Blog
Jensen Huang leaves NvidiaNvidia CEO AI industry impactOpenAI Nvidia dependence

Jensen Huang Leaving Nvidia: What It Means for AI Compute

What a Jensen Huang exit from Nvidia could mean for AI compute: OpenAI's heavy Nvidia reliance, Anthropic's mix of chips, and what to watch.

Portrait of Luis Chavez-Mattos
Edited by Luis Chavez-Mattos, Director, Product··RSS

For Executives, Procurement and Engineering

Minimalist aerial photograph looking straight down at a woman sorting processor chips directly onto a giant data center server blade on an enormous empty flat ground, casting long hard shadows in bright sunlight across a single thin painted line.

How could Jensen Huang leaving Nvidia affect AI labs' compute supply?

In the short term, probably very little. Nvidia's chips for the next few years are already designed, ordered and in production, and AI labs have signed long commitments. The bigger effects would come later and through confidence: investor sentiment, Nvidia's spending plans, and how labs think about relying on one supplier. Labs that run almost entirely on Nvidia, like OpenAI, carry more exposure than labs that spread across several chip makers, like Anthropic.

A caution first. The source for this article is commentary from a video creator describing a leadership change at Nvidia, citing an SEC filing. I could not verify the details independently, and the commentary mixes in personal allegations I will not repeat. Treat the specifics as unconfirmed and check Nvidia's own filings and major news outlets. The more useful question holds either way: what happens to AI if the person at the center of its main chip supplier changes?

TL;DR

  • Nvidia is the main chip supplier for most AI labs, and its GPUs are the default hardware in most AI data centers.
  • OpenAI is described as almost entirely on Nvidia, while Anthropic is described as around a third, with the rest split across Google TPUs, Amazon chips and AMD.
  • Short-term supply is unlikely to change, because chips take years to design and build and orders are already placed.
  • The real risk is confidence. If investors lose faith, Nvidia may spend less on expansion, and compute availability is a main limit on how fast models improve.
  • Expect more labs to design their own chips and to avoid depending on any single supplier or any single famous CEO.
  • If you use AI tools at work, the practical move is to avoid locking your workflows to one model vendor.

Why does one CEO matter so much to the chip market?

Nvidia started in 1993 making graphics chips for gaming. Its GPUs (graphics processing units) turned out to be very good at the math behind training and running AI models. Today they are the default hardware in most AI data centers.

Huang co-founded the company and has been its public face for over three decades. Three things tie the company's value to him personally:

  • Vision and fundraising. He has been the voice that persuades customers, investors and governments to keep betting on AI infrastructure.
  • Supply chain relationships. Nvidia designs chips but does not manufacture them. Taiwan Semiconductor Manufacturing Company (TSMC) makes the vast majority of the world's most advanced chips, and Huang, who was born in Taiwan, has long-standing ties there.
  • Market reputation. When a company and its founder are seen as one thing, news about the person moves the stock and the mood around the company.

None of this means Nvidia cannot function without him. It means the market has priced in his judgment and relationships, and a transition invites doubt about an interim leader until that person proves themselves.

How dependent is each AI lab on Nvidia?

The commentary draws a sharp contrast between the two best-known labs.

LabReported Nvidia relianceOther compute
OpenAIAlmost 100%Little reported
AnthropicAround one thirdGoogle TPUs, Amazon chips, AMD, and smaller deals

The video's creator cites Anthropic deals of roughly $250 billion tied to Nvidia-powered chips, roughly $250 billion with Google TPUs, and around $100 to $110 billion with Amazon. I could not confirm those dollar figures, so read them as rough claims rather than audited numbers. The pattern is consistent with public reporting that Anthropic trains and runs Claude on Google's TPUs and Amazon's Trainium chips as well as Nvidia GPUs.

OpenAI's concentration was described a year earlier as "everything OpenAI does runs on Nvidia," a line attributed to Huang himself in the commentary. If that is still broadly true, any disruption to Nvidia's supply, pricing or goodwill hits OpenAI harder than it hits a lab with several suppliers.

What is a TPU, and why does it matter here?

A TPU (tensor processing unit) is a chip Google designed in-house specifically for AI workloads. Amazon makes its own AI chips too, and AMD sells GPUs that compete directly with Nvidia's.

For a lab, each option offers a different trade:

  • Nvidia GPUs are the most widely supported. Most AI software is written with them in mind.
  • Google TPUs and Amazon chips can be cheaper or easier to get at scale, but they require adapting software.
  • AMD gives a second source of standard GPUs, which helps with price negotiation and supply risk.

