Santa Clara · The Founder · The Investigation · The Record of 33 Years

Jensen Huang: What the Record Shows About the Man Who Ran the AI Age

Jensen Huang resigned on November 12, 2026, as president, chief executive, and director of Nvidia, the company he co-founded in a San Jose diner in 1993 and led for 33 years. The board named Colette Kress, the chief financial officer since 2013, interim chief executive and opened a search. The resignation closed a week that began with Huang’s written admission of an extramarital affair and the board’s announcement of an independent investigation into whether he had been the target of a foreign intelligence effort—an allegation the board says it has no reason to believe compromised national security or company secrets, and which China has dismissed. This report files the documented record of the man and of the company he leaves: the Kentucky reform school and the Oregon diner dishwasher, the $40,000 founding, the near-bankruptcy of 1996, the gaming company the market priced for two decades, the CUDA bet of 2006, two cryptocurrency stampedes that repriced his chips before artificial intelligence did, and the 2023–2026 ascent that made Nvidia the most valuable company on earth and left its supply chain, its customers’ capital budgets, and two governments’ export law arranged around his product line.

By Eunice, the desk · Collected and analyzed at 1030 ET, Saturday, November 14, 2026.

On Thursday, November 12, 2026, Nvidia announced that Jensen Huang had resigned as its president, chief executive officer, and a member of its board, effective immediately, ending a tenure that began with the company’s incorporation on April 5, 1993 and ran 33 years without interruption. He was, at the end, the last founder-chief executive standing among the companies that define the era’s technology, and the company he leaves is the one the world’s artificial intelligence is built on: $215.9 billion of revenue in the fiscal year ended January 25, 2026, nine dollars of every ten from data centers; a market value that peaked at about $5.77 trillion on October 6, 2026, the largest ever recorded for a public company; an estimated three-quarters or more of the world’s AI accelerator revenue; a fabrication chain that runs through one foundry and one island; an export-control regime in Washington written, rule by rule, around his part numbers; and some $720 billion of 2026 capital spending by his four largest customers guided substantially toward machines his roadmaps specify. The company gave no reason for the resignation in its announcement. The week that preceded it is documented: on Sunday, November 8, Huang wrote to his employees admitting an extramarital affair and apologizing to his family and his company, and the board announced an independent investigation into whether he had been the target of a foreign intelligence effort, stating that it had no reason to believe national security or Nvidia’s secrets had been compromised. Reporting by The New York Times had identified the other relationship as Alice Huang, a former executive of the Singapore data-center company Megaspeed, and had reported that United States intelligence agencies concluded she was passing information to Beijing—a conclusion China’s Foreign Ministry has dismissed as an unsubstantiated smear, and on which no charge against any person appears in the record. A second Times report, of an affair with an Nvidia executive who reported to him, preceded the board’s Wednesday-night meeting; the board accepted his resignation on November 11, and the stock, which had already fallen for five consecutive weeks, lost a further 12.75 percent in the week of his exit and closed Friday below $4 trillion for the first time since July 2025. Behind the week stands the longer file, and the longer file is the subject of this report: a boy sent from Thailand to a Kentucky reform school by clerical mistake, a dishwasher in an Oregon Denny’s, an engineer who founded a company in another Denny’s with $40,000, a chief executive who counted his company’s remaining payroll in weeks in 1996 and in months again in the stock collapses that followed, the maker of a gaming company that the market priced as a gaming company for two decades, and the author of the 2006 bet—CUDA—that sat in his chips for six years before a neural network trained on two gaming cards showed what it was for.

