As the artificial intelligence revolution accelerates, a new constraint has emerged — electricity. OpenAI CEO Sam Altman and Microsoft CEO Satya Nadella are voicing growing concern that power availability, not silicon, could be the defining bottleneck for the next phase of AI’s evolution.

In a candid discussion on Microsoft’s BG2 podcast, the two executives revealed that the AI industry’s demand for power is beginning to outstrip its ability to deliver it, highlighting a disconnect between the rapid scaling of data centers and the slower development of energy infrastructure. Nadella even admitted that Microsoft, one of the largest cloud providers in the world, is currently sitting on a surplus of GPUs with nowhere to plug them in.

From Compute Scarcity to Power Scarcity

For years, the AI sector’s focus has been on securing enough high-performance chips — specifically, Nvidia’s GPUs — to train and deploy increasingly complex models. But that bottleneck is now shifting. “The biggest issue we are now having is not a compute glut, but a power one,” Nadella said. “We can’t build data centers fast enough near power sources.”

The issue marks a striking reversal for the U.S. power market. After more than a decade of flat electricity demand, the rise of AI and cloud computing has triggered a massive surge in energy consumption. Data centers, which already account for roughly 4% of total U.S. electricity use, are expected to double that share by the end of the decade, according to estimates from the International Energy Agency (IEA).

This imbalance is forcing tech companies to adopt unconventional approaches. Many are now pursuing “behind-the-meter” energy solutions, where electricity from solar farms or small-scale gas plants is routed directly into data centers — bypassing the traditional grid.

Altman’s Energy Gamble: Betting on Nuclear and Solar

Sam Altman, whose company OpenAI operates some of the world’s most power-intensive AI systems, has been preparing for this challenge. Beyond his leadership at OpenAI, Altman has invested in several next-generation energy startups, including nuclear fusion firm Helion, fission developer Oklo, and solar thermal innovator Exowatt.

Altman argues that without breakthroughs in cheap, scalable energy, AI’s growth could stall. “If a very cheap form of energy comes online soon at mass scale, then a lot of people are going to be extremely burned with existing contracts they’ve signed,” he warned, referencing the enormous upfront costs tech firms are incurring to secure power capacity years in advance.

Yet, the alternatives remain limited. Fossil-fuel-based generation takes years to build, while renewable options like solar and wind, though fast to deploy, are intermittent and dependent on vast land and material resources.

Ironically, solar power may be the most psychologically appealing option for Silicon Valley — a parallel technology to semiconductors. Both are built on silicon substrates, both modular, both capable of scaling quickly. But even solar farms cannot be deployed at the same exponential pace as AI adoption.

An Exponential Challenge Ahead

The tension between AI’s exponential growth and the physical limits of energy infrastructure underscores what Altman calls the “infrastructure paradox.” Every improvement in compute efficiency, he argues, only drives demand higher — a phenomenon known as Jevons Paradox.

“If the cost of compute fell by a factor of 100 tomorrow, usage would rise by much more than that,” Altman said. “There’s no ceiling for how much intelligence the world wants to generate.”

That prediction points to a future where the intersection of energy and intelligence becomes one of the most important strategic frontiers in global technology. For investors, policymakers, and utilities, the question is no longer just how to power cities — but how to power cognition itself.

The next phase of the AI race may not be decided by algorithms or GPUs, but by who can build, buy, or innovate enough power to keep the machines running.