How powerful is enough for AI? Nobody knows, not even OpenAI CEO Sam Altman or Microsoft CEO Satya Nadella.
That leaves software-first companies like OpenAI and Microsoft in a bind. Much of the technology industry has focused on computing as a major barrier to AI adoption. And while tech companies are scrambling to secure power, their efforts have lagged behind GPU purchases, with Microsoft apparently ordering too many chips for the amount of power it contracted for.
“The cycle of supply and demand in this particular case is really unpredictable,” Nadella said on the BG2 podcast. “The biggest problem we have right now is not too much computing power; it’s the capacity, the ability to get some kind of data. [data center] Builds complete at near power speeds. ”
“If you can’t do that, you could actually have a lot of chips in your inventory that you can’t connect to. That’s actually my problem today. It’s not a chip supply issue. It’s the fact that you don’t have a warm shell to connect to,” Nadella added, referring to a commercial real estate term referring to buildings that are ready for tenants.
In some ways, we’re seeing what happens when companies used to working with silicon and code, two technologies that can scale and deploy more quickly than large power plants, need to step up their efforts in the energy world.
For more than a decade, U.S. electricity demand has been flat. But over the past five years, demand from data centers has started to increase, outpacing utility companies’ new generation capacity plans. This has led data center developers to add power directly to the data center without going through the grid, so-called behind-the-meter methods.
Altman, who also appeared on the podcast, believes there may be a problem. “If a very cheap form of energy comes online at scale soon, it will seriously hurt the existing contracts that many people have signed.”
tech crunch event
san francisco
|
October 13-15, 2026
“If we can continue to reduce this incredible cost per unit of intelligence, let’s say we’re averaging about 40 times a year at a given level, that’s a pretty scary metric from an infrastructure perspective,” he said.
Mr. Altman has invested in nuclear energy, including nuclear fission startup Okro and fusion startup Helion, as well as Exowatt, a solar power startup that concentrates the sun’s heat and stores it for later use.
But none of them are ready for widespread deployment today, and fossil-based technologies like natural gas power plants take years to build. Additionally, today’s orders for new gas turbines likely won’t be fulfilled until later this decade.
This is partly why technology companies have been rapidly adopting solar power, attracted by the technology’s low cost, emission-free power, and ability to deploy quickly.
Subconscious factors may also be at play. Solar power is in many ways a technology similar to semiconductors, a de-risked and commoditized technology. Both solar power and semiconductors are built on silicon substrates and come off the production line as modular components that can be packaged together and combined into parallel arrays, making the finished parts more powerful than individual modules.
Due to the modular nature and speed of deployment of solar power, the pace of construction will be much closer to that of data centers.
But both still take time to build, and demand can change much faster than either 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 could find themselves with power plants shut down.
However, based on his other comments, he seems to think that is unlikely. Rather, he seems to strongly believe in the Jevons paradox. The idea is that using resources more efficiently increases utilization and increases overall demand.
“If the price of computing per unit of intelligence or whatever you want to do were to drop by a factor of 100 tomorrow, usage would increase by a factor of well over 100, and there would be a lot of things that people want to do with that computing that don’t make economic sense at today’s costs,” Altman said.
Source link
