Reed’s view
In a long, revealing interview with Dwarkesh Patel this week, Microsoft’s Satya Nadella gave the best argument I’ve heard for believing that the AI buildout may be sustainable.
Nadella was asked whether one AI model would win. “If there’s one model that is the only model that’s most broadly deployed in the world and it sees all the data and it does continuous learning, that’s game-set-match and you stop shop,” he responded. That, he said, is wildly unlikely to happen — and he argued that a focus on the models is the wrong way to look at the industry.
To take advantage of AI, developers need a lot of other compute infrastructure built around it — like databases and web services. You might be able to create chatbots without much of this stuff. But the powerful, agentic apps and automations that companies and startups are hoping to create will take a lot more than a GPU.
And in this world that Nadella envisions, there will be many, many different kinds of AI models. That requires building data centers that are versatile — and not tied to one kind of AI model.
Nadella said that’s why he didn’t opt to build all the infrastructure that Microsoft’s partner, OpenAI, required. That sent OpenAI into the arms of Oracle, a competitor — which doesn’t bother Nadella. “We didn’t want to just be a hoster for one company and have just a massive book of business with one customer. That’s not a business,” he said.
Instead of building as much compute capacity as possible, Microsoft has been trying to meet demand for compute as it materializes all over the world. If the company can do this successfully, it can avoid buying a bunch of expensive GPUs that will sit idle while they depreciate in value.
By the end of the interview, you get a sense that this reckless AI buildout that we’re all hearing about is really much more like the same old cloud business — but fast-growing, massive, and transformative.
Some people may end up getting burned if they aren’t deliberate and methodical. And there could be a black swan event, like some insane, overnight leap in capability. Barring those things, the AI buildout is a lot more pragmatic than it seems.
“Ultimately, you’re allowed to build ahead of demand, but you better have a demand plan that doesn’t go completely off-kilter,” Nadella said.
In this article:
Room for Disagreement
In an opinion piece for Fierce Network last month, analyst Stephen M. Saunders wrote that there is too much money going into data centers, with underinvestment in virtually everything else. “These workhorses of the AI revolution need power, water, and a network,” he said, but the US is “nowhere near” delivering on those three elements.
Notable
- In a wide-ranging interview with Semafor earlier this year, Nadella emphasized the need for human leadership qualities as AI progresses: “You can’t just come in and say, ‘I’m smart. I have a bunch of ideas, but I don’t know exactly what to do.’ No, you have to know what to do when it is ambiguous, when it is uncertain. So that bringing clarity is super, super important,” he said.