Inference Growth Sharpens the Focus on AI Power and Infrastructure Constraints

TrendForce expects the combined AI inference computing capacity of the five largest North American cloud providers to increase by approximately 122% in 2026, according to figures cited in a September 25 Washington Examiner report. The research firm also expects combined capital expenditure by nine major global cloud providers to exceed $886.7 billion in 2026.

The forecasts illustrate the infrastructure consequences of an AI market increasingly oriented toward inference workloads. Scaling those workloads requires not only accelerator supply, but also available power, grid interconnection, land, water and financing. The report notes continuing public and regulatory disagreement in the United States over data-center consumption of electricity, water and land.

Texas provides an indication of the gap that can emerge between requested and deliverable capacity. Regulators were considering more than 700 GW of data-center power requests. That figure represents requests for electricity rather than commissioned or guaranteed contracted data-center capacity, and the report flags the possibility that some of the stated demand may be speculative.

At the hardware layer, the article cites an earlier Data Center Dynamics report that Argentum AI secured an agreement valued at $4.1 billion for approximately 27,000 Nvidia GB300 GPUs. The reported amount describes the GPU agreement, not confirmed Argentum AI capital expenditure or data-center construction cost.

For operators and investors, the trend could shift competition beyond accelerator procurement toward projects with demonstrable demand, a realistic interconnection path and adequate financing. In markets with exceptionally large power-request pipelines, those attributes may carry greater weight than headline queue volumes.