Flex CEO: US Grid Is the Real AI Bottleneck—'Forget All the Frontier Models'

By 2030, roughly 11.8% of all U.S. electricity could flow to data centers, according to a 2025 Berkeley study. Revathi Advaithi, CEO of supply chain giant Flex, says the country's power infrastructure is nowhere near ready for that demand—and the consequences would hit the entire AI economy.

"We have a grid that's just not ready for what's in front of us," Advaithi told executive editorial director Diane Brady onstage at Fortune's annual Most Powerful Women Summit in Santa Barbara, California.

Her framing cuts against the industry's current obsession. Chips and frontier models dominate the AI conversation. Advaithi argues they're the wrong place to look for risk.

What does this change for the AI buildout?

Advaithi's point is about timelines. Building a chip takes about five years, she explained. Building a power grid capable of meeting the country's growing energy needs takes decades. That asymmetry means grid capacity, not model capability, becomes the binding constraint on AI deployment.

"When I think about AI readiness, everyone's focused on technology: whether it's semiconductors or frontier models and things like that. And I'm focused on, 'Hey, if you don't get the energy infrastructure and the grid ready, forget all the frontier models and forget all of the semiconductors'... Because you don't have a country that can take advantage of all of that," she said.

For operators and supply chain planners, the implication is direct: compute deployment schedules now depend on power readiness the U.S. doesn't have. Advaithi estimates America will still face a power gap over the next decade as frontier models keep growing, because deploying that compute requires significant power capacity—and the country is already running a deficit.

Why the grid lags so far behind

Advaithi was blunt about the pace of innovation in power. "Power doesn't seem to have changed since the [Thomas] Edison days, and so it really is going to take a long time to innovate and change what's happening in the power infrastructure," she said.

She also identified the specific friction. There's a "tremendous amount" of approval processes for creating new power technology, and data centers themselves aren't moving with haste or adopting new innovations. That combination—slow regulators plus slow adopters—stretches the timeline further.

"It is no joke when I talk about the fact that energy and power innovation hasn't happened in a long time, and it hasn't happened because there's safety issues, there's regulatory issues," Advaithi explained. "You don't want to mess around with things like power, because it impacts people's lives."

Those constraints aren't arbitrary. Power is regulated tightly because failures are dangerous. But the result is a permitting environment that Advaithi says must accelerate if the U.S. wants to close the gap.

The deficit scenario

Advaithi predicts a continued slump over the next decade unless the country speeds up its standards and approvals for power innovation. If the lumbering process keeps stalling progress, AI's growing appetite for electricity threatens to outpace the infrastructure needed to sustain it.

"Unless we move forward pretty fast with new generation—upgrading our transmission lines which are all old and outdated, really getting to next generation distribution—we are going to be operating in a deficit," she said. "We don't want that deficit to impact what will happen in AI."

The takeaway for anyone betting on AI-driven growth: the model race may be decided in labs, but the deployment race gets decided at the grid. Advaithi's warning suggests that permitting reform, transmission upgrades, and generation buildout deserve as much attention from AI investors as the next chip cycle.