OpenAI Has Talked to Broadcom About Developing New AI Chip
The Takeaway
• OpenAI has hired former Google chip staffers
• ChatGPT maker is talking to Broadcom about working on chip
• Sam Altman has had talks with TSMC
Last year, as the world’s top artificial intelligence developers were racing to speed up their work using ever-bigger clusters of computers, OpenAI CEO Sam Altman was trying to play a longer game. He decided to start a new company that could develop and produce a new AI chip and help set up chip factories to make them and data centers to house them.
His plan has taken numerous twists and turns since then. But there are signs his effort to build a new chip is taking shape, starting within OpenAI rather than via a separate venture.
The ChatGPT maker has been hiring former members of a Google unit that produces Google’s AI chip, the tensor processing unit, and has sought to develop an AI server chip, according to three people who have been involved in the conversations. OpenAI has been talking to chip designers including Broadcom about working on the chip, according to the people.
The team has discussed how the eventual chip could help the new venture Altman has envisioned, which aims to increase the amount of computing power for AI developers such as OpenAI, said one of the people.
A new server chip that would rival the kind made by Nvidia is a long shot that would take years to come to fruition. And in trying to develop a chip, OpenAI risks upsetting Nvidia, OpenAI’s most important chip supplier.
But it could also provide OpenAI with potential leverage in future pricing negotiations with the company. Nvidia has been generating unprecedented profit margins and sales on its AI-focused graphics processing units because customers such as OpenAI don’t have viable alternatives.
Altman’s efforts reflect OpenAI’s concerns about competition from cash-rich tech companies such as Google and Meta Platforms, which are investing heavily in data centers and specialized servers to develop the best AI. Altman has long said that to beat them and achieve its dream of “superintelligence,” OpenAI will need many times the computing capacity it has today. That type of AI could, say, help humans colonize Mars or develop fusion energy, and it’s a long way from what OpenAI’s conversational AI can currently do.
Developing a new type of chip has only been one part of Altman’s plan. In conversations that started in 2023 and continued this year, Altman spoke to executives at key chip manufacturers and suppliers, such as Taiwan Semiconductor Manufacturing Co., which makes Nvidia’s chips, about whether they could boost their capacity to make more Nvidia chips or Altman’s proposed new AI chip, said a senior TSMC manager.
Altman discussed with TSMC and others whether a possible new venture of his could finance dozens of new chip factories that such manufacturers would operate. Altman has also gotten to know Intel CEO Pat Gelsinger, whose firm runs a chip foundry that aims to compete with TSMC.
But Altman’s factory expansion ambitions struck numerous executives as improbable, because it would require a lot of capital and specialized labor.
Altman was “too aggressive for me to believe,” TSMC CEO CC Wei said in a news conference last month when he was asked about the prospect of new factories.
Substantial Orders
Still, behind the scenes, TSMC executives have told Altman that they were open to expanding chip production if he or OpenAI could commit to a substantial amount of orders for the new chips, according to a senior TSMC manager who was briefed about the conversations.
That type of arrangement would be more manageable for TSMC because it could use existing factories in Taiwan rather than trying to set up dozens of new ones.
Another key part of Altman’s plan involves building new data centers to house the chips. Recently, he told one industry executive that he aimed to set up one or more companies with outside investors to pay for real estate, power, data centers and the specialized AI chip servers that would be housed in them. He said OpenAI would commit to renting those servers, this person said.
Altman has sought approval from the U.S. Commerce Department to work on such projects with foreign governments such as in the Middle East, said one person who was formally briefed on the conversation. The agency has been concerned about foreign involvement in critical U.S. sectors such as AI. (Multiple news outlets previously reported on the conversations.)
Altman hasn’t spoken publicly about the specifics of the data center and chips effort, but he has privately told industry executives he would unveil the effort this year. It isn’t clear whether the venture has been incorporated as a company.
As he tried to sway chip manufacturers last year, Altman’s plan became one of many bones of contention between him and OpenAI’s board of directors, which briefly ousted him last fall.