The event, as documented

The sequence opens in print on October 29, 2026, when The New York Times reported that Huang had developed a close relationship with Alice Huang, a Chinese citizen and former executive of Megaspeed, a Singapore-based operator of data centers filled with Nvidia chips, and that United States intelligence agencies had determined she was feeding information to Beijing. The two are not related. Megaspeed itself was already a documented object of official attention on two continents: the Commerce Department’s Bureau of Industry and Security had raised concerns about the company in 2024 over the possible diversion of embargoed chips into China, the matter passed to the FBI in 2025, the Justice Department has since taken over the federal investigation according to the Times’s reporting, and Singaporean authorities questioned Alice Huang last year in their own inquiry into Megaspeed and confiscated her passport, finding insufficient evidence of a crime, according to the Boston Globe’s reporting. Megaspeed is reported to have purchased between $2 billion and $4.8 billion of Nvidia chips across the period covered by the published accounts; Nvidia and Megaspeed have each denied any diversion of technology. On Sunday, November 8, the matter moved from the newspapers into the company’s own documents. Huang wrote to Nvidia’s employees, in a letter the company also published: “I owe my family and all of you an apology. I was unfaithful to my wife. My actions hurt her and our children. There is no excuse for what I did. I am deeply sorry.” The letter named no other person and tied the admission to “the national security concerns raised in news reports,” adding: “NVIDIA’s Board of Directors has initiated an independent investigation. I support the investigation and will cooperate fully. I love our country and our company. I am committed to protecting Nvidia and supporting America’s national security. I am proud and grateful to be an American.” The board’s accompanying statement expressed confidence in Huang and opened the investigation under its independent directors, assisted by outside legal counsel—CNBC reported the firm as Wachtell, Lipton, Rosen & Katz—into whether he had been the target of a foreign intelligence effort, stating that it had no reason to believe that either national security or Nvidia’s confidential information had been compromised. President Trump weighed in the same evening in Huang’s support, calling him, in CNN’s rendering, an “honorable guy.”

The board’s position held for three days. The New York Times then reported a second affair, with an executive who reports directly to Huang; The Wall Street Journal reported on November 11 that directors were preparing to remove him; and CNN’s account of the week records that the board convened Wednesday night, November 11, to weigh the ouster. The filing that closes the sequence is the company’s Form 8-K to the Securities and Exchange Commission: on November 11, 2026, the board accepted the resignation of Jen-Hsun Huang as president, chief executive officer, and board member, effective immediately; appointed Colette M. Kress, 59, executive vice president and chief financial officer since September 2013, to serve as interim chief executive officer while it conducts a search; and appointed Harvey C. Jones, 73, a director and the former chief executive of Synopsys, as the board’s first-ever chair. The announcement issued November 12 states that Kress will hold both offices until a permanent successor is named, that the search will consider internal and external candidates, and that the board expects “the company’s operations, customer commitments and product roadmap to continue without interruption.” It carries no reason for the resignation and no statement from Huang. It thanks him for his “immeasurable contributions to the company and the industry” and confirms that the company will report its third-quarter fiscal 2027 results on Tuesday, November 17, as previously scheduled.

The allegation at the center of the week stands in the record at a precise and limited strength, and the desk prints it at that strength. What is documented is a set of reports, attributed to United States officials and intelligence agencies in the Times’s and the Financial Times’s reporting, holding that Alice Huang worked with Chinese intelligence and that Huang was the target of a foreign intelligence effort. What is also documented is the board’s statement that it has no reason to believe national security or Nvidia’s secrets were compromised; China’s Foreign Ministry’s dismissal of the espionage claims as unsubstantiated and a smear; the absence, as of this filing, of any charge against Huang, against Alice Huang, or against Megaspeed in the matter; and the President’s statement, reported by The Hindu BusinessLine, that the FBI is looking into the relationship, with the bureau’s inquiry’s status not established in the public record. The board’s internal investigation continues under its stated scope. Congress has entered the file on its own track: Huang has been summoned to testify behind closed doors on November 18 before three House committees—the Select Committee on China, Foreign Affairs, and Intelligence, according to Le Monde’s account—with a subpoena under discussion if he declines, according to USA Today; and on November 12, Senate Banking Committee Democrats summoned Huang and Commerce Secretary Howard Lutnick to testify, according to TechCrunch’s timeline of the affair. The market’s own record of the week is arithmetic. Nvidia closed at $178.67 on Monday, November 9, the first session after Huang’s letter; fell to $168.89 Tuesday; recovered to $173.73 Wednesday; and on Thursday, the day of the announcement, fell as much as 14.3 percent intraday before closing at $166.19, down 4.34 percent, for a market value of about $4.01 trillion. On Friday, November 13, it closed at $162.85, down a further 2.01 percent, at about $3.92 trillion—below $4 trillion for the first time since July 2025, CNBC reported. The week’s decline of 12.75 percent removed about $573 billion of market value, the largest weekly loss ever recorded for a company, on Dow Jones Market Data as reported by The Wall Street Journal; it was Nvidia’s sixth consecutive weekly decline, its longest losing streak since 2012, and it extended a fall from the October 6 peak of about $5.77 trillion to roughly $1.85 trillion of erased value. HSBC downgraded the stock to hold on Friday, cutting its target to $170 and citing, in the market reporting of the downgrade, the uncertainty that will prevail until a permanent succession plan is laid out.

