Nvidia Without Jensen Huang: What His Resignation Changes Technically, and What It Doesn’t
Jensen Huang has resigned as Nvidia's CEO, with CFO Colette Kress as interim chief. A technical look at what his exit means for CUDA, the Rubin-to-Feynman roadmap, Nvidia's flat org, succession, competition and supply chain, and what to watch next.
Written on November 13, 2026 (Sydney time) from public filings and reporting available that day. The story is still moving: the board’s independent investigation and the congressional inquiries have not concluded. This is not investment advice, and the judgments below are my own.
What actually happened
On November 12 (US time), Nvidia announced that Jensen Huang had resigned as president, CEO and board member, effective immediately. The 8-K filed with the SEC says the board accepted his resignation on November 11. CFO Colette Kress is interim CEO and keeps the CFO job. The board will look at internal and external candidates for the permanent role, and Harvey Jones, a former Synopsys CEO who has sat on the board since 1993, becomes Nvidia’s first-ever board chair.
The trigger was personal conduct, not the business. According to NPR and the BBC, Huang admitted in an email to staff to an affair with a businesswoman whom US intelligence reportedly believes was working for Chinese intelligence. Beijing denies it, and the Justice Department says it has reached no conclusion. The New York Times then reported a second relationship with an executive who reported to him, and the board met overnight. Nvidia says it has “no reason to believe” national security or company secrets were compromised.
The stock opened roughly 13% lower on November 12, touched $148.80, and closed at $166.19, down 4.34%, for a market value of about $4.01 trillion (Motley Fool). AMD rose 5.9% and Broadcom 5.5% the same day. Reuters noted Nvidia now trades at about 12 times expected earnings, its lowest multiple since 2012 (Reuters via MarketScreener). Analysts who cut ratings mostly cited execution and governance risk, not weaker demand (CNBC roundup).
That is the news. The more useful question is what a founder’s exit does to a chip company’s engineering machine.
Year one: the pipeline is already set
Chips are slow. A data center GPU typically takes three to four years from architecture definition to volume production, so whatever Nvidia ships over the next year or two was specified, taped out and validated long ago.
The roadmap Nvidia laid out at GTC 2026 (DCD, The Next Platform) looks like this:
- Second half of 2026: Vera Rubin in volume, alongside the NVLink 6 switch, ConnectX-9 SuperNIC, BlueField-4 DPU, Spectrum-6 Ethernet with co-packaged optics, and the Groq-derived inference chips.
- 2027: Rubin Ultra, with four compute chiplets and 1 TB of HBM4E per package, in the Kyber NVL144 rack.
- 2028: the Feynman GPU with die stacking and custom HBM, the Rosa CPU, the LP40, and the first NVLink switches built on co-packaged optics.
Rubin is scheduled to ramp this half, Rubin Ultra is probably close to design-complete, and Feynman has been in development for a while. Those programs are run by architecture, verification, packaging and software teams that do not stop because the corner office changes hands. Huang said at GTC he expected Blackwell and Rubin purchase orders to reach $1 trillion through 2027 (CNBC). Those orders are for products that already exist, and no rival can swap in a full rack-scale alternative at that volume on short notice.
The same goes for CUDA. Its value lives in nearly two decades of libraries (cuDNN, cuBLAS, NCCL, TensorRT), compilers, drivers and developer habit, maintained by thousands of engineers. Even a cautious Korean broker, Daol Securities, wrote that it is hard to read the resignation as a rollback of CUDA or the wider system (ChosunBiz). My read: over the next 12 to 18 months, the direct hit to what Nvidia can deliver is small.
Years two to five: tempo and integration
Huang’s fingerprints are rarely on a specific circuit. They show up in two places.
The first is tempo. Moving from a two-year to a one-year cadence was his call, and it is brutally hard. Every year the GPU, CPU, NVLink, NICs, switches, rack design and software have to turn over together, and if any one slips the whole system waits. DIGITIMES put it plainly: the roadmap is set through 2028, but the tempo and the talent bench were built around him.
The second is integration. Nvidia no longer sells a chip; it sells racks and, increasingly, whole data centers. Deciding what comes after Feynman (how far to push optics, how to split work between GPUs and LPUs, how deep to go on its own CPUs) means someone has to arbitrate between the GPU, networking, software and systems groups and place large bets early. That is exactly where Daol says the uncertainty starts.
