Altman and Nadella need more AI power, but they’re not sure how much

How powerful is enough for artificial intelligence? No one knows, not even OpenAI CEO Sam Altman or Microsoft CEO Satya Nadella.

This has put software-first companies like OpenAI and Microsoft in a bind. Much of the technology world has focused on computing as a major barrier to the deployment of AI. As technology companies raced to secure power, the effort to procure GPUs lagged to the point that Microsoft apparently ordered too many chips for the amount of power it had contracted for.

“The supply and demand cycles in this particular case you can’t really predict,” Nadella said on the BG2 radio broadcast. “The biggest problem we face now is not the abundance of computing, but the power and kind of ability to get it [data center] Construction is done quickly enough close to power.

“If you can’t do that, you may actually have a bunch of chips in stock that I can’t deliver. In fact, that’s my problem today. It’s not the supply issue of the chips; it’s the fact that I don’t have warm envelopes to deliver them to,” Nadella added, referring to the commercial real estate term for prefabricated buildings for tenants.

In some ways, we’re seeing what happens when companies accustomed to silicon and software, technologies that are expanding faster than massive power plants, need to step up their energy efforts.

For more than a decade, electricity demand in the United States has remained flat. But over the past five years, demand from data centers has begun to surge, outpacing utilities’ plans for new generation capacity. This has prompted data center developers to add power in so-called behind-the-meter arrangements, where electricity is fed directly into the data center, bypassing the grid.

Altman, who was also on the podcast, believes problems may be brewing: “If a very cheap form of energy soon becomes widely available, a lot of people will suffer greatly from the current contracts they have signed.”

“If we can continue this incredible reduction in the cost of each unit of intelligence — let’s say it was an average of 40 times a certain level per year — you know, that’s a very scary percentage from the standpoint of building infrastructure,” he said.

Altman has invested in nuclear energy, including fission startup Oklo and nuclear fusion company Helion, along with Exowatt, a solar startup that concentrates the sun’s heat and stores it for later use.

However, none of them are ready for large-scale deployment today, and fossil fuel-based technologies such as natural gas power plants take years to build. In addition, orders placed today for new gas turbines will likely not be filled until later this decade.

That’s partly why technology companies are adding solar power so quickly, because of the technology’s cheap cost, emissions-free power, and ability to deploy quickly.

There may be unconscious factors at play as well. Solar PV is in many ways a parallel technology to semiconductors, de-risked and commoditized. Both solar PV and semiconductors are built on silicon substrates, and both are rolled onto production lines as modular components that can be packaged together and linked into parallel arrays that make the completed part stronger than any individual module.

Given the modularity and speed of deployment of solar energy, the pace of construction is much closer to that of a data center.

But both still take a long time to build, and demand can change much more quickly than a data center or solar project can be completed. Altman acknowledged that if AI becomes more efficient or if demand doesn’t grow as expected, some companies may find themselves saddled with idling power plants.

But from his other comments, he doesn’t seem to think that’s likely. Instead, he seems to be a firm believer in Jevons’ paradox, which says that more efficient use of resources will lead to greater utilization and greater aggregate demand.

“If the price of computing per unit of intelligence or whatever — whatever you want to think about it — dropped by a factor of 100 tomorrow, you would see usage go up by more than 100, and there would be a lot of things people would like to do with that computing that don’t make any economic sense at the current cost,” Altman said.