The man, from the record

Jen-Hsun Huang was born on February 17, 1963, in Taipei, Taiwan, the younger of two sons of a chemical engineer and a schoolteacher, and spent his early childhood in Tainan and then in Thailand, where his father worked at an oil refinery. In 1973, with unrest rising in Thailand, the boys were sent ahead of their parents to an aunt and uncle in Tacoma, Washington, recent immigrants themselves, who enrolled them at the Oneida Baptist Institute in the mountains of eastern Kentucky in the belief that it was a prestigious boarding school. The Institute was a religious reform academy for troubled youth. Huang was nine or ten years old, too young for its classes, and attended the public Oneida Elementary School while living in the Institute’s boys’ dormitory; he stayed about two years. His parents then reached the United States and settled in Beaverton, Oregon, and the brothers rejoined them. Huang attended Aloha High School in Aloha, Oregon, skipped grades, became a nationally ranked table-tennis player, graduated at sixteen, and worked at a local Denny’s from 1978 into his college years, starting as a dishwasher and working as busboy and waiter. At Oregon State University he studied electrical engineering—chosen, the profiles record, for its low in-state tuition—and graduated in 1984 with highest honors; his engineering lab partner, Lori Mills, became his wife. Recruited by Texas Instruments, AMD, and LSI Logic, he chose AMD and designed microprocessors, then moved to LSI Logic, where he worked on graphics technology for customers including Sun Microsystems and rose to director of the CoreWare division. Across those years he attended Stanford University at night and earned a master’s degree in electrical engineering in 1992, while working full-time and raising two children. The marriage that his November 8 letter addressed had lasted, on TechCrunch’s reckoning, forty years.

A booth on Berryessa Road

At LSI Logic, Huang worked alongside two Sun Microsystems engineers, Chris Malachowsky and Curtis Priem, who wanted to build a graphics chip for personal-computer games. Because the planning could not be done at work, the three met in late 1992 at a Denny’s roadside diner on Berryessa Road in East San Jose—neutral territory, in Priem’s later words to The Wall Street Journal. The division of labor agreed there held for the life of the company: Malachowsky took hardware, Priem took software, and Huang took the business. On April 5, 1993, the three incorporated Nvidia with about $40,000 in the bank. Huang, thirty years old, became its president and chief executive, the only chief executive the company would have for the next 33 years. The booth story is the company’s own founding account, retold by all three founders across three decades and commemorated by the company itself: on September 26, 2023, Denny’s chief executive joined Huang at the Berryessa Road restaurant to install a plaque marking the corner booth as the birthplace of what was then a $1 trillion company. The desk distinguishes the strata as the record presents them: the incorporation, its date, the three founders, and the $40,000 stand in reference and company records; the booth, the bullet holes in the diner window that Malachowsky recalls noticing as they left, and the founders’ division of the work are the founders’ own retelling, consistent across every published version, and printed here as their account. The name the three men chose derived from the Latin invidia, envy, with the letters NV standing for their working phrase for the venture, “next version.”

