Chapter 03 · The Operating System

The Jensen Huang
Playbook

How Nvidia's Founder Actually Operates

Jensen Huang has run the same company for 33 years. It survived a first chip that failed, a lifeline from Sega, a decade in which CUDA ate the profits, and an AI boom nobody at Nvidia planned for in 2006. I traced the habits he credits: candor in public, low expectations, platforms over products. Some of it transfers. A lot of it rests on timing, a gaming cash machine and a stock that made leaving too expensive.

Explore the playbook The numbers
Portrait of Jensen Huang Peter Dasilva / © European Union, 2026 · CC BY 4.0 (cropped)
01 / The Numbers

The $5 Trillion Company

~$5.5T
Nvidia market value

As of 12 Oct 2026 (peak close ~$5.77T on 6 Oct 2026). First company to pass $4T (Jul 2025) and $5T (29 Oct 2025); $1T to $5T took about two and a half years. Moves daily.

$215.9B
FY2026 revenue

Up 65%, with Data Center $193.7B (about 90% of sales) and GAAP net income of $120.1B. Fiscal year ended 25 Jan 2026.

$96.2B
Latest quarter

Q2 FY2027 revenue (quarter to 26 Jul 2026), up 106%; net income $59.7B in one quarter, about the same as Nvidia's entire FY2024 revenue ($60.9B). Gross margin 75%.

~42,000
People

Employees in 38 countries, ~31,000 in R&D; turnover 3.7% (FY2026). Roughly $5.1M of revenue per employee (our arithmetic).

1993
Founded in a Denny's booth

Huang, Chris Malachowsky and Curtis Priem planned Nvidia at a Denny's in East San Jose; incorporated 5 Apr 1993 with Huang as CEO at 30.

33 years
CEO since day one

The only CEO Nvidia has had, through the NV1 flop, the 2008 and 2022 downturns and the AI boom (as of 2026).

$5M
Sega's lifeline

In 1995–96 Sega invested about $5M after Huang admitted the chip it had commissioned was on the wrong architecture ("six months to live", he later said).

~60
Direct reports

His own figure (Stripe Sessions, 2024), with no one-on-ones; reported as ~50 in 2023.

Huang's Bets

Bet Since Key milestones Outcome Role
NV1 1995 First chip; curved-surface rendering stranded when DirectX standardized on triangles. Flop; staff cut ~100 to ~40 (1996) First product bet
Sega rescue & RIVA 128 1996–97 ~$5M Sega investment bought ~6 months; RIVA 128 shipped Aug 1997 after emulator-based prototyping, reportedly ~1M units in 4 months. Survival CEO, negotiator
IPO & GeForce 256 1999 Nasdaq IPO 22 Jan 1999 at $12 ($43.3M net); GeForce 256 marketed as "the world's first GPU", Nvidia's own coinage. $43.3M raised CEO
Xbox 2000 Microsoft deal signed 5 Mar 2000, $200M prepaid; chips shipped from Jul 2001. $200M prepayment Supplier
CUDA 2006 General-purpose GPU programming on every GeForce; consumed gross profit for years; AlexNet trained on two GTX 580s in 2012. The install base behind the AI boom Bet owner
DGX-1 to OpenAI 2016 Huang hand-delivers the first DGX-1 (~$129k list) to OpenAI in San Francisco. Early tie to the lab that built ChatGPT Supplier
Mellanox 2019–20 Announced at $6.9B, closed ~$7B (27 Apr 2020); networking now $14.8B a quarter (Q1 FY27). China's SAMR later made a preliminary finding over approval conditions (Sep 2025). Networking business Acquirer
Arm (abandoned) 2020–22 $40B deal from SoftBank; UK CMA, EU and US FTC objected; terminated 8 Feb 2022. $1.36B charge Failed acquirer
The AI boom 2023–26 H100 demand after ChatGPT; $1T (May 2023) to first $5T company (Oct 2025); Vera Rubin unveiled at GTC 2026. ~$5.5T (Oct 2026) CEO
UK build-out 2021–25 Cambridge-1 (Nvidia's own ~$100M system, 2021); Isambard-AI (£225M UK-government facility on Nvidia GH200 chips, 2025); up to £11B partner AI factories and a £2B start-up fund pledged with partners (Sep 2025). Partner pledges, delivery in progress Builder / supplier / investor

Huang's 'portfolio' is one company and a sequence of bets inside it, several of which nearly killed it. Figures are as of October 2026; market value moves daily, and partner-announced totals (like the UK's £11B) are pledges, not Nvidia spending.

