Products You May Like
The CEO of OpenAI has a plan to revolutionize the semiconductor industry, according to a new report (paywalled) from the Wall Street Journal. The CEO, whose company is behind industry-changing ChatGPT, wants to secure up to $7 trillion in funding for a new global network of foundries and power plants to produce custom silicon used for artificial intelligence. It’s a bold and ambitious plan that would forever change the landscape of the semiconductor industry if it ever comes to fruition, and that’s a big if.
The report states that Altman has already begun talking to foundries, though it’s unclear which ones would be interested in such an offer. According to Tom’s Hardware, TSMC is mentioned in the report, but it doesn’t reference Intel, Samsung, or Global Foundries. Altman is reportedly hoping to build a network of fabs around the world to produce AI chips for OpenAI and other companies. To do that, he is going to need between $5 and $7 trillion, which is an amount that seems difficult to secure, even from dozens of sources.
OpenAI used Nvidia’s H100 accelerator to train ChatGPT, but it might be using its own GPUs in the future.
Credit: Nvidia
The dollar amount also dwarfs current industry spending; as Tom’s Hardware notes, the semiconductor industry is expected to be a trillion-dollar industry, but not until 2030. It’s about half of that amount in its current form, so it’s unclear why Altman thinks that much money will be necessary. He may envision all the associated costs of building multiple fabs, including infrastructure supporting them. However, even fabs constructed from scratch, like Intel’s new fab in Ohio, are estimated to cost just $100 billion over time. Micron has also announced a new memory fab in New York that will cost $100 billion.
It remains to be seen how successful Altman will be in this ambitious venture. It exists right now purely as discussions between Altman and potential investors and chip-makers, reportedly. Still, it’s a novel idea that would take many years to come to fruition. It also seems like a risky investment given the current race by TSMC, Intel, and Samsung to diversify their fab locations to protect against geopolitical instability and future supply chain disruptions. Plus, a global network of fabs producing custom AI silicon sounds related to what Nvidia plans to do, but on a much smaller scale. Since it’s already using Nvidia hardware, it makes us wonder if it would also be involved in Altman’s plans at some point.