America
Jensen Huang exit aids Anthropic and OpenAI as open models suffer
Jensen Huang’s resignation from Nvidia has cast a shadow over companies developing “open” artificial intelligence models and shifted the balance of power toward “closed” frontier laboratories such as Anthropic and OpenAI.
According to a report in the Financial Times, the former chief executive of the US chipmaker had directed Nvidia’s resources toward building a broad AI ecosystem.
As part of this effort, he invested in and acquired startups involved in developing open-source and open-weight models that users can customise and run on their own hardware.
Huang was a fierce advocate for the spread of these models. While supporting the development of alternatives in the US, he lobbied against moves by Washington to restrict the use of Chinese-made open models.
This summer, in his first post on the X platform, Huang wrote: “Our AI leadership will not be measured by a single frontier AI model, but by whether the US builds a robust and open ecosystem that spans every industry.”
His ignominious departure from Nvidia has stripped the open-model sector of its most vocal and influential backer in the US.
A venture capitalist who had invested in AI startups alongside Nvidia said:
“I think Jensen was the biggest economic beneficiary and personally committed to diversity at the model layer. Without him, and with a more stingy Nvidia on investment, we could end up in a world where model providers are far more concentrated.”
It is unclear whether Nvidia will sustain its rapid pace of dealmaking under new leadership.
The company was expected to participate in or lead four startup transactions currently under negotiation: investments in Thinking Machines Lab and Groq, a potential partnership with Reflection AI, and a $12.9 billion acquisition of open-model repository Hugging Face.
In a post on X on Saturday, Elon Musk said: “Jensen masterfully managed the supply chain at Nvidia, but many of these deals were a handshake with Jensen, not a written contract. Without Jensen, at least some of these critical deals will be lost.”
Huang framed these investments as a way to cement US dominance in the technology.
Others point to more direct commercial motives: these deals helped broaden Nvidia’s customer roster and served as a buffer against OpenAI and Anthropic.
A senior figure at a leading Silicon Valley AI company said: “In a way, [Huang’s departure from Nvidia] is bad for the industry, but in another way, Jensen is the enemy of frontier labs because he wants models to be commoditised so he can retain the commanding heights of the value chain.”
A decade ago, Huang delivered Nvidia’s first “AI supercomputer” to OpenAI, setting the then-fledgling startup on a trajectory that ultimately led to ChatGPT.
At the time, the laboratory’s leaders included Musk, Sam Altman, and Dario Amodei.
Huang’s relationships with this trio laid the groundwork for the AI boom and contributed to the rise of OpenAI, as well as SpaceX and Anthropic, the companies founded by Musk and Amodei.
Earlier this year, Nvidia invested $30 billion in OpenAI and committed more than $100 billion in resources to support the laboratory’s flagship data centre.
Yet friction also emerged: frontier lab founders grew uneasy over their reliance on Nvidia and have sought to reduce that dependency by forging alternative partnerships and investing in their own chips.
A senior official at an AI laboratory said:
“That is why you see every lab and frontier company investing in their own chip programmes: it is because of Nvidia’s profit margins. It makes sense to expect Jensen’s removal to further concentrate computing power in the hands of frontier labs and hyperscalers. This is good for the labs.”
OpenAI collaborated with Broadcom to develop its custom chip, Jalapeño. OpenAI chief financial officer Sarah Friar has frequently remarked that the company should not be “single-threaded” or dependent on a single chip supplier.
Anthropic diversified its AI infrastructure suppliers earlier than its rival, having secured early support from Google and Amazon in the form of access to their chips.
Even so, while announcing plans to secure a $10 billion investment from Nvidia alongside its initial public offering, Anthropic executives highlighted during their roadshow an ability to switch within days to rival AMD chips, a process that typically takes months.
At an event last month, Huang praised OpenShell, an open-source security platform developed by Nvidia for AI agents, while delivering a sharp critique of the two frontier labs’ secrecy:
“How can you have an opaque agent system, locked inside an opaque company, inside an opaque box? It is clearly unsustainable.”
In the near term, these laboratories could also draw practical benefits from the turmoil at Nvidia.
Next week, Anthropic will conduct one of the largest IPOs of all time, aiming to raise a total of around $100 billion, while OpenAI is in the final stages of a $20 billion fundraising round.
According to industry sources, investors pulling back from the chipmaker may redirect their capital toward these laboratories.
One source said: “I don’t think open-source models will disappear… but losing Jensen, their biggest financial backer and champion, will slow their progress, and I don’t think that hurts Anthropic’s IPO.”
Huang also clashed with the frontier laboratories over his defence of lower-cost competitors, including those from China.
Amodei of Anthropic has been one of Silicon Valley’s staunchest advocates for tighter curbs on China’s access to Nvidia chips, even as Huang lobbied the Trump administration for broader access to the Chinese market.
Earlier this year, Amodei remarked that exporting chips to China was “a bit like selling nuclear weapons to North Korea.”
Nvidia lost billions of dollars in sales to China as the Biden and Trump administrations imposed export controls on its most powerful processors.
Nevertheless, analysts note that the company collaborated with Chinese AI firms as part of its wider strategy to champion open AI systems.
Some figures in and around Nvidia insist that support for open systems will endure despite Huang’s departure.
Thomas Wolf, co-founder of Hugging Face, the open-source AI platform Nvidia agreed to acquire in September, said: “If open models had only one champion, it would all be over this week. These models have millions of developers, including those inside Nvidia.”