Bloomberg interviewed Tsu-Jae King Liu, president of the National Academy of Engineering and former UC Berkeley engineering dean, on the infrastructure demands of AI scaling. Drawing on decades in chip design and board-level experience at Intel, Liu outlined how energy procurement and grid capacity now rival semiconductor supply as the limiting factor in AI deployment.
The AI race has shifted from a chip shortage story to an infrastructure story. If power supply and transmission become the choke point, the winners aren't just Nvidia and model makers. Utilities, grid operators, and whoever can secure megawatts fastest stand to gain. That changes which companies win, which regions boom, and how regulators think about AI investment.
Listen to earnings calls and investor presentations for the real signal. It's not who orders more GPUs. It's who starts talking openly about energy procurement timelines and grid partnerships. By 2026, the bottleneck moves from the fab to the substation.
For two years, Wall Street sold AI as a chip shortage. That framing was always incomplete. Semiconductors still matter, but the harder problem is turning intelligence into an industrial load and then convincing grids, regulators, and local politics to absorb it. You can compress a product launch. You cannot compress a substation.
Filed to the Technology desk · Aug 21