Reading note: Nvidia's fiscal year ends in late January, so FY2026 is roughly calendar 2025. Revenue, margin and headcount come from Nvidia's SEC filings; market value and net worth are point-in-time snapshots. For scale: Huang holds about 3.58% of Nvidia (870.6M shares as of 23 Mar 2026, per the proxy), behind a Forbes-estimated fortune of roughly $198B on 14 Oct 2026; his base salary is $1.5M.

Bigger than the UK economy, with an asterisk: At about $5.7T in mid-May 2026 Nvidia's market value exceeded UK GDP (~$4.26T per Euronews; the IMF's Oct 2025 estimate was ~$3.96T). Market value is a stock; GDP is an annual flow. Vivid, not like-for-like.

One quarter's profit is about a whole year's sales: Q2 FY2027 net income ($59.7B) against FY2024 revenue ($60.9B), both from Nvidia's own results.

02 / The Playbook

The Operating System

How Huang actually operates: a company run as if thirty days from failing, candor over information hoarding, platform bets that took a decade, and markets that did not exist yet. A taste below; open the full system to filter by domain.

Survival

Stay Thirty Days From Going Out of Business

Huang has run Nvidia for more than three decades as if it could die next month, and says the fear of failing drives him harder than the wish to win.

Read principle
Survival

Lower Your Expectations, Raise Your Resilience

Huang argues that people who expect things to go well break when they don't, and that character is built by setbacks, not talent.

Read play
Survival

Confront the Mistake, Ask for Help

When Nvidia's first chip bet on the wrong technology, Huang told his biggest customer the truth, and asked them to pay anyway.

Read play
Survival

Shake It Off and Skate to the Next Opportunity

Treat setbacks as part of the route: absorb them fast and move to where the next opening is.

Read play
Survival

Know the Real Price Before You Start

Asked whether he would found Nvidia again, Huang says no: the most honest thing a founder of a multi-trillion-dollar company has said about what it costs.

Read play
Candor

Kill the Information Monopoly

Power at Nvidia isn't supposed to come from knowing things others don't, so information goes to everyone at once.

Read principle
See all 22 plays

Read alongside Elon Musk Sam Altman Plays that echo or argue with each other are linked at the bottom of each page.

03 / Errata

The costs and the asterisks

Nvidia won so completely that its history now reads like a plan. It wasn't. The honest record keeps the failed chips, the disclosure fines, the blocked deal, the political entanglements and the culture questions next to the wins.

NV1: the wrong bet

NV1 (1995) chose curved surfaces just as Microsoft's DirectX standardized on triangles, and the follow-on Sega console chip was on the wrong architecture. Huang told Sega so and asked it to invest anyway: about $5M in 1996, as headcount fell from roughly 100 to 40. Sega later sold its stake for ~$15M. Nvidia survived because one counterparty chose to trust him.

Bumpgate: defective laptop chips (2008)

On 2 Jul 2008 Nvidia took a charge of roughly $150–200M for notebook GPUs with "abnormal failure rates". A shareholder suit alleged the company knew from about Aug 2007; a consumer class action involved Dell and HP laptops; Nvidia cut 6.5% of its workforce in Sep 2008 and the consumer case settled around 2010. The claims about when it knew are allegations, not findings.

Crypto: a fine and an unfinished lawsuit

On 6 May 2022 the SEC fined Nvidia $5.5M for not disclosing crypto mining's effect on Gaming demand in two FY2018 10-Qs, without Nvidia admitting or denying; the SEC's Kristina Littman said the failures "deprived investors of critical information to evaluate the company's business in a key market." A shareholder class action was revived by the Ninth Circuit (25 Aug 2023), the Supreme Court dismissed Nvidia's appeal (11 Dec 2024), and the FY2026 10-K shows it still being defended. No finding has been made against Huang personally; the outcome is open.