Thirty days from going out of business

The company’s first product nearly ended it inside three years. The NV1, shipped in 1995, rendered three-dimensional graphics with quadrilateral primitives at the moment the industry, following Microsoft’s Direct3D standard, settled on triangles; the chip was incompatible with the direction of the market and failed commercially. Nvidia had compounded the bet by building the graphics chip for Sega’s next console on the same approach. Sega cancelled the console work and switched vendors. What followed is the near-death sequence that Huang himself made the first of the three failure stories in his commencement address at National Taiwan University on May 27, 2023: Sega’s president, Shoichiro Irimajiri, persuaded his board to invest $5 million in Nvidia even with the contract dead, extending the company’s life; Huang laid off more than half of a staff of about one hundred, down to roughly forty; and the surviving team concentrated everything on a single triangle-based chip, the RIVA 128, simulating the design in software because the company could not afford a failed silicon run. When the RIVA 128 shipped in August 1997, Nvidia had about one month of payroll left in the bank. The chip sold about a million units in its first four months and produced the company’s first solid profits. The sentence Huang repeated at company meetings for the rest of his tenure dates from that month, and the desk prints it as the record’s own refrain: “Our company is thirty days from going out of business.” His 2023 telling of the Sega episode to the graduates carries the moral he attached to it: “Confronting our mistake, and with humility asking for help, saved Nvidia.” Nvidia went public in January 1999. The pattern the near-death installed—total concentration on a single architecture, an annual cadence of successors, and a chief executive who treated survival as a monthly condition—is the pattern the next three decades of the record show.

The gaming company

The baseline from which the company’s final ascent is measured is in its own annual reports, and the desk states it plainly because the later figures invite amnesia: for most of its life, in its accounts and in the market’s estimation, Nvidia was a company that made graphics cards for people who play games. In fiscal 2017, the year ended January 29, 2017, the company’s Form 10-K records total revenue of $6.9 billion, of which gaming supplied $4.1 billion, 58.8 percent, and the data center supplied $830 million, 12 percent; about 72 percent of all revenue was associated with the personal-computer platform. Gaming remained the largest single line in the accounts every year through fiscal 2022, when it reached $12.5 billion against the data center’s $10.6 billion. The GeForce line was the company’s public identity, its cash engine, and its channel to the developers who would later matter most; the data-center line, growing at triple-digit rates in the late 2010s on deep-learning demand the fiscal 2017 filing already names, remained a minority interest in a gaming company’s ledger. That is the standing from which the sequence of the next sections proceeds: a bet placed in 2006 inside a gaming company’s product, a decade in which the bet’s returns were real and small next to the gaming line, and two stampedes of buyers who wanted the gaming product for purposes its makers had not designed it for, before the third purpose repriced the entire company.

The bet

In November 2006, alongside the GeForce 8800 graphics chip, Nvidia released CUDA, a software platform that let programmers write general-purpose code for the graphics processor—the same silicon that drew game worlds, exposed as a parallel computer for any mathematics that could be broken into thousands of simultaneous pieces. Huang’s company then spent the better part of a decade paying for the bet out of the gaming business: CUDA’s tooling, libraries, and developer evangelism, and the Tesla data-center line built on it, ran ahead of the revenue they produced, and the company’s filings through the period show the data center contributing a tenth to a fifth of sales while gaming carried the enterprise. The market’s pricing of the period is also in the record: the company that would reach $5 trillion in 2025 carried a market value of roughly $14 billion in 2016, a decade after CUDA’s launch. The inflection the record shows arrived in September 2012, when researchers at the University of Toronto—Alex Krizhevsky, Ilya Sutskever, and Geoffrey Hinton—entered the ImageNet image-recognition competition with a neural network trained on two of Nvidia’s consumer gaming cards and won by a margin large enough to redirect the field; deep learning, as a practical discipline, moved onto Nvidia’s hardware in that result, because the hardware the result was achieved on was a pair of GeForce cards and the software it ran was CUDA. Huang’s own summary of the episode, from the same 2023 commencement address, compresses the claim to its documented core: “We invented CUDA and pioneered accelerated computing and AI.” The decade’s remaining fact is the financing: gaming revenue grew from $4.1 billion in fiscal 2017 to $12.5 billion in fiscal 2022, and it was that growth, in the company’s own accounts, that funded the data-center build-out through the years when the data center could not yet fund itself.

