The real test for the artificial intelligence build-out is what happens in 2028, analysts say

JPMorgan's Michael Cembalest is looking past the immediate headlines around artificial intelligence and focusing on a specific window: 2028. His concern centers on the pace of data center construction, the physical backbone of the AI build-out, and what happens to that momentum after the U.S. midterm elections.

The logic worth understanding is timing. Data center projects are long-cycle commitments—permitting, power contracts, and construction can stretch across years. That means decisions made now will show up in capacity figures and earnings later. Cembalest's two-year-ahead framing suggests the market may be underestimating how policy shifts, financing conditions, or power constraints could alter the trajectory of that spending once the political calendar turns.

For investors, the takeaway is less about any single forecast and more about what to monitor. Watch data center capex guidance from hyperscalers, power availability and grid interconnection queues, and any post-election changes to incentives or regulation affecting large-scale infrastructure. These are the inputs that will determine whether the build-out proves durable or merely front-loaded. The 2028 question is really a question about whether today's AI investment cycle has staying power beyond the current political and financing environment.