OpenAI CEO Sam Altman has been raising billions of dollars in the background from global investors with his adventures into chip manufacturing, and now he's using those billions of dollars to set up a network of factories to manufacture semiconductors.
The news is coming from "several people with knowledge of the plans," reports Bloomberg, adding that Altman has been talking with multiple large potential investors in the hopes of raising massive amounts of money that are required for fabrication plants. They're not cheap, into the tens of billions of dollars... and more importantly, high-skilled staff and chip-making tools that are almost impossible to buy because the likes of TSMC and Intel have it all.
Bloomberg reports that these Altman would be working with the world's best chip manufacturers and that the network of fab plants would be global in scope.
- Read more: Sam Altman was raising billions in funding for codename Tigris, for a new chip to rival NVIDIA
Altman knows there is a future AI GPU shortage coming, as companies like NVIDIA, AMD, and Intel can't get them AI GPUs fast enough. NVIDIA is ramping up production of its AI GPUs with next-gen "Blackwell" B100 AI GPUs coming this year and a beefed-up "Hopper" H200 AI GPU coming soon. AMD just launched its new Instinct MI300X AI GPU, and Intel is ramping up its AI GPU efforts... but it's not enough.
Meta is putting 600,000 x NVIDIA H100 AI GPU-equivalent compute performance into its AGI (artificial general intelligence) efforts, soaking up huge quantities of AI GPU hardware in 2024 and beyond. Every big company is pushing into AI, thus the need for AI GPUs is insatiable right now.
OpenAI knows this and is preparing with a stockpile of billions and billions of dollars and the name on the street to push this kind of idea. Constructing, staffing, and maintaining fabrication plants to manufacture semiconductors isn't cheap, with OpenAI peers in the likes of Amazon, Google, and Microsoft -- the latter is OpenAI's largest investor -- designing their own custom silicon and outsourcing the manufacturing to other companies like TSMC (Taiwan Semiconductor Manufacturing Company).
Altman could build a single next-gen fabrication plant with tens of billions of dollars, but creating a network of these state-of-the-art plants wouldn't be quick... it would be years, probably 5-10 before we see something truly revolutionary that is a large step ahead of something TSMC has in the oven for the years to come (which is some super-advanced technology).
OpenAI could work with Intel, TSMC, or Samsung in the fabrication market, which I can see Intel coming up to the plate given their relationship with Microsoft and Windows. Microsoft is pushing heavy into the AI PC market, while Intel is pushing heavy into the AI processor market with its new Meteor Lake CPUs featuring an NPU (Neural Processing Unit) for AI workloads.