Ambitions are rich, but reality is harsh! Nearly half of U.S. data center projects for 2026 face cancellation or delays.

```

The grand narrative of the expansion of U.S. artificial intelligence infrastructure is facing a harsh reality check.

According to the latest "2026 Data Center Outlook" report released by Sightline Climate, of the approximately 16 GW of new data center capacity planned in the U.S. this year, 30% to 50% are expected to be delayed or canceled, and only about 5 GW have actually entered the construction phase.

This figure stands in stark contrast to the more than $700 billion in annual capital expenditures by hyperscale cloud computing firms, revealing the deep-rooted difficulties in AI infrastructure construction in terms of electricity supply, supply chain, and sociopolitical factors.

Canaccord Genuity analyst George Gianarikas described the situation as "the U.S. data center boom hitting a formidable wall of logistical resistance." According to Bloomberg, a severe shortage of electrical equipment such as transformers, switchgear, and batteries is one of the main causes of delays, with U.S. domestic manufacturing capacity falling far short of demand, forcing builders to rely on imports. Meanwhile, community opposition, permitting hurdles, and delays in grid integration are collectively shrinking the feasible space for project execution.

The Supply-Demand Gap is Shocking, and Prospects After 2027 Look Even Bleaker

The Sightline Climate report shows that in 2026, there are 140 data center projects planned to go into operation across the U.S., with a combined capacity of about 16 GW. Of these, 53% will be connected to the grid, 3% will rely on self-supplied power, and another 25% have not yet disclosed their power supply plans.

However, of this planned 16 GW, 11 GW remain at the "announcement stage" with no sign of construction—given a typical data center construction cycle of 12 to 18 months, it means these capacities are very unlikely to come online as scheduled.

Looking ahead to 2027, the gap will widen further. Announced planned capacity reaches 21.5 GW, but the actual construction underway is only about 6.3 GW. Between 2028 and 2032 the situation is even more severe: the vast majority of planned projects have not broken ground, and 37 GW of planned infrastructure don’t even have a confirmed completion date, with only 4.5 GW actually under construction.

Transformer Bottleneck is Critical, with High Dependence on Imported Supply Chains

The shortage of power infrastructure is the core bottleneck restricting data center construction. The rapid expansion of data center scale requires larger and more powerful transformers to safely convert high-voltage grid electricity for delivery to chips. GE Vernova Electrification CEO Philippe Piron pointed out that prior to 2020, the delivery cycle for high-capacity transformers was typically 24 to 30 months, which was "completely acceptable" under the old model, but AI companies often require delivery within 18 months.

The surge in demand has extended transformer lead times to up to five years, with prices soaring as well. Some companies have been forced to adopt stopgap measures, such as Crusoe, which has begun refurbishing old transformers from abandoned power plants to meet urgent needs.

The pressure on demand for transformers and other electrical equipment is coming not only from data centers, but also from the popularity of electric vehicles and heat pumps, which are likewise driving up demand for grid expansion. U.S. domestic manufacturing capacity simply cannot keep up, deepening dependence on imports. This predicament reflects deep structural problems from decades of offshoring manufacturing—despite repeated policy calls in recent years for reshoring, there has been little substantive increase in capacity.

Nuclear Power Commitments Unmet, Funding Gap Reaches Trillions

The outlook on the power supply side is also grim. The Trump administration's promises about a nuclear renaissance remain largely rhetorical, with almost no new nuclear plants breaking ground. Small modular reactors are seen as a beacon of hope, but large-scale deployment will still take years.

A handful of ultra-large self-powered projects are trying to bypass grid limitations, including New Era Energy & Digital's planned 7 GW project in Lea County, New Mexico; a 4.5 GW coal-to-gas project in Homer City, Pennsylvania; and a 1.8 GW hybrid natural gas and renewable energy project by Crusoe in Cheyenne, Wyoming. The report notes that waiting for the grid to offer this level of power delivery "might take ten years."

On the funding side, J.P. Morgan analysis indicates that supporting this AI cycle will require no less than $5 trillion, and even if hyperscale cloud companies continue ramping up capital spending, the U.S. government will still need to fill a gap of over $1 trillion.

Social Resistance Rises, Public Sentiment is Shifting Fast

Beyond infrastructure problems, sociopolitical resistance is also rapidly accumulating. The Maine House of Representatives passed a moratorium by 82–62 to halt construction of large data centers until 2027, and simultaneously established a Maine Data Center Coordination Committee to evaluate the impact of data centers on the state's resources, environment, and finances.

In a recent report, Goldman Sachs executive director Shreeti Kapa wrote that she sensed a strong consensus at investor dinners: "There’s simply not enough compute power. Every participant is facing severe compute constraints—from fabs to data center permits, to power, memory, and labor. The bottlenecks are real, and they will last for quite some time."

A recent Quinnipiac University poll shows that public concerns about the deep integration of AI into healthcare, education, and daily life are rising quickly, further increasing the social friction cost of industry expansion.

Canaccord summarizes: "Energy constraints are intensifying, and so are sociopolitical constraints. Something has to change." Faced with abundant capital but real-world implementation roadblocks, the path to expanding U.S. AI infrastructure is much more tortuous than the market had anticipated.

Risk warnings and disclaimerThe market involves risks, and investments should be made cautiously. This article does not constitute personal investment advice and does not take into account individual users' specific investment objectives, financial status, or needs. Users should consider whether any opinions, views, or conclusions in this article fit their particular circumstances. Investments made accordingly are at the user's own risk. ```