When he first broached the chip idea with the board last year, he initially said it would be independent from OpenAI, according to two people with knowledge of the pitch. Some board members felt his growing array of side projects, including a separate AI device company, would draw his attention away from OpenAI.
Earlier this year, though, he told colleagues that OpenAI would own a stake in the new venture and that OpenAI’s new board of directors has had a chance to review the idea.
“OpenAI is having ongoing conversations with industry and government stakeholders about increasing access to the infrastructure needed to ensure AI’s benefits are widely accessible,” said Liz Bourgeois, a spokesperson for OpenAI. “This includes working in partnership with the premier chip designers, fabricators and the brick and mortar developers of data centers.”
She didn’t comment on the specifics of this article. Broadcom spokespeople didn’t immediately respond to a request for comment.
Funding data center infrastructure is a well-trod path for private equity firms and other investors, and Altman has been forging relationships with numerous firms that might participate. They include SoftBank, which has also discussed a separate Altman venture to develop a consumer AI-powered device; UAE-based sovereign wealth funds; and large private equity firms.
Several governments including the UAE have said they want to develop AI-focused data centers to boost their economy, keep local data protected from outsiders, and increase their standing in the tech industry.
It isn’t clear whether Altman has received commitments from such investors for the plan. The OpenAI spokesperson said the company is working closely with Microsoft, OpenAI’s exclusive cloud server provider, but did not elaborate, and a Microsoft spokesperson did not immediately comment.
The Wall Street Journal earlier this year reported Altman was in talks with investors to raise funds for a chip venture, noting it could require raising up to $7 trillion. But Altman privately has told people that figure represents the sum total of investments that participants in such a venture would need to make, in everything from real estate and power for data centers to the manufacturing of the chips, over some period of years.
A Challenge to Nvidia?
Then there’s the question of whether and how Altman plans to make a new AI chip. OpenAI’s chip-focused team, led by Richard Ho, who also worked on TPUs at Google, will likely choose an American company, such as Broadcom, to help it develop the proposed new chip, said the people who have been involved in the conversations. Broadcom works with Google on the TPUs it makes and has talked to the OpenAI chip team, these people said.
Taiwanese rivals to Broadcom, such as AIchip Technologies, also pitched this type of service to OpenAI, some of these people said.
The OpenAI team doesn’t appear to have started designing the chip, and it wouldn’t be produced until 2026 at the earliest, one of the people said. The team is considering various chip packaging and memory components to maximize the chip’s performance, this person said.
Altman discussed his chip plans with memory chip makers Samsung and SK Hynix earlier this year, according to two people with knowledge of the conversations. Both make high-bandwidth memory chips, which provide a high-speed connection between the processor and the memory, allowing for faster data transfer and processing as the chips make AI computations.
Such memory chips have been in short supply, posing challenges for Nvidia's chips.
If OpenAI or the new venture proceeds with designs for a new AI chip, it might be viewed as an affront to Nvidia. The chip designer has supplied hundreds of thousands of its graphics processing units to Microsoft so that OpenAI could use them to develop ChatGPT and other products.
For Nvidia, however, a customer moving onto its turf would be a familiar sight. For instance, Microsoft, like Google and other large Nvidia customers, also develops its own AI-specific server chip as it tries to lessen reliance on Nvidia chips. There’s no indication OpenAI wants to use the Microsoft chip, though.
Altman took to X in February to lament about not having enough Nvidia GPUs to power his company’s AI development.
Since then, OpenAI has gotten access to more Nvidia-powered servers from Microsoft. And by the middle of 2025, Oracle and Microsoft will provide OpenAI with one of the most powerful clusters of Nvidia servers in the world, which would cost about $2.5 billion per year to rent.
But that’s hardly enough computing capacity for Altman. Already, Microsoft and OpenAI have discussed a future data center that could cost up to $100 billion and was known within OpenAI as “Stargate.” The prospective size of the facility and its server cluster means it would likely require a breakthrough in energy or else face a shortage of power.
It isn’t clear whether the Stargate discussions have become part of Altman’s separate AI data center talks.