OpenAI CEO Sam Altman is seeking to safeguard the future supply of advanced AI chips by bringing the manufacture of chips in-house, the FT reports.
The newspaper says that Altman is in talks “with key Middle Eastern investors” and the Taiwanese chip giant TSMC, as he looks to launch a venture to design and build semiconductors designed to drive AI processes.
Altman’s move, says the FT, reflects the growing importance of custom hardware in the rapidly expanding field of AI.
It is said discussions “remain in the early stages”, and the complete list of partners and backers is as yet unconfirmed.
Altman's chips initiative is reportedly driven by concern over the long-term shortage of AI-suitable chips, as the projected demand for AI chips is predicted to outstrip their production as demand for AI-driven solutions skyrockets. It is said Altman wants secure AI chip supplies in place by 2030.
Altman has in the past complained about the scarcity of AI chips, a market dominated by Nvidia, which controls more than 80% of the global market for the chips best suited to run AI applications.
OpenAI tech built on 10,000 Nvidia GPU chips
Since 2020, OpenAI has developed its GenAI technologies on a supercomputer constructed by Microsoft, using an estimated 10,000 Nvidia graphics processing units (GPUs).
Running ChatGPT is inordinately expensive for OpenAI. It is said that each query costs roughly 4 cents, meaning that if ChatGPT queries grow to a tenth of the scale of Google search, it would require roughly $16bn of chips a year to keep operational.
As well as Nvidia’s dominance of the AI chips market, supplies of AI-capable chips are also subject to disruption because of increasing levels of geopolitical instability.
The supply of powerful microprocessors destined for use in AI applications will grow in geopolitical strategic importance, says Glenn Steinberg, EY Global Supply Chain and Operations Leader.
If OpenAI’s chip manufacturing plans play out as reported it will mark a departure from its industry peers, including Amazon, Google and Microsoft, who design their own silicon but outsource its production, reports Forbes.
Building chip fabrication plants will also require substantial investment, with a single plant costing in the region of tens of billions of dollars.
Sheikh Tahnoon, is said to be a key figure in the negotiations. He oversees substantial investment funds and entities in Abu Dhabi and is involved with AI technology group G42, which collaborates both with Microsoft and OpenAI.
The Forbes report concludes that whatever shape Altman’s chips manufacturing strategy takes it will “align OpenAI’s hardware needs closely with its ambitious AI development roadmap”.