Power Grid Constraints Emerged As AI Growth Bottleneck
The widening gap between power supply and data center demand threatens AI development, with a 19-gigawatt deficit projected by 2027.
Updated on Oct. 12, 2026 in Data Centers
Live Poll
Do you expect your monthly electricity bill to rise due to increased regional data center construction?
UBS analysts identified that power grid limitations have become the primary constraint for the AI industry's growth in the United States. While 57 gigawatts of capacity are required by 2027, only 38 gigawatts of power are currently confirmed.
Why it matters
The massive energy requirements for AI data centers face severe delays from regulatory hurdles, supply-chain bottlenecks for components like transformers, and local community opposition. This imbalance is driving a shift in market valuations for colocation data center stocks as firms struggle to secure baseload energy.
Current infrastructure lead times for large transformers have reached 210 weeks, severely hindering capacity expansion. By contrast, new power generation requires up to 10 years to commission, dwarfing the 12 to 24-month timeline required for constructing the data centers themselves.
The players
UBS
A global financial services firm that provides investment research and banking, recently signaling a shift in its outlook on data center infrastructure.
The details
Data centers require a constant, stable baseload power supply similar to that of a mid-sized city to maintain server operations. The delay in scaling this capacity stems from complex interconnection processes—the formal application for a power plant to connect to the grid—which currently take a median of 6 years to reach commercial operation. Supply-chain constraints, specifically for high-voltage transformers, further exacerbate these delays. Additionally, regulatory barriers have begun to emerge, such as New York's one-year moratorium on data centers exceeding 20 megawatts.
Timeline
2024 marks the baseline for US installed power capacity.
September 23, 2026, UBS initiated coverage on colocation stocks.
2027 is the projected target year for data center capacity demand.
2027-2029 represents the forecasted EBITDA growth period for colocation firms.
2030 is the target date for $511 billion in new power generation investment.
The Tech Race
The US interconnection queue holds 2.3 terawatts of capacity, yet the industry faces a significant bottleneck in converting these plans into active power. This grid saturation follows a pattern where infrastructure development consistently lags behind the exponential growth in demand for high-performance computing.
The current power deficit will likely favor large, established colocation firms capable of securing long-term power purchase agreements ahead of smaller competitors. Regional moratoriums, like the one in New York, may force developers to prioritize locations with existing grid surplus, shifting where new AI infrastructure is deployed.
The takeaway
The bottleneck in AI growth is now less about chip compute and more about grid interconnection and transformer supply. Investors and developers should monitor the $511 billion investment goal for 2030 as the critical benchmark for whether the industry can overcome these physical constraints.
Further reading
Learn more about infrastructure trends in our Data Centers section.
Live Poll
Do you expect your monthly electricity bill to rise due to increased regional data center construction?
