- The race is on to address the AI chip shortage.
- SoftBank’s Masayoshi Son is the latest tech leader who plans to invest heavily in chip production.
- Chips are needed to train the complex models that underpin AI — but there’s not enough to go around.
Tech leaders at the vanguard of the AI revolution are being forced to confront a difficult reality: they don’t have anywhere near enough computing power to fulfill their ambitions.
The scale of the issue became clear earlier this month following a Wall Street Journal report that said OpenAI boss Sam Altman was seeking up to $7 trillion to fund a venture that could boost the production of microchips, known as GPUs.
Since then, it’s also emerged that SoftBank CEO Masayoshi Son, who met Altman in Tokyo last year, is seeking to raise $100 billion to finance a chip venture codenamed Izanagi, after the Japanese deity of creation, Bloomberg reported.
The incentive to throw so much cash at chips is simple.
Training future versions of large language model GPT or text-to-video model Sora to become ever smarter depends on them. So too are more speculative ambitions being pursued by both Altman and Son, such as the development of artificial general intelligence.
At a technical level, demand for chips is driven by the so-called scaling law, which, as Wharton professor Ethan Mollick puts it, explains why “the more compute you throw at training an AI, the better it is.”
There’s just one problem: the industry’s access to chips is limited.
Nvidia, a lead designer of AI-specialized chips such as the H100 and A100, has struggled to meet demand. Meanwhile, the manufacturing of these designs is done by just three companies — TSMC, Samsung, and Intel — and all are struggling to meet demand.
That’s created a frenzied scramble for supply across the tech sector.
Mark Zuckerberg made it a point last month to highlight Meta’s plans to have stockpiled some 600,000 chips by the end of the year, as the Instagram owner works towards its vision of building “general intelligence” and making it “widely available so everyone can benefit.”
Meanwhile AI startups, which raised $22.4 billion last year per Dealroom data, have been busy fundraising so they can direct as much cash as possible towards securing compute power.
As things stand, however, there simply aren’t enough chips to go around.
How Altman and Son go about addressing this problem remains to be seen. Altman’s multi-trillion dollar plan would need to figure out not just more efficient chip designs for increasingly complex AI workloads, but manufacturing capacity too.
The SoftBank chief, meanwhile, has already benefited from the AI buzz given its majority control of ARM. The value of the chip has more than doubled since it went public in September to about $130 billion amid a surge in demand for AI applications.
But Son’s reported plans signal his intent to invest a whole lot more into segments of the chip sector that ARM is not involved with.
If the world wants better AI, triggering one of the biggest and most expensive hardware buildouts in tech history is probably the only option.
Source link
