Summary
Approximately half of US data centers scheduled to open in 2026 face cancellation or delays, per Bloomberg reporting on Sightline Climate analysis. Of 12 GW planned for 2026, only 4 GW is actually under construction. The critical bottleneck is not computing hardware but electrical components (batteries, transformers, circuit breakers) manufactured overseas — representing less than 10% of construction costs but gating entire projects. Additional pressures include helium shortages, cash constraints, and community opposition.
Key Points
- Of 12 GW of data center capacity planned for 2026, only 4 GW (one-third) is under construction.
- 2027 projections are worse: 6.3 GW under construction vs. 21.5 GW announced.
- 2028-2032: 37 GW planned but only 4.5 GW has broken ground.
- The critical constraint is electrical components from overseas (Canada, Mexico, South Korea, China) — batteries, transformers, circuit breakers.
- These components represent less than 10% of construction costs but are impossible to substitute and gate entire projects.
- Andrew Likens (Crusoe Energy) stated: “If one piece of your supply chain is delayed, then your whole project can’t deliver.”
- Additional pressures: helium shortages, cash constraints, community opposition to data center construction.
- Data center firms source components from Canada, Mexico, South Korea, and China, creating extended timelines.
Newsletter Angles
- The infrastructure gap is real: The AI buildout is not a software problem — it is a physical infrastructure problem constrained by transformers, circuit breakers, and electrical components. Less than 10% of costs, but 100% of the gating factor.
- China dependency in AI infrastructure: The components bottlenecking US data centers come from China. This creates a direct tension with decoupling rhetoric — you cannot build sovereign AI infrastructure with Chinese electrical components and simultaneously pursue trade restrictions.
- Community opposition as friction: Local political pushback is an underreported constraint on data center expansion. This connects power/infrastructure themes to local politics.
- Energy as the real AI bottleneck: Not compute, not memory, not algorithms — electrical infrastructure is what’s actually limiting AI scaling.
Entities Mentioned
- Crusoe Energy — Andrew Likens quoted on supply chain constraints; building AI factory in Abilene
- Sightline Climate — analytics firm whose data underpins the Bloomberg reporting
Concepts Mentioned
- AI Infrastructure Spending — the gap between announced capacity and actual construction
- Data Center Buildout Delays — systemic delays driven by supply chain, energy, and political constraints
- Supply Chain Concentration — dependency on overseas electrical components as a single point of failure
- Chokepoint Control — China’s role in supplying critical electrical components for US data centers
Quotes
“If one piece of your supply chain is delayed, then your whole project can’t deliver.” — Andrew Likens, Crusoe Energy
Notes
- Based on Bloomberg reporting and Sightline Climate analysis — credible sourcing.
- The 4 GW vs. 12 GW figure for 2026 is a stark metric. The 2027-2032 figures are even more dramatic.
- Helium shortages mentioned here corroborate the Strait of Hormuz/helium supply chain described in Big A — The Crisis Got Weirder (RAM Apocalypse Update).
- The “less than 10% of costs but gating” framing is the key insight — the bottleneck is not where the money is.