Energy Demand vs Grid Capacity Collision
conceptAI & Compute · Money & Finance
We're building a wave of power-hungry AI on top of an electric grid that physically can't keep up.
Who they are
The looming clash between soaring electricity demand and a US power grid that can't deliver it fast enough[1].
What they do
The engine treats this as a hard physical bottleneck: what's being built needs far more power, far sooner, than the grid can supply.
How it works
By the end of 2024 more than 10,300 projects (1,400 GW of generation, 890 GW of storage) were stuck in line waiting to connect; grid upgrades take 5-10 years and small nuclear reactors won't be ready until the 2030s, while AI computing demand roughly doubles every 18-24 months. The stopgap is natural gas, which means an AI industry that claims to fight climate change is actually driving a spike in fossil-fuel burning. Updated 13 Sep 2026 from Berkeley Lab's own figures: by the end of 2025 it was about 8,200 projects — 1,312 GW of generation and roughly 749 GW of storage — and the total waiting in line actually FELL 10% on the year, as failed projects dropped out faster than new ones joined. The mix is the real story: gas in the queue jumped 86% to 253 GW while solar fell 19%, storage 16% and wind 19%. Another 549 GW has its paperwork done but still isn't running.
Why it matters
The engine warns this collision speeds up a pollution-driven path toward a 2040 crisis point, because the grid simply cannot support what's being built on it. The gas jump puts a number on the contradiction this entry already named. A shrinking queue could mean the rules are clearing out projects that were never real, or it could mean people are giving up; the engine doesn't pick between those.
The engine's record — word for word
US grid interconnection backlog: 10,300 projects (1,400 GW generation, 890 GW storage) awaiting connection as of end 2024. Grid upgrades take 5-10 years. AI compute demand doubling every 18-24 months. SMRs not at scale until 2030s. Natural gas bridge = contradiction: AI purportedly solving climate change while driving massive hydrocarbon spikes. BAU2 collision: AI compute demand accelerates pollution-driven collapse timeline, compressing path to 2040 disruption. PJM Interconnection forecasting extreme load growth. The grid cannot support what is being built on it. [2026-09-13 primary-source refresh — LBNL Queued Up: 2026 Edition, data through 2025] The end-2024 figures above are superseded, and the direction of travel is not what the raw project count suggests: ~8,200 projects were actively seeking interconnection at end-2025, representing 1,312 GW of generation and ~749 GW of storage — fewer projects, and total active queue volume DOWN 10% year over year on high withdrawal rates and fewer new requests[2]. THE MIX IS THE FINDING: active natural gas capacity rose to 253 GW (+86%) in 2025 while solar (773 GW, -19%), storage (749 GW, -16%) and wind (220 GW, -19%) all fell — this node's existing line 'Natural gas bridge = contradiction' now carries its quantity. 549 GW already holds a draft or executed interconnection agreement without reaching commercial operation, per the same source. Held: a shrinking queue is consistent both with FERC Order 2023 clearing speculative projects and with demand-side withdrawal; this node does not collapse between those readings.
Follow the trail
Walk this on the live map →