Data-Center Power Demand Meets Contracts and Supply Limits
Companies detailed customer-backed power commitments and construction schedules, alongside warnings that electricity availability could constrain AI expansion.
The power-equipment provider Babcock & Wilcox Enterprises, Inc. (BW) omitted an earlier reference to discussions for up to 2 GW of power for a new hyperscaler customer from its August market update. The company had included those discussions in May; both updates retained a project to provide 1.2 GW for an AI factory in North Dakota. The omission narrowed the opportunities specifically described in August, without explaining the status of those discussions.
Other companies added concrete customer commitments during the survey window. Atlas Energy Solutions Inc. (AESI), which was securing power-generation assets, said in May that it had executed a framework agreement with Caterpillar covering 1.4 GW of incremental generation assets through 2030. In September, Atlas announced equipment purchase agreements backed by separate cost reimbursement agreements with a leading frontier AI lab. The company said those agreements allocated equipment to that customer and facilitated near-term financing, tying procurement to specific data-center projects.
The power supplier DTE Energy Co. (DTE) maintained the same distinction between approved demand and additional opportunities in July and September. Its 1.4 GW Oracle project was approved and included in its plan; an executed 1 GW Google agreement remained “upside to plan.” For Oracle, the company said contracts had received regulatory approval and construction had begun, with demand expected to ramp in 2027/2028. Minimum monthly charges, termination fees, and credit and collateral requirements accompanied that commitment, which the company said protected existing customers.
The digital infrastructure company MARA Holdings, Inc. (MARA) continued to describe available power as a constraint across its May and August updates. In May, the company said it had announced a definitive agreement to acquire Long Ridge, adding land and power for a data-center campus adjoining its existing operations. In August, MARA said AI infrastructure demand continued to accelerate and the supply of “power-ready sites” was failing to keep pace. Its later language sustained the power-availability argument behind the earlier acquisition announcement.
The data-center operator Applied Digital Corp. (APLD) moved from an April presentation describing power availability as a constraint on AI rollout to a July warning about its own dependence on third-party electricity suppliers. Inadequate supply could prevent it from meeting customer obligations or growing, and outages could outlast backup arrangements, the company said. The on-site power supplier Capstone Energy Plus, Inc. (CEPL) paired May language that AI demand was outpacing grid expansion with a June qualification: it “may see a significant increase” in product demand, and slower AI adoption or changes in customer spending priorities could adversely affect its business. Those disclosures preserved uncertainty around both access to electricity and demand for power equipment.
The edge data-center provider Duos Technologies Group, Inc. (DUOT) made its capacity constraint more specific between May and August. In May, the company reported a contract for 4.8 MW of critical compute capacity supporting a hyperscaler GPU cluster and said demand was now measured in megawatts. In August, a company speaker described enterprise customers moving from 20 kW cabinets to 60 kW as they replaced components with AI equipment. The speaker said legacy facilities lacked sufficient cooling and space for generators, extending the company's description of the constraint beyond electricity supply.
The AI infrastructure provider WhiteFiber, Inc. (WYFI) supplied a concrete operating milestone in September: initial billing had commenced at NC-1. The company described the site's 99 gross MW capacity as fully contracted and identified an initial 40 MW of IT load contracted to Nscale. It also specified that total power capacity of at least 99 gross MW was expected by May 2029, giving separate measures for contracted compute load and the site's planned power capacity.
The AI infrastructure company Digi Power X Inc. (DGXX) placed its next delivery milestones closer. In August, the company said construction of its Alabama Tier III AI data center remained on schedule and described a 10-year AI data-center agreement. It expects to deliver the first 15 MW of IT load in December 2026, followed by an additional 25 MW in March 2027.