Hyperscalers Are Becoming Power and Network Builders, Ex-Luminar CTO Warns

Microsoft signed a 20-year power agreement that is helping make possible Constellation's restart of Three Mile Island — a deal that, according to a former Luminar executive, signals hyperscalers have stopped waiting for utilities and started building infrastructure themselves.

Jason, cofounder and former chief technology officer of Luminar, makes that argument in a new essay on the AI infrastructure race. He draws a direct parallel to the autonomous vehicle boom he witnessed from inside the industry in 2022, when Luminar's lidar technology was helping enable the next generation of self-driving cars.

At an automotive conference that year, Jason said he was excited to have a front-row seat to one of the most significant technological transformations in the industry's history. An automotive executive pulled him aside afterward.

"Jason," he said, "you don't have a front-row seat. You're part of the show. What you're doing for autonomy, and how you're forcing us to think, is exactly what's needed."

That conversation stuck with him because it captured something larger than autonomous driving, he writes. When an industry is transformed by technology, the companies that built the old infrastructure are not always the ones that define the new one.

What happened in the autonomy race?

The pattern played out clearly in autos, according to Jason. Tech companies threw themselves into lidar, radar, computer vision, and AI software with genuine urgency. Traditional automakers invested enormous sums — tens of billions of dollars collectively — through internal programs, acquisitions, and partnerships.

Ford created Argo AI and later Latitude AI. General Motors built Cruise. Volkswagen invested heavily in Argo AI and Cariad. Toyota, Mercedes-Benz, and Nissan each launched major autonomous initiatives of their own. Despite those investments, many of the highest-profile programs were scaled back, reorganized, or shut down as the complexity of true autonomous driving became clear.

Meanwhile, Waymo and Zoox — born in the technology industry, not the automotive industry — kept advancing because they reimagined the entire system around software, AI, and compute. The result is visible on the roads today: Waymos and Zooxes that don't come from Detroit.

How is the AI buildout different?

Today Jason has shifted his focus from optics for autonomous vehicles to optical networking for AI infrastructure. He sees the same game unfolding again, with different players and far higher stakes.

Hyperscalers — the Googles, Microsofts, Amazons and Metas of the world — are racing to build the data centers that will power the AI economy. Power has emerged as one of the defining constraints. These facilities are extraordinarily power hungry, and the grid isn't keeping up.

So the technology companies aren't waiting. They're investing directly in:

• Power generation
• Private optical networks
• Advanced cooling
• Technologies that connect geographically distributed AI campuses into what increasingly functions as a single AI factory

Elon Musk has pursued natural gas generation for xAI's data center campuses for much the same reason as Microsoft's Three Mile Island deal. When a single AI campus can justify billions of dollars in dedicated energy investment, the economics of infrastructure begin to change, Jason argues, and the companies building AI begin to look a lot like infrastructure companies themselves.

Texas has even moved toward requiring many new large data centers to bring their own generation capacity — a recognition that the grid alone cannot keep pace with AI's growth.

Why aren't utilities keeping up?

None of this is because utilities are failing, Jason writes. They are doing exactly what society has asked them to do: provide reliable, affordable electricity within a highly regulated framework. Permitting, environmental review, transmission planning, and rate-case proceedings all take years. AI infrastructure is being deployed in months.

"The mismatch isn't competence. It's speed," he writes. While utilities work within those constraints, the hyperscalers are increasingly building around them.

Telecommunications is experiencing the same shift. Much of today's carrier infrastructure was designed for an internet built around consumers, enterprises, and mobile devices. AI changes the traffic patterns entirely. Rather than wait for carriers to evolve, hyperscalers are building private optical networks between their own campuses and purchasing networking equipment directly from manufacturers.

A new generation of innovators is asking a fundamentally different question, Jason writes: What if the underlying architecture itself needs to change?

What comes next?

Jason doesn't argue for utilities or telecom providers to step aside — quite the opposite, he says their role will become even more important. But institutions built around long investment cycles, regulated returns, and legacy infrastructure cannot be expected to move at the pace of AI.

Having helped build technology through two major infrastructure transitions, he draws one lesson: technology almost always advances faster than the institutions surrounding it. Eventually those institutions adapt — but rarely before innovators have forced them to rethink what's possible.

"The show has already started," Jason concludes. "The question isn't whether that infrastructure will change. It's who will build it."