Hyperscaler Capex Near $700 Billion Is Crowding the Rest of the Investment Cycle

Hyperscaler capital expenditure is on track for about $700 billion this year, a jump of roughly 60 percent from 2025. That figure now sits at the center of the U.S. business-investment outlook. Bonus depreciation and a friendlier tax treatment of equipment help the broader industrial base. They do not create extra transformers, extra high-voltage interconnects, or extra advanced-packaging lines. Those are the binding constraints.

When a handful of buyers can absorb an entire year of incremental supply, everyone else becomes a residual claimant. That is already visible in lead times for switchgear, liquid cooling, and specialized contractors. Regional utilities are being asked to underwrite loads that look like new cities. Some will. Some will queue the request behind other large offtakers.

A practical allocation strategy

Companies that need AI capacity should stop thinking in terms of “our own cluster someday.” Most will rent. The strategy question is reserved versus on-demand, multi-cloud versus single-cloud, and training versus inference. Inference closer to the customer often has a clearer ROI than another foundation-model experiment. Buy that first.

Industrial firms should treat the same crowding as an opportunity if they make power equipment, thermal systems, or construction services — and as a threat if they need those inputs for their own plants. A factory delayed for want of a substation is a strategy failure, not a utility footnote. Lock interconnection studies early. Pay for design work before groundbreaking PR.

Investors watching only model announcements will miss the real bottleneck story. The next leg of returns may sit with the unfashionable suppliers of electricity, cooling, and installation labor. Boards should ask a simple question in every capex review: if the hyperscalers take the next increment of this input, do we still have a plan?