The AI Buildout Meets Resistance as OpenAI Pullback Exposes Hidden Constraints

The data center gold rush is getting its first collective hangover. OpenAI has hit pause on its UK Stargate data center plans as expansion plans across the sector begin to cool. Just weeks after raising at an $852B valuation, it’s a sign that even cash-rich AI players are getting more selective about where they spend.

Tightening the belt: The ChatGPT maker pointed to soaring energy costs and regulatory hurdles as it shelved plans to deploy up to 31K NvidiaNVDA

chips in northeast England. But the pullback isn’t happening in isolation. OpenAI has also scrapped parts of its Texas expansion with OracleORCL

and shut down its Sora video app in March as pressure from Anthropic and GoogleGOOGL

builds. The pattern points to a broader reset, with companies potentially trimming back ahead of a potential IPO.

• OpenAI says it will “move forward when the right conditions… enable long-term infrastructure investment,” signaling a wait-and-watch approach.
• That hesitation is already hitting policy goals, with UK Prime Minister Keir Starmer’s AI hub ambitions facing setbacks as grid delays stall major projects.

The AI Boom Hits Bedrock

The bigger picture points to an industry running into real infrastructure limits. Bloomberg reports that nearly half of planned US data center projects for 2026 could face delays or cancellations, as bottlenecks shift from demand to buildout. The main constraint is power equipment, with shortages of transformers and switchgear stretching lead times from roughly 24–30 months before 2020 to nearly five years today, effectively putting a cap on how fast AI capacity can scale.

• Research from 10a Labs shows that data center projects worth $156B faced local opposition in 2025, driven by concerns over noise, water, and air quality.
• China supplies over 40% of US battery imports and ~30% of key grid equipment, meaning trade tensions could further slow deployment timelines.

Slamming into the infra wall: Morgan Stanley’s Todd Castagno told The New York Times that “a lot of the commitments and build-out of data centers in easily accessible areas has already occurred,” making new projects harder to execute. That constraint is starting to matter more than demand itself, even as AI hyperscalers plan to pour roughly $710B into North American data centers in 2026. It’s an early sign the AI buildout is running into physical limits, not just capital ones.