- The Fed sees U.S. data center construction potentially reaching US$360B–US$930B annually by 2027, far above historical norms and driven by AI and cloud demand.
- Massive build-outs by hyperscalers like Meta and AWS are straining land, skilled labor, specialized equipment, and especially power infrastructure.
- Grid capacity limits, crowded interconnection queues, and regulatory friction could delay projects and leave many AI data centers power-constrained.
- Speculative announcements and possible overbuild raise risks of stranded assets, higher costs for ratepayers, and instability for both utilities and investors.
Read More
The Fed’s recent projections, as reported in CoStar’s December 2025 article, depict an explosive growth trajectory for the U.S. data center sector through late 2027. The best‐case scenario assumes roughly a doubling of current momentum, pushing annual construction investment toward the upper bound of US$900 billion; the base case centers around US$600 billion annually. However, a conservative scenario—assuming a 75% slowdown in new announcements—still implies investment of several hundred billion dollars annually—well above historical norms.
This scale is being driven heavily by generative AI and cloud infrastructure demand. Some of the largest projects are already underway: Meta’s 2.4 million square foot “Project Tropical” in Virginia, Meta’s 35 projects totaling 17.3 million square feet, and Amazon (AWS) handling 54 properties totaling 17 million square feet based on CoStar data through 2027. But this rapid build‐out places stress on multiple levers of supply: industrial land, electrician and contractor labor, production of specialized equipment, and especially on power infrastructure.
Independent analyses underscore similar concerns. Bain & Company projects U.S. data center electricity demand could double to ~409 terawatt-hours by 2030 under a baseline scenario, with construction delays, skilled labor shortages, permitting friction, and utility interconnection latency being top risk factors. Gartner estimates that by 2027 about 40% of AI data centers may be constrained by power availability. Studies of grid and interconnection bottlenecks further note that utilities in fast‐growing data center hubs like Texas, Virginia, and the Southeast are facing crowded interconnection queues and soaring transmission and distribution (T&D) upgrade demands.
Several strategic implications emerge for investors, operators, and policymakers:
- Project selection and staging will matter: sites with firm power commitments, modular design, and prefabricated construction may mitigate delays.
- Utility partnerships will become a key competitive advantage; regulators and utilities must balance approving growth with protecting ratepayers from speculative overcommitment.
- Sustainability and ESG commitments will be increasingly difficult to maintain; access to clean power and emissions from backup generation or fossil fuel sources are critical risks.
- There is significant risk of mismatch between build announcements and actual delivery—potential for stranded assets or overinvested capacity especially in markets promising cheap power but with long grid upgrade lead times.
Open questions to monitor:
- How many announced projects are speculative versus having secured power, equipment, permitting, and tenants?
- How will grid regulatory reform evolve at state and federal levels to handle accelerated load growth?
- What role will behind‐the‐meter power sources (on-site generation, battery storage, gas turbines) play in bridging supply gaps?
- How will rising electricity prices or carbon policies affect operating margins for data center operators?
Supporting Notes
- Fed projection ranges: US$360B-US$930B annually by late 2027; mid-case ~US$600B.
- Current investment is accelerating sharply from roughly US$100B annually in early 2024.
- CoStar data shows 125 data centers totaling 33.5 million square feet completed in 2025; 219 totaling 76.1 million in 2026; 183 totaling 73.4 million in 2027.
- Meta is leading with 35 projects totaling 17.3 million square feet; Amazon/AWS with 54 properties totaling 17 million square feet through 2027.
- Fed identifies constraints: land, labor, equipment, power, and capital.
- Bain forecasts U.S. demand doubling to ~409 TWh by 2030; global capacity reaching ~163 GW.
- Gartner predicts 40% of AI data centers will be power‐constrained by 2027.
- S&P Global notes that grid T&D constraints, interconnection queues, and renewable curtailment are major risks.
- Utilities like Georgia Power propose 50% expansion in capacity over six years, mostly to serve data centers; regulators challenge overbuilding absent firm contracts.
- Industry concern over speculative or duplicate project filings that inflate infrastructure planning pipelines.
Sources
- www.costar.com (CoStar Group) — December 30, 2025
- www.prnewswire.com (Bain & Company via PR Newswire) — October 22, 2025
- www.gartner.com (Gartner) — November 12, 2024
- www.spglobal.com (S&P Global Market Intelligence) — October 16, 2025
- www.bcg.com (Boston Consulting Group) — January 20, 2025
- apnews.com (AP News) — December 2025
- www.ft.com (Financial Times) — November 2025
- www.datacenterfrontier.com (Data Center Frontier) — December 1, 2025