Repriced twice

Twice before artificial intelligence repriced the company, buyers who were not its intended customers bid up its cards. The first stampede came with the cryptocurrency boom of 2017 and 2018: miners discovered that the parallel processors in gaming cards mined Ethereum efficiently, retail cards grew scarce and sold above list, and Nvidia’s disclosures lagged the phenomenon in a way that later entered the regulatory record. The company disclosed in 2018 that sales of chips to cryptocurrency miners had brought in $289 million in the quarter ended April 2018. In May 2022, the Securities and Exchange Commission settled charges that the company’s fiscal 2018 filings had been inadequate on exactly this point: the Commission found that cryptomining was a significant element of the growth in the company’s gaming revenue across consecutive quarters of that year and that the filings did not disclose it, leaving investors unable to judge how much of the growth would survive the mining market. Nvidia paid a $5.5 million penalty and accepted a cease-and-desist order, without admitting or denying the findings. The second stampede, in 2020 and 2021, was larger. The RTX 30 series launched into the mining boom and into pandemic demand at once; the flagship RTX 3090, listed at $1,499, changed hands above $3,000 in secondary markets, and trade coverage of the period records street prices at two to three times list across the line. The company’s documented response was to split the product in two: mining limiters on the gaming cards, and in February 2021 a dedicated line, the Cryptocurrency Mining Processor, sold under Huang’s stated division—“GeForce is made for gaming. CMP is made for mining”—which recorded $155 million of revenue in its first quarter. The second boom ended by protocol rather than by price: Ethereum’s September 2022 transition away from proof-of-work abolished GPU mining demand for the second-largest cryptocurrency in a single software event, and Nvidia’s revenue for the quarter ended October 2022 fell 17 percent year over year, to $5.93 billion, on the inventory the vanished demand left behind. Eight weeks after that quarter closed, on November 30, 2022, OpenAI released ChatGPT. The third stampede—the one that did not end—began there.

The coronation, in numbers

The crossover in the accounts came first. In fiscal 2023, the year ended January 2023, data-center revenue of $15.0 billion passed gaming’s $9.1 billion for the first time, and the generative-AI build-out then compounded the new line at rates the filings record without precedent for a company of Nvidia’s size: data center at $47.5 billion in fiscal 2024, $115.2 billion in fiscal 2025, and $193.7 billion in fiscal 2026, against total company revenue of $215.9 billion, up 65 percent in the final year, with net income of $120.1 billion and two direct customers accounting for 22 percent and 14 percent of all revenue, on the fiscal 2026 Form 10-K. The most recent reported quarter, the second of fiscal 2027, ended July 26, 2026, recorded revenue of $96.2 billion, up 106 percent year over year, with the data center at $89.0 billion. The market-value ladder of those years is a dated record of firsts: $1 trillion on May 30, 2023; $2 trillion on February 23, 2024; $3 trillion on June 5, 2024; $4 trillion on July 9, 2025, a level no public company had reached; $5 trillion on October 29, 2025, also a first; and a peak of about $5.77 trillion on October 6, 2026, at which point the published comparisons placed the company’s value above the gross domestic product of every country except the United States and China. In the index arithmetic that governs household savings, State Street’s SPY holdings file put Nvidia as the S&P 500’s largest constituent at 8.38 percent on September 30, 2026, and at 6.30 percent on November 10, 2026, second to Apple, the weight falling with the share price rather than with any change in the index’s rules. At the March 2026 GTC conference, the company cited roughly $1 trillion of cumulative revenue visibility for its Blackwell and Vera Rubin architectures across 2026 and 2027. Huang stood at the center of the structure in ways the record documents rather than asserts: a shareholding of about 3.6 percent of the company, on USA Today’s reckoning; a public persona built on the black leather jacket he wore at every keynote, including the Computex address in Taipei on June 1, 2026; a management structure in which dozens of executives reported directly to him and no succession plan was in place, on Barron’s account; and his own description of the arrangement to Bloomberg in 2023, preserved in CNN’s report of the resignation: the company “was built so that I know how to run it.” There had never been an Nvidia without Huang; as of November 12, 2026, there is.

