Not enough AI chips? Make them yourself

OpenAI CEO Sam Altman has said numerous times that there aren’t enough computer chips out there to match his company’s goals, and it looks like he might be taking matters into his own hands.

Driving the news: Altman has reportedly restarted efforts to rally investors for a new chip-making venture. Bloomberg previously reported that Altman was seeking between US$8 billion and $10 billion for something chip-related last year, but new details suggest he wants to use that money to build a network of fabrication plants.

  • This would be a break from what Google, Amazon, and Microsoft have done: design chips themselves and leave manufacturing to the likes of Intel, TSMC, and Samsung.
     
  • It’s unclear whether Altman’s factories would make chips designed in-house or make their capacity available to companies OpenAI currently buys chips from.

Why it matters: Chips are where the money’s at in AI right now, but demand has made them both more expensive and hard to obtain. If Altman gets the sizeable time and cash commitment that might scare others from starting a new manufacturing operation, it could prevent manufacturing bottlenecks from slowing AI growth (and maybe make chips cheaper).

  • Demand isn’t going to ease any time soon. Meta plans to have 350,000 Nvidia H100 chips by the end of 2024 for its AI ambitions. Estimates of what Nvidia charges put the pricetag for that between US$8 billion and $14 billion.

Catch-up: Big tech has largely relied on Nvidia for its AI chips, which has been a boon for the company. Its full-year financial results won’t be out until next month, but revenue was up a staggering 206% year-over-year in Q3. Its stock price has also risen by 221% in the last year, pushing its market value above US$1 trillion for the first time.