Data Center At the AI Infra Summit, experts said the future of data centers hinges less on AI itself and more on securing electricity and community trust.
Broadband Breakfast
SANTA CLARA, Calif., Sept. 16, 2026 — Data center builders see their political challenge less as a referendum on artificial intelligence than as a race to build the energy, grid and public trust needed for an industry growing faster than its rules.
At the AI Infra Summit, Ed Nelson, strategy director and co-founder of the AI Infra Summit, described an industry that has moved in less than a decade from experimental chips to enormous campuses whose power needs shape permitting and community debates.
“In 2018, nobody really had any conception of where this thing was going to go,” Nelson said. Early AI hardware conversations centered almost entirely on speed, he said, while today power is a defining constraint.
The shift helps explain why data centers have become a political issue in the United States’ midterm elections.
“Back in those days, it was all about performance,” Nelson said. “This year, I think a lot of the conversations have been around the energy itself and how you generate that, the grid and all that sort of stuff.”
Power moves to center stage
The summit itself mirrors that transformation. Launched as the AI Hardware Summit at the Computer History Museum with about 300 attendees, the event has grown to 9,000 people, Nelson said, and expects 12,000 when it moves to the San Jose Convention Center for its 10th anniversary next year.
Its agenda now stretches from energy and physical data center infrastructure through chips, networking, memory, interconnects and enterprise software.
No single layer of digital infrastructure can expand alone, said Nelson and Joe Wells, founder and chief growth officer of information provider Data Center Signals. Faster processors require denser racks, denser racks require more electricity, and new generation requires permits, transmission and local consent.
“Energy is top of mind for everyone,” continued Nelson. The metric once focused on performance figures, he added, but now includes the power required to produce them. “The conversations have just gone way beyond performance.”
That expansion is also changing who builds capacity. Hyperscalers still hold much of the computing infrastructure, Nelson said, but emerging AI cloud companies are constructing facilities quickly and sometimes supplying capacity to the same giants they ostensibly challenge.
“Everybody’s building very, very fast,” Nelson said. The result is a market with new business models but few settled expectations for communities, utilities or regulators.
A playbook built in public
The industry’s political standing may depend on whether it can translate that speed into local benefits. Well added that operators have been slow to respond as opposition hardened around water, electricity and secrecy.
“There’s no playbook right now,” Wells said. Still, he argued that confidentiality need not block accountability: “You can still have NDAs, but you can still have open discussion. You can have both.”
Broadband is another part of the bargain. Moderator Drew Clark, CEO of Broadband Breakfast, highlighted the move from urban and suburban interconnection points to rural data centers.
Large training campuses can sit farther from population centers, but inference facilities, which deliver AI responses, need lower latency and closer connections. Wells said facilities without fiber are “just big empty buildings.”
Akul Saxena, a Broadband Breakfast reporter covering the summit, said the most notable transition was “the change from chatbots to agents and what that requires at a physical infrastructure level.”
Training can occur almost anywhere because it is not time-sensitive, Saxena said, while “inference needs to be low latency.” More distributed facilities, Saxena added, mean the issue “essentially touches every American state at this point.”