa Road in San Jose. Huang ran it for 33 years until November 2026.

What is CUDA, and why does it matter?

CUDA, launched in November 2006, lets programmers use graphics chips for general computing. It built an ecosystem — 15-plus years of software libraries, courses, and trained programmers around Nvidia’s way of doing things — and that ecosystem, more than any single chip, is why the AI boom ran on Nvidia hardware.

What happened to Jensen Huang in November 2026?

On November 8, 2026, Huang told employees he had been unfaithful to his wife, and the board opened an independent investigation the same day after reports of a foreign intelligence operation targeting him. The board accepted his resignation on November 11, announced November 12. The spying claim remains an allegation — unnamed officials, denied by Beijing, unverified by Reuters, no charges brought. CFO Colette Kress became interim CEO, and the board says it does not intend to reinstate Huang.

Is Nvidia still selling chips to China?

Mostly no, for data-centre chips. Kress confirmed Nvidia has stopped shipping its licensed Hopper chips to China, after selling $2.4 billion of them the previous quarter, and the next quarter’s forecast assumes no data centre revenue from China.

Who runs Nvidia now?

Colette Kress, the CFO, is interim chief executive while keeping her finance role. Harvey Jones, a director since 1993, became the board’s first-ever chairman. A search firm is considering internal and external candidates for a permanent CEO.

Does any of this affect graphics card prices in Canada?

Not directly. Gaming cards are under a tenth of Nvidia’s revenue, and production plans did not change. What affects Canadian buyers is data-centre demand competing for the same manufacturing capacity, and the exchange rate — GPUs are priced in US dollars. Watch availability and street prices, not boardroom headlines.

🛍️ Shop this guide

Nvidia’s story starts with gaming hardware — and the cheapest way to understand what a GPU actually does is to have one in your own machine. These picks are the practical starting points for anyone the story inspired to build or upgrade:

  • NVIDIA GeForce RTX graphics card — the modern descendant of the GeForce 256: the card that plays games today and runs local AI models tomorrow — the heart of any gaming or AI PC.
  • High-wattage PC power supply (750W+) — modern graphics cards are power-hungry; a quality 750W or larger unit is the unglamorous part every GPU upgrade needs first — check wattage before you buy the card.
  • 2TB NVMe SSD — local AI models, game libraries, and datasets all eat storage; a fast 2TB drive is the single best quality-of-life upgrade for a PC — room for models, games, and everything else.
  • Mid-tower PC case with mesh airflow — GPUs run hot, and a mesh-front case with room for full-length cards keeps temperatures (and fan noise) down — cooling is performance.
  • 1440p 144Hz gaming monitor — a capable GPU deserves a display that can show what it renders; 1440p at high refresh is the sweet spot for most desks — see what the card can actually do.

As an Amazon Associate we earn from qualifying purchases — at no extra cost to you. Read our affiliate disclosure.

Sources and verification notes

All historical and November 2026 figures in this guide were verified on Nov 18, 2026, and are logged with sources in the companion fact-check file (58 figures). Key sources:

New figures added for this guide (verified Nov 18, 2026):

  • Geoffrey Hinton shared the 2024 Nobel Prize in Physics with John Hopfield (announced Oct 8, 2024), and co-founded Toronto’s Vector Institute — NobelPrize.org via Beginners in AI ; corroborated by TellWorld
  • Canada’s Sovereign AI Compute Strategy: Budget 2024 committed C$2 billion over five years (announced Dec 5, 2024) — up to $700M AI Compute Challenge, up to $1B public supercomputing, up to $300M AI Compute Access Fund — Cassels ; corroborated by Prompt20 sovereign AI guide
  • AI Sovereign Compute Infrastructure Program, April 2026: ~$890M over seven fiscal years for a large Canadian-owned AI supercomputer — canada.tedlee.ca ; corroborated by AIGNITRON
  • Nvidia named Canada among the countries driving its sovereign AI business — Prompt20 sovereign AI guide (summary of Nvidia investor disclosures; single source, stated with attribution in the guide)
  • Nvidia is fabless; TSMC in Taiwan fabricates its advanced GPUs (Hsinchu/Tainan fabs) — Disintermediate ; corroborated by TariffTracker and VaaSBlock

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top