# Nvidia after Jensen Huang Jensen Huang resigned November 12, 2026. Colette Kress is Nvidia's interim CEO. What it means for Vera Rubin, CUDA, AI chips, and productivity. - Published: 2026-11-14 - Author: Neeraj Nathany, Founder, Faraday - Canonical: https://faraday.email/blog/nvidia-ceo-transition-ai-chips-2026 ![NVIDIA CEO transition, November 2026, on a dark copper gradient](https://faraday.email/assets/blog/blog-nvidia-ceo-transition-2026.jpg) **Jensen Huang resigned as Nvidia’s president, chief executive, and a member of the board, effective immediately.** The board accepted it on November 11, 2026, and Nvidia announced the **CEO transition** on November 12. **Colette Kress**, CFO since 2013, is interim CEO and keeps the finance job. **Harvey C. Jones**, a director since 1993, is the company’s first chair. The permanent search covers candidates inside and outside Nvidia. Third-quarter fiscal 2027 results stay on the calendar for Tuesday, November 17. > The racks already sold are still arriving. The annual rhythm he kept in person is what can slip. **Vera Rubin** stayed. Google Cloud’s A5X instances, built on Vera Rubin NVL72, reached general availability on November 6. What left is the habit of putting foundries, memory makers, clouds, and governments on one person’s calendar. Read the week as three clocks — silicon, capital, and the work on your desk — and then decide which one you can actually move. ## What the board actually put in writing The 8-K names Jen-Hsun Huang and dates the acceptance to November 11. The press release is November 12. Kress, 59, signs as interim chief executive, executive vice president, and chief financial officer. She has run finance since September 2013, after nearly 25 years at companies including Cisco and Microsoft. Jones, 73, has been a director since the founding year. He is managing partner of Square Wave Ventures and a former chief executive and executive chairman of Synopsys. Nvidia calls him the first chair the board has had. His line, and hers, are the operating promise: customer commitments and the product roadmap continue, on the foundation of 33 years, still aimed at accelerated computing and AI. The documents name no permanent successor and give no reason. Reuters, The Guardian, and CNBC tied the exit to allegations of extramarital affairs, including relationships with Nvidia employees and one with a woman those reports described as a suspected Chinese spy. The filings do not discuss that reporting. The next hard date is earnings, about 1:20 p.m. Pacific on November 17, with the call at 2 p.m. — Kress’s first as chief executive. On November 13 the board told CNBC it was still choosing a search firm and had “not formally engaged any potential candidates.” The Financial Times had reported an approach to Arm CEO Rene Haas, a former Nvidia executive. Other coverage named Ian Buck, who helped build **CUDA**. The board called individual speculation premature. > Until a search firm is hired, the succession is a clock. ## Three clocks inside one headline **Silicon is already in motion.** Vera Rubin is a cloud product. A data center is a power contract before it is a slide, so the handoff from Blackwell does not happen on the morning of a press release. TrendForce’s November 13 note said the supply chain was not cutting orders, and that Rubin takes Blackwell’s place as the mainstream high-end GPU. **Capital can turn this quarter.** Huang sold the chips and also helped customers pay for them. August filings described up to $105 billion of Nvidia support for an OpenAI data center in Ohio, capacity phased from 2028. CNBC, citing the summer quarterly report, put off-balance-sheet agreements backing AI cloud providers at about $36 billion. Reuters reported that Kress and her staff were the counterweight on deals that large. She now holds the check and the signature, so a guarantee, an investment, or a buyback can move before any new chip is taped out. Smaller AI clouds feel a pullback first. **Monday’s inbox did not get a new calendar.** The model subscription still bills. Waiting for the chip story to settle means waiting on a chair and a search firm that has not been hired. > A rack can be reserved. A morning is due anyway. ## What is locked, and what is still a slide On October 21 Nvidia and Google Cloud described A5X, bare-metal **Vera Rubin NVL72** instances, for general availability the next month. November 6 was that day. The announcement described a path toward as many as 960,000 Rubin GPUs across sites: an ambition with a real first step, with capacity that shows up region by region. NVL72 is Rubin GPUs and Vera CPUs in one liquid-cooled rack. A successor inherits that queue. The editable sequence is the public roadmap he had already presented: a denser **Rubin Ultra** generation in the second half of 2027, then **Feynman** on a 2028 horizon. He was also the metronome for TSMC’s leading process, high-bandwidth memory, CoWoS packaging, NVLink, and the rack builders behind them. A first chair and an interim CEO can keep that convoy moving, and they will start fewer experiments that existed because the founder wanted them on a Tuesday. ## CUDA outlasts the CEO The advantage was one stack in one pair of hands: chip, link, rack, **CUDA**, the developer who already assumes it, the deals closed in person, and the financing around them. CUDA is the piece a press release cannot reassign, which is why near-term results can look intact while the story around the stock gets loud. The opening for everyone else is a budget cycle. Google’s TPUs and Amazon’s Trainium were already the custom hedge, and AMD was already the second source. Single-supplier risk used to be a keynote argument. It is now a governance memo too. Counterpoint Research has forecast that server ASICs can pass GPUs in unit shipments next year. Treat that as a forecast. The evidence you can actually check is a 2027 order book with more than one chip vendor on it. Reuters, citing people familiar with Kress, reported she is likely to deal with Washington more like a conventional technology company than Huang did over export rules. Policy still decides who can buy which chip. Developers still pick the platform they already know how to ship on. > The transistor schedule is published. The underwriting schedule is the live one. ## What this does to AI, if you buy the output OpenAI, Anthropic, and