A rapid wave of data‑center construction in the United States has become one of the largest capital undertakings in 2026, reshaping technological competition, energy demand and local politics. The five largest hyperscalers — Amazon, Microsoft, Google, Meta and Oracle — plan to spend more than $750 billion on capital expenditures this year, a 67% increase from last year; roughly 75% of that spending is earmarked for AI infrastructure.
The stakes
The scale of spending, competition for physical inputs (land, cooling systems, turbines) and the strain on the power system have economic and political consequences. Data‑center investment has supported over $21 billion of new private construction spending in the past year even as the rest of the construction industry contracted by more than $100 billion.
Three core realities
- They compete for everything: Hyperscalers are bidding for sites and infrastructure, not just GPUs. For example, Amazon purchased 188 acres in Northern Virginia last year for $700 million (about $3.7 million per acre), a parcel that a homebuilder had assembled for $51 million to build roughly 500 homes.
- They are stressing the grid: Utilities now project a sixfold increase in 2030 peak electricity demand growth compared with forecasts three years ago, driven largely by data centers. Aging infrastructure is struggling to keep pace.
- They are a defining political issue: Both Republicans and Democrats have embraced restrictions or moratoria as an electoral strategy. At least 75 projects nationwide, representing roughly $130 billion in potential investment, were affected by political delays in Q1 2026.
Key numbers and averages
- Total online data centers in the U.S.: about 4,000 and growing, including 580 hyperscale facilities.
- Average site capacity for large projects in the pipeline: 62 MW (compared with an average Virginia site of roughly 34 MW at the end of 2024).
- Average build cost for a 62 MW project: roughly $2.1–2.4 billion.
- Average permanent staff: about 50 people per typical Virginia site; on‑site construction workforce can peak around 1,500 during the 12–18 month build period.
- Water use: a hyperscale data center’s direct water use is roughly comparable to that of a major U.S. golf course. Even accounting for indirect water used to generate power (about 12x higher), total data‑center water use nationwide in 2023 was less than 1% of agricultural irrigation — though local watershed impacts can be significant.
- Electricity share: data centers consumed 4.4% of U.S. electricity in 2023 and are forecast to approach about 12% by 2030.
Supply chain and labor pressures
Every input is under pressure. Hitachi Energy’s North American transformer supply‑chain lead reports that 44% of data‑center leaders see utility wait times beyond four years. The construction sector needs roughly 350,000 additional workers this year to meet demand; analysts say there are not enough specialized workers to finish cooling and electrical installations.
Price effects and grid policy
The capacity shortage is pushing up electricity costs: PJM, the largest U.S. grid operator, has seen capacity prices spike 11x in a few years, increases that utilities pass through to consumer bills. PJM proposed prioritizing the disconnection of data centers without on‑site power plants during periods of high system stress. OpenAI’s job postings for roles like a power trading lead indicate how companies are preparing to manage these energy risks internally.
Major projects and their implications
Meta’s Hyperion project in northeastern Louisiana is one of the most visible examples. Meta has committed more than $50 billion to build the campus, which aims for about 5 GW of capacity upon completion around 2032 — roughly 80 times the average project noted above. Meta is financing 10 new natural‑gas plants to supply the site. Meta only owns 20% of the project; the remainder is owned by funds managed by Blue Owl Capital, with Meta leasing the campus, a structure that offers Meta flexibility if the AI boom cools.
Nvidia pledged over $100 billion in guarantees for an 8 GW Ohio project intended to power OpenAI — another signal of how large these builds can become.
The politics: public opposition and local backlash
Gallup polling shows more than 70% of Americans oppose a data center being built near them — a higher level of opposition than for nuclear plants. In Virginia, often considered the U.S. data‑center heartland, public support has halved in three years. Many competitive House districts this fall have data centers already planned or under construction, and virtually all congressional and gubernatorial ad mentions tracked by AdImpact are negative toward data centers.
Concrete political actions have followed: Pennsylvania Governor Josh Shapiro (D) imposed stricter requirements on new data centers after previously celebrating a large project; Michigan Republican Senate nominee Mike Rogers backed a one‑year moratorium; and Texas Governor Greg Abbott temporarily froze new data centers.
Bottom line
The U.S. data‑center boom is a proxy fight over AI, jobs, the environment, energy production and energy use. It is driving unprecedented private spending and supply‑chain competition, stretching the electric grid and provoking a bipartisan local backlash that could influence elections and reshape how communities negotiate incentives and permits for tech infrastructure.
