Not Just “Power Shortages,” Delays Will Become The Key Theme For U.S. Data Centers In 2026

As construction of artificial intelligence infrastructure advances into more complex phases, project “delays” are emerging as the industry’s primary concern, supplanting the narrower issue of power shortages. Multiple executives and investors have indicated that postponements will define 2026, creating a divergence that separates world‑class operators from ordinary participants.

The term “delays” now covers a broad spectrum of causes and manifestations. Problems range from utilities failing to meet promised power deliveries and late arrivals of critical equipment to on‑site labor shortages that slow construction schedules. For developers building large‑scale AI facilities, the interplay between delivery speed and scale has become increasingly apparent: lead times for generators, transformers and liquid‑cooling components have lengthened, and liquid‑cooling—required for Nvidia’s most advanced hardware—remains a technically immature area with unresolved challenges.

This pervasive slowdown is producing knock‑on effects in capital markets. Recent examples demonstrate that not all delays have equal consequences; the most damaging cases involve political opposition, financing withdrawals and the collapse of binding contracts. Such developments have not only pressured share prices of the companies involved but also prompted some investors to reassess the economic viability of affected projects.

Industry leaders are attempting to address bottlenecks at their source. Nvidia’s chief executive convened a closed‑door summit at the company’s California headquarters last week to promote behind‑the‑meter power generation, a model in which data centers build on‑site generation capacity to avoid lengthy grid interconnection waits. The industry’s ability to resolve power‑supply and construction delays rapidly will be a decisive factor in determining which firms prevail in the AI infrastructure race.

Political resistance and financing constraints have already delayed several projects. Local opposition can be particularly acute when data center proposals scale from hundreds of megawatts to gigawatt levels. In Saline Township, Michigan, residents voted last September against rezoning farmland for a large AI data center, prompting developer Related Digital to sue the township; officials reversed course weeks ago. The project, planned by Related Digital and Oracle for OpenAI, is designed to total 1.4 gigawatts, and Michigan’s energy regulator approved an expedited power request for the site this week.

Despite that regulatory approval, local resistance has had tangible financial consequences. Before construction can commence, Oracle and Related Digital must secure more than USD 10 billion in project financing. According to The Information, investment firm Blue Owl declined last week to participate, citing concerns that the lease and debt terms were less attractive than for other Oracle‑backed projects and that persistent local opposition could cause further delays. Oracle stated that Blue Owl’s decision would not affect the timeline and that alternative financing channels may be available, but the announcement nevertheless triggered a 5% decline in Oracle’s share price.

Another common source of delay arises when developers and prospective tenants sign letters of intent or enter exclusive negotiation periods but fail to convert those agreements into binding leases. Fermi, a data center developer co‑founded by former Texas governor Rick Perry, disclosed in regulatory filings that its AI server campus in West Texas did not secure tenant leases by the contractual deadline. Business Insider reported that the withdrawing tenant was Amazon Web Services (AWS), and Fermi’s stock subsequently plunged more than 40%.

Such outcomes are not rare. Earlier in the year, Microsoft chose not to sign additional data center agreements with CoreWeave in Texas, and both AWS and Microsoft abandoned plans to lease facilities developed by Applied Digital in North Dakota. Because lenders typically release large construction loans only after developers obtain binding agreements with top‑tier tenants, letters of intent alone are insufficient to unlock the billions in financing required.

More severe delays involve projects that have already secured funding and advanced into construction. The Information reports that a small number of AI data center developments have experienced repeated problems that led customers to exit binding contracts before completion. These situations threaten project viability. In extreme cases, developers have been forced to offer Nvidia server rentals at half price to retain tenants. Industry veterans note that data center contracts often favor tenants; if facilities fail to meet delivery milestones, customers may be entitled to exit or obtain substantial discounts, undermining the project’s economics. For investors, the central question has shifted from whether delays will occur to whether project teams possess the capability to resolve issues and deliver on schedule.

Confronted with these infrastructure constraints, Nvidia has taken a more proactive stance. Jensen Huang convened roughly 100 executives at the company’s Santa Clara headquarters for a closed meeting focused on electricity as the core constraint on AI expansion. Attendees reportedly included major energy and electrical infrastructure firms such as Schneider, General Electric, Hitachi and Siemens Energy, along with representatives from cloud service providers. Huang emphasized that AI has transformed data centers into factories that consume power to produce intelligence, and he conveyed a clear message: entities that can rapidly expand power supply will capture the next wave of AI demand.

The summit underscored the growing importance of behind‑the‑meter generation. Rather than waiting months or years for public‑grid interconnection, some operators are turning to on‑site generation; OpenAI’s facility in Abilene, Texas, is an example of this approach. Huang also urged the power sector to improve data center energy efficiency to increase the AI output per watt.

Other notable developments in the infrastructure landscape include Alphabet’s acquisition of Intersect for USD 4.75 billion, a deal that secures land adjacent to high‑quality wind, solar and battery resources often described as “electrified land.” At the same time, regulatory scrutiny has intensified, with Vermont Senator Bernie Sanders calling for a pause in AI data center construction. New project collaborations are also emerging: Texas Pacific Land Corporation announced a strategic agreement with Bolt Data & Energy, a company co‑founded by Eric Schmidt, to develop large‑scale data center campuses.

The industry now faces a pivotal test. The capacity to overcome political obstacles, secure robust financing and accelerate delivery of power and critical equipment will determine which operators can scale effectively and which projects will falter amid the growing complexity of AI infrastructure deployment.