Google keep training, and they reread supply contracts while they do it. [Anthropic’s S-1](https://faraday.email/blog/anthropic-ipo-ant-2026), for a proposed Nasdaq listing under ANT, landed in the same season. A filing and a CEO change in one quarter is an industry growing up in public. Feature lists stay where they were. Inference is where the mix shows up. Next year’s token can run on Nvidia, a TPU, Trainium, or AMD, and the customer sees the answer, not the die. They notice if your product was welded to one stack. The compute already queued still arrives. A six-month pause “until Nvidia is settled” is a tax. [ChatGPT Work and Claude Cowork](https://faraday.email/blog/chatgpt-work-vs-claude-cowork-inbox-2026) still finish a project you assign. [Genie 4](https://faraday.email/blog/google-genie-4-world-model-2026) still tries to keep a scene running. A project agent completes a job. The inbox is the stream that arrives anyway. Most companies lose the hour in the second one. ## Productivity is the layer you still choose Enterprise AI has been a hopeful sequence of purchases: chips, a model, a chatbot seat, then a slide about hours returned. Coding tools earned a number. Much of the rest became another window beside the window where work already lived. That window is email. Microsoft’s 2025 Work Trend Index put a knowledge worker at 117 messages a day — the sourced picture is in [the state of the founder inbox](https://faraday.email/blog/state-of-founder-inbox-2026). A new rack classifies none of them. [Faraday](https://faraday.email/) is the client for that count. It signs into the Gmail or Outlook account you already have, and the address stays. Mail is classified on arrival — a person, a bill, a booking, a newsletter, the noise — so the morning starts smaller. Glance is the short version of the day. Auto-write drafts from the thread, in the way you already write, without a prompt per sender. [Prompt fatigue](https://faraday.email/blog/prompt-fatigue-promptless-email) and this transition are the same argument from opposite ends of the stack: tokens got abundant, and attention did not. A daily surface has to be intelligent on its own. The iPhone app is on the [App Store](https://apps.apple.com/in/app/faraday-tidy-ai-email-client/id6814509563). The Mac app is native and about 10 MB. The web app is the same inbox. In the US the published price is $14 a month, or $12 a month billed annually, after a one-week trial. Confirm the live number on [pricing](https://faraday.email/pricing). Give projects to a computer-use agent and the stream to an [agentic inbox](https://faraday.email/blog/what-is-agentic-inbox-2026). When silicon becomes a committee, that split is still yours. The category page is [email management software](https://faraday.email/blog/email-management-software-2026); the ranking is [the best AI email clients](https://faraday.email/blog/best-ai-email-clients-2026). > Chips set the price of a token. The client decides whether the token did any work. ## What to watch, in order **November 17.** Rubin volumes, margin, and the verbs around investment, guarantees, and capital return. A restated roadmap is the promise already made. A new adjective is the tell. **The search, then the customers.** A firm, then a person. Until the board says otherwise, Haas, Buck, and every other name are rumors. Kress can be the bridge, the candidate, or both. Rubin Ultra and Feynman become real when a cloud publishes a region and a reservation, the way A5X did. In 2027 budgets, new lines for TPUs, Trainium, and AMD are the share shift you can count. If the only AI result you can point to is a subscription, leave the chip headlines and fix the inbox. Shareholders are reading this week for a different line. This page is for the operator. ## Who is Nvidia’s CEO now? Colette Kress. She is interim chief executive and remains CFO, named on November 12, 2026, after the board accepted Jensen Huang’s resignation on November 11. A permanent search is open. ## Why did Jensen Huang resign? Nvidia’s release and 8-K record the resignation and give no reason. Reuters, The Guardian, and CNBC reported allegations of extramarital affairs, including relationships with employees and one with a woman described as a suspected Chinese spy. ## Who is Colette Kress? Colette M. Kress, 59, has been Nvidia’s CFO since September 2013, after nearly 25 years in technology finance, including Cisco and Microsoft. She holds the interim CEO role and the CFO role together. ## Who is Harvey C. Jones? Harvey C. Jones, 73, is Nvidia’s first chair. A director since 1993, he is managing partner of Square Wave Ventures and a former CEO and executive chairman of Synopsys. ## When does Nvidia report earnings? Tuesday, November 17, 2026. Third-quarter fiscal 2027 results at about 1:20 p.m. Pacific Time, call at 2 p.m. The date was set before the transition. ## Does the CEO change affect Vera Rubin? The near-term platform is already shipping. Google Cloud’s A5X instances, on Vera Rubin NVL72, became generally available on November 6, 2026. On the roadmap already presented, Rubin Ultra is aimed at the second half of 2027 and Feynman sits on a 2028 horizon. Those are the parts a new CEO can still reshape. ## What happens to CUDA? CUDA stays. Clusters already built on it do not move because a CEO resigned. The open question is whether 2027 and 2028 budgets add AMD, Google TPU, or AWS Trainium beside it. ## Will custom AI chips gain on Nvidia? They already were. TPU, Trainium, and AMD are the existing hedges, and the transition adds a governance reason to use them. Unit forecasts that put ASICs ahead of GPUs next year are forecasts. Watch the 2027 order book. ## What does the Nvidia transition mean for AI productivity? Chips and financing set the price and supply of tokens. Productivity is which daily surface got intelligent. For most operators that surface is email, and a new GPU does not sort it. ## Where does Faraday fit after the Nvidia news? Faraday is an AI email client for Gmail and Outlook. It classifies mail on arrival, shows the day in Glance, and drafts with Auto-write, without a prompt per sender. Your address stays. General agents do projects. Faraday does the stream. ## Is this investment advice about Nvidia stock? No. Read the 8-K and the November 17 release before any decision about NVDA. This page is about chips, AI products, and the workday.