The Arm deal that regulators stopped

The $40B purchase of Arm from SoftBank (Sep 2020) drew "significant competition concerns" from the UK's CMA, an EU investigation and an FTC suit (Dec 2021). It was terminated on 8 Feb 2022 with a $1.36B charge. A UK-relevant reminder that scale invites scrutiny.

Caught between two governments

A US licence requirement on H20 chips (Apr 2025) produced a $4.5B charge. In Aug 2025 Nvidia agreed a 15% revenue share with the US government (the legal basis is unclear; officials called it voluntary); H200 sales were approved with 25% (Dec 2025/Jan 2026), yet the FY2026 10-K reports no H200 revenue. China's SAMR made a preliminary antitrust finding on 15 Sep 2025, and Nvidia says it is "effectively foreclosed" from China's data-center compute market.

Circular money: a $105B lease guarantee for OpenAI, and concentrated customers

A $100B letter of intent to invest in OpenAI (Sep 2025) shrank to a reported ~$30B investment, and its Aug 2026 10-Q discloses guarantees capped at $105B in total, covering "defined portions of lease and power payments" on an OpenAI affiliate's 20-year leases at SB Energy's Ohio campus. That is credit support payable on certain tenant defaults, not cash paid out, and the guarantees end on events including OpenAI achieving a satisfactory credit rating. Nvidia also holds stakes in CoreWeave, Intel and Nscale and put $17.5B into private companies in FY2026. Two direct customers accounted for 22% and 14% of FY2026 revenue, and supply obligations stand at $95.2B. Critics call the loop circular; defenders, including CoreWeave's CEO, say demand is real.

A demanding culture, paid for in stock

Feedback in public, about 60 direct reports and seven-day weeks; former staff told Bloomberg of shouting in meetings, and Nvidia declined to comment. Huang's line that he would "rather torture you into greatness" came with his own "tongue-in-cheek" gloss. Turnover was 2.7% in 2023 (Nvidia's fiscal 2024 sustainability report, via Bloomberg; semiconductor average ~17.7%) and 3.7% in FY2026 while shares rose more than 3,000% from the start of 2019. Retention was bought partly with equity. We found no sourced burnout data either way.

Antitrust questions, not findings

France's competition authority raided Nvidia's offices (Sep 2023) and has examined CUDA and cloud investments; a US DOJ inquiry was reported in 2024, including Run:ai; the FY2026 10-K says regulators in the EU, US, UK, China and South Korea have requested information. Nvidia's position: "Nvidia wins on merit." These are questions, not findings.

Noted for completeness: a Dec 2024 New York Times report said Huang's use of trusts and his foundation could cut his family's estate taxes by about $8B (legal planning, not alleged wrongdoing). And on 27 Jan 2025 Nvidia lost about $589B of market value in a day on the DeepSeek scare, the largest one-day loss on record. We keep them on the page so the record stays honest.

04 / Luck vs Skill

What actually transfers

Strip out the gaming cash machine, the AI timing and a stock that made leaving too expensive, and ask which of Huang's habits an ordinary founder can run.

Actually transfers

  • Tell the truth early when a project fails
  • Share information widely instead of hoarding it
  • Ask for a short written top-five instead of status meetings
  • Verify before you bet: prototype in simulation
  • Choose hard problems that take years to copy

Needs his platform, capital, or timing

  • Retain people with a rising stock
  • Run 60 direct reports with no one-on-ones
  • Put your platform on every chip you sell
  • Invest in your own customers
  • Negotiate market access with heads of state

The through-line of the caveats: Huang's most copied habits sit on top of three things nobody can copy: a gaming business that paid for a ten-year bet, deep learning arriving on hardware Nvidia already sold, and a share price that made staying rational. The honest claim is 'unusually well positioned and persistent'. Nobody can fairly say he 'foresaw everything'.

Q1

If your company were thirty days from failing, what would you stop doing tomorrow?

Q2

Which information in your organization is power only because it's hidden?

Q3

Is your 'zero-billion-dollar market' empty because it's early, or because nobody needs it?

Huang's most repeatable lesson is no tactic. A founder who stays long enough, admits mistakes fast and keeps the company ready can be standing in the right place when the luck arrives. He also says he wouldn't do it again.