Spreading across all of them takes engineering work. It also means that if one supplier stumbles, the lab can shift load to another. That is the advantage the commentary attributes to Anthropic.

Will AI labs start building their own chips?

The commentary predicts yes, and that is the most plausible long-term shift. Building your own chips is called vertical integration: owning more steps of the process from design to deployment. Google, Amazon and others already design custom AI chips. The commentary says Anthropic has been assembling a custom silicon team in recent months, and that it plans to use unreleased, more capable versions of Claude to help design chips. I could not verify that claim, so treat it as reported rather than confirmed.

If Nvidia's standing weakens, expect OpenAI to feel more pressure to follow, because it has the most to lose from a single-supplier setup. The creator is explicit that this is speculation.

Chip design is slow and expensive. A new chip takes years to move from plan to production, and it still depends on a manufacturer like TSMC. So custom chips reduce dependence on Nvidia, but not on Taiwan's manufacturing base.

Could Nvidia stop being the default choice?

Probably not soon, and the commentary admits as much: demand remains huge and these things take years. But the argument has a logical shape worth understanding:

  1. A leadership shock lowers market confidence in Nvidia.
  2. A lower valuation and weaker sentiment can make the company more cautious about capital spending (money put into new factories, capacity and supply deals).
  3. Compute availability is one of the main limits on how capable the next generation of models can be.
  4. Labs respond by spreading their bets, which gradually reduces Nvidia's share even if its chips stay excellent.

Each step is plausible, but none is certain. Nvidia's products and order book matter more to near-term supply than who holds the title of CEO.

What about the risk of a famous-founder CEO?

The commentary makes a point beyond chips: when a company's reputation sits in one person, trouble for that person becomes trouble for the company. It predicts more companies will spread public attention across a leadership group, and points to Anthropic's CEO appearing alongside other executives as an example.

You do not need to accept the prediction to take the lesson. Any business that depends on one supplier, one tool or one person carries concentrated risk.

What does this mean for you?

Most people reading this do not buy GPUs. Here is how the story reaches ordinary work.

If you use AI tools daily: Prices, speed limits and new features depend partly on compute supply. A tighter market can mean usage caps, higher prices or slower rollouts of new models. A looser one can mean the reverse. You cannot control that, but you can avoid being stuck.

If you manage a team or budget:

  • Ask which AI vendors your tools depend on. Many products are built on one model provider.
  • Prefer tools that let you switch models without rebuilding your workflow.
  • Keep your prompts, templates and data in formats you own, so a vendor change is an inconvenience, not a crisis.

If you invest or follow tech news:

  • Separate confirmed facts (filings, company statements) from commentary and rumor.
  • Watch for concrete signals: changes in announced data center spending, new chip deals by major labs, and any statements about manufacturing partnerships in Taiwan.

If you work in procurement, IT or strategy: Treat a single-supplier AI stack the way you would treat a single-source part in manufacturing. It is fine until it is not.

Can I track AI vendor dependence without another subscription?

If your team wants a simple view of which AI models and vendors each of your workflows relies on, you can describe that tracker in plain English to Remy and get a working internal app: a table of workflows, the model each uses, a fallback option, an owner and a review date. Remy builds apps with a database and sign-in, and it can reach nearly 400 AI models from about 60 vendors, so a fallback you list can be a real option. If you want a news feed alongside it, the app can pull in current coverage of chip and AI-supply news.

Frequently Asked Questions

5 questions

Has Jensen Huang actually left Nvidia?

The source for this article is a video describing the departure as reported through an SEC filing, and I could not verify it independently. Check Nvidia's investor relations page and major news outlets for the current situation before relying on it.

Will GPU supply for AI labs drop if the CEO changes?

Not immediately. Chips in production and already-signed orders continue regardless of who runs the company. The risk is slower: reduced investor confidence could lead to more cautious spending on future capacity.

Why is OpenAI considered more exposed than Anthropic?

According to the commentary, OpenAI runs almost entirely on Nvidia, while Anthropic gets roughly a third of its compute from Nvidia-based systems and the rest from Google TPUs, Amazon chips, AMD and others. More suppliers means less damage if one runs into trouble.

Does building custom chips remove the dependence on Nvidia?

It reduces it, but custom chips still need a manufacturer, and most advanced chips are made by TSMC in Taiwan. Custom silicon also takes years to develop, so any shift would be gradual.

Should I change which AI tools I use because of this?

Not because of this alone. The sensible step is to avoid lock-in: keep your prompts and data portable and know which vendor each important workflow depends on.

How we source, write, and correct our articles. Editorial standards

Related articles

Straight to your inbox.
New pieces as they publish. No noise.