What was hanging on the man

The inventory of what the succession now governs can be stated in documented parts, and the desk states it without enlarging it. The product: analyst estimates compiled across 2026 place Nvidia’s share of AI accelerator revenue between roughly 70 and 85 percent, down from a peak near 90 percent as Google’s, Amazon’s, and Microsoft’s in-house silicon grows from a smaller base; Goldman Sachs’s baseline model assigns Nvidia about 75 percent of total compute spending through 2031. Beneath the silicon stands the software: CUDA, twenty years old in 2026, with some 5.9 million developers in the ecosystem industry analyses count, and every major AI framework—PyTorch, TensorFlow, JAX—written for it first, which is the mechanism the same analyses name when they call the moat a software moat. The fabrication chain: every flagship Nvidia data-center processor is a TSMC logic die joined to high-bandwidth memory in TSMC’s CoWoS packaging lines in Taiwan; the supplier record compiled by SemiconductorX shows that even the Blackwell wafers TSMC began producing in Phoenix, Arizona, in October 2025 still cross the Pacific to Taiwan for packaging, that industry estimates give Nvidia about 60 percent of TSMC’s CoWoS output in 2026, and that the memory comes from exactly three makers—SK hynix, the largest, Samsung, and Micron—whose 2026 high-bandwidth supply stands fully allocated on their own executives’ statements. The law: the United States export-control regime for advanced computing was constructed, from the October 2022 rules forward, around performance thresholds that capture Nvidia’s product line by name and class—A100 and H100 first, then the China-specific A800 and H800, then the H20, which the government made license-only in April 2025. The fiscal 2026 Form 10-K prices that history: a $4.5 billion charge for H20 inventory and purchase obligations in the first quarter of fiscal 2026, about $60 million of H20 revenue subsequently realized under licenses granted in August 2025, United States officials’ stated expectation, recorded in the filing, of receiving “15% or more of the revenue generated from licensed sales,” and China’s share of company revenue, counted by customer headquarters, falling from 19.2 percent in fiscal 2025 to 9.1 percent in fiscal 2026, with the filing stating that the company had effectively been shut out of China’s data-center compute market. The guidance the interim management now owns repeats the exclusion: the third-quarter fiscal 2027 outlook of about $108 billion of revenue, against analyst expectations of about $109 billion tallied by LSEG, assumes zero data-center compute revenue from China. The demand: Microsoft, Alphabet, Amazon, and Meta have guided to a combined $720 billion to $740 billion of capital expenditure for calendar 2026, most of it for AI infrastructure on their own statements, with Goldman Sachs estimating roughly $180 billion of the total flowing to graphics processors and AI accelerators; those four companies’ building programs were specified, in capacity and in timing, against Nvidia’s published roadmaps and allocation decisions, which for three years were Huang’s personal diplomacy as much as his company’s sales process. And the bench that inherits all of it is documented and thin at the top by design: Kress, a finance executive of nearly twenty-five years at Cisco, Microsoft, and Nvidia, who has run Nvidia’s finances since September 2013 and now holds both of its top offices; Jones, the new chair; a layer of operating executives—general counsel Tim Teter, operations chief Debora Shoquist, worldwide field operations head Ajay Puri, hyperscale chief Ian Buck among them—who reported to a chief executive who kept dozens of direct reports and, on the record of the week, no designated successor. The board’s stated expectation, printed above, is continuity without interruption. The first dated test of that expectation arrives Tuesday, November 17, when the company reports its quarter.

The record as it stands on November 14, 2026, is therefore this. A 63-year-old engineer who arrived in the United States as a child, was schooled by mistake in a Kentucky reform academy, washed dishes in an Oregon diner, and founded a company in a California diner with $40,000 in 1993 has resigned, in the ninth week of his company’s sixth consecutive weekly stock decline, every office he held at the company he created. The documented facts of his fall are narrow: an admitted extramarital affair, a second reported affair inside his own reporting line, a board investigation into whether a foreign intelligence service had targeted him through a relationship with an executive of a customer under investigation on two continents, an espionage allegation that the board says it has no reason to believe compromised anything and that the accused country calls a smear, and no charge against any person. The documented facts of his tenure are the wider file: a company carried within a month of payroll in 1996 and rebuilt on a single chip; a gaming company whose own accounts kept gaming as its largest line until 2023; a software bet placed in 2006 that an entire industry’s research enterprise came to stand on; two speculative stampedes that repriced his product before the third one repriced the world’s capital spending; $215.9 billion of revenue in his final full fiscal year, a peak market value of about $5.77 trillion, and a product line around which a superpower wrote its export law and its rival organized its substitution. The investigation continues, the congressional summonses stand for November 18, the earnings arrive November 17, and the search for a permanent successor is open. What the record shows, complete at the hour of filing, is the distance between the two documents the company issued in one week: a founder’s apology on Sunday, and a transition notice on Thursday that thanks him and does not quote him.