China is killing the US on energy. Does that mean they’ll win AGI? — Casey Handmer

Casey Handmer argues solar’s 43% learning rate and battery temporal arbitrage will dominate AI data center power by the early 2030s, making natural gas a transitional dead end.

• Solar’s Wright’s Law coefficient is 43%: every doubling of production cuts cost by 43%, with demand growing 6x faster than supply additions.
• All gas turbine capacity through ~2030 is already spoken for; supply constraints will force AI data centers to solar beyond that.
• Amortized Brayton cycle cost alone is ~$35/MWh before fuel, cooling, or grid delivery — structