AI Will Hit a Wall in 2026, if nothing changes.

TL;DR

AI progress will stall not from lacking smarter models, but because power grid bottlenecks mean we can't physically turn the data centers on.

Key Points

• 1.Energy supply, not chips or money, is the core limiter of AI scaling. The IEA forecasts global data center electricity use doubling by 2030, growing at 15% annually — four times faster than overall electricity demand.
• 2.Half of planned 2026 data centers may face delays. Of 16+ GW planned, only 5 GW is under construction; Morgan Stanley warns developers expect power constraints by 2027–2028 due to grid underinvestment.
• 3.The US grid has a 2 TeraWatt backlog of power projects awaiting connection. Median wait time is five years, with some regions like Northern Virginia facing 9–12 year queues, driven largely by transformer shortages and poor planning.
• 4.Proposed fixes like small modular nuclear reactors are unlikely to save the timeline. Elon Musk confirmed chip production will soon outpace available electricity, and SMRs have historically run a decade late and a billion dollars over budget.