The org chart makes this harder. Huang ran the company with around 60 direct reports and no routine one-on-ones, pushing information through large group meetings (Fortune). That kept decisions fast. It also meant many cross-team calls ended at one desk, and Nvidia never built a formal number two. NPR quoted analysts calling succession his blind spot. A new CEO either keeps 60 reporting lines, which few people can manage, or adds a management layer. Either way, decisions probably slow down for a while.
Who runs the technology now
Kress knows the business and investors trust her, but she is a finance executive, not an engineer. The most discussed permanent candidate is Ian Buck, who joined in 2004, led the early development of CUDA, and now runs the hyperscale and HPC business (Reuters via Investing.com). Tae Kim, author of The Nvidia Way, argues the board should choose between Buck and Arm CEO Rene Haas, and that keeping Jonah Alben, senior vice president of GPU engineering, matters just as much (Investing.com).
I agree that retention is the more urgent problem. Alben owns the GPU architecture, chief scientist Bill Dally runs research, and much of the networking group came from Mellanox. Many of these people hold vested stock worth a lot of money. If the search drags on, or engineering priorities start losing arguments to financial ones, losing a handful of them to AMD, a hyperscaler or a well-funded startup would do more damage than an empty CEO chair.
A window for competitors, but a narrow one
Rivals will push. AMD has a 6-gigawatt deal with OpenAI, with the first gigawatt of MI450 deploying in the second half of 2026 (AMD). Google is selling TPUs outside its own cloud (RCR Tech), and Amazon’s Trainium, Meta’s MTIA and Broadcom’s custom ASIC business keep growing.
Still, a CEO change does not rewrite anyone’s training stack. Moving from CUDA to ROCm or XLA means rewriting kernels, retuning collectives and redoing performance validation. Big customers were already going multi-vendor; this probably speeds that up rather than setting off a stampede. What I would worry about more is the deal table. Huang personally carried relationships with the heads of the cloud providers and model labs, and that kind of trust does not transfer with a title.
Export controls and supply chain
The allegations involve Chinese intelligence, members of Congress have asked for a briefing, and Reuters reports fresh questions about how Washington oversees chip sales to China. Expect Nvidia to be more conservative on China-specific products and licence applications, and expect more regulatory scrutiny regardless.
On supply, TSMC’s leading-edge capacity and CoWoS packaging, and HBM supply from SK hynix and others, rest on multi-year contracts and teams that manage them daily. Those will not break over a CEO change. The harder part is the next round of joint bets, such as custom HBM and co-packaged optics, where suppliers want a clear, long-term commitment from the top.
What to watch
- Q3 fiscal 2027 results on November 17, 1:20 p.m. Pacific (8:20 a.m. November 18 in Sydney). Listen for any change in how management talks about Rubin ramp timing and next-quarter guidance.
- Whether the permanent CEO is an engineer or an operator, and internal or external.
- Whether Alben, Buck and Dally stay.
- Whether the roadmap shown at GTC 2027 still runs on a one-year cadence.
- The findings of the board investigation and congressional inquiries, and any knock-on effect on export licences.
My take
On engineering alone, Nvidia is very unlikely to miss product deliveries over the next year and a half because of this. The real risk sits after 2028. Without one person who can make the final call across GPUs, networking, software and systems, can Nvidia keep the annual cadence and keep picking the next computing shift early? Nobody can answer that yet. It depends on who the board picks and whether the core engineering team stays.
FAQ
Will Huang’s resignation delay Rubin?
Nothing public suggests so. Vera Rubin is scheduled for volume in the second half of 2026, Nvidia has announced no change, and board chair Harvey Jones said he expects operations, customer commitments and the product roadmap to continue without interruption.
Is CUDA at risk?
Not in the short term. CUDA is a large engineering system maintained by big teams. Over the long term, it depends on whether new leadership keeps software as a top priority.
Who could be the next Nvidia CEO?
The board says it will consider internal and external candidates. Ian Buck, Colette Kress and Arm’s Rene Haas are the names analysts mention most, but for now that is speculation.
