Microsoft CEO Satya Nadella and OpenAI CEO Sam Altman have identified power availability and data center construction speed as critical impediments to artificial intelligence (AI) expansion, according to statements made on the BG2 podcast. Nadella specifically indicated that Microsoft currently possesses a significant inventory of AI chips that cannot be deployed due to insufficient "warm shells" and power infrastructure.
"The biggest issue we are now having is not a compute glut, but it’s a power and it’s sort of the ability to get the [data center] builds done fast enough close to power," Nadella stated. He elaborated that this has led to a situation where "a bunch of chips sitting in inventory that I can’t plug in," highlighting that the constraint is no longer chip supply but rather the physical and electrical infrastructure needed for deployment.
The remarks underscore a broader trend in the United States, where electricity demand from data centers has significantly increased over the last five years, diverging from a decade of flat demand. This surge has outpaced the new generating capacity planned by utilities, prompting data center developers to explore "behind-the-meter" power arrangements that bypass the traditional grid.
Altman echoed concerns regarding the pace of infrastructure development against the rapid evolution of AI. He observed that if the cost per unit of intelligence continues its current rate of reduction, which he cited as an average of "40x for a given level per year," the "infrastructure buildout standpoint" presents a "very scary exponent." Altman also suggested that a rapid emergence of "very cheap form of energy... at mass scale" could financially impact companies with existing power contracts.
Despite these potential challenges, Altman reportedly subscribes to Jevons paradox, believing that increased efficiency in AI compute will drive even greater demand, rather than reduce overall energy consumption. His investments in advanced energy technologies, including nuclear fission startup Oklo, fusion startup Helion, and solar startup Exowatt, reflect a strategy to address future energy needs. However, these technologies are not yet ready for widespread deployment, and traditional power plant construction faces multi-year timelines.
Tech companies have increasingly adopted solar power solutions, which offer lower costs, zero emissions, and faster deployment compared to conventional large-scale power plants. This modular approach aligns more closely with the construction speed of data centers, though both still require time to build, often less rapidly than demand shifts.
