5 hours ago
7
America wants to build the future of computing. It may not have the electricity to plug it in.
That was the warning from Pat Gelsinger, the former Intel CEO who is now a general partner at Playground Global. On an a16z podcast appearance on October 9, 2026, he argued that US energy capacity has been effectively flat for 15 years. In his view, that stagnation is now a hard ceiling on economic growth in an AI-driven economy.
His sharpest line cut straight to the problem facing every company racing to build AI infrastructure.
“Why build a new data center and buy the million GPUs if I can’t power them?”
Fifteen years of running in place
Gelsinger’s core point is about net growth, not gross additions. The US has been adding renewable energy capacity. It has also been retiring coal plants. According to Gelsinger, those two trends have nearly cancelled each other out.
There has been some recent movement. Gelsinger acknowledged that capacity growth has picked up to about 4% annually. He also said that pace falls well short of what AI and data center operators will need.
The timelines for adding new power are the other half of the problem. Gelsinger said gas turbine deliveries now take approximately eight years. He also noted that the last new US nuclear reactor came online around two decades ago.
Defaults on the horizon
Gelsinger did not stop at diagnosing the bottleneck. He said the industry should expect more defaults on data center projects as power needs go unmet on schedule.
He is not alone in seeing a gap between what AI needs and what exists. Executives at companies including Cerebras and Nebius have made similar observations, saying demand for AI compute is running far ahead of available supply.
Oracle’s remarks have also been read as a sign of potential underwriting caution among operators. The concern is about committing to new chip purchases without energy already secured.
Gelsinger has been involved with energy-focused ventures including Alva Energy and PowerLattice, which work on efficiency and new capacity. That gives his warning a practitioner’s perspective, and it is worth keeping in mind that he has backed companies positioned to address the very gap he describes.
What this means for the AI buildout
The immediate pressure lands on hyperscalers and chipmakers that have made large capital expenditure commitments. Those plans assume the power will be there when the hardware arrives. Gelsinger’s comments have raised concerns in the tech community about whether those commitments remain viable if generation and transmission cannot keep pace with project timelines.
Recent policy discussions around permitting reform and grid modernization are closely tied to these capacity concerns. Faster approvals for generation and transmission projects would directly affect how quickly new power can reach data center sites.
The things worth watching are fairly concrete. Track whether permitting reform actually advances, whether turbine lead times shorten, and whether more operators follow Oracle in tying chip purchases to secured power. Any reported data center project defaults would be the clearest test of Gelsinger’s prediction.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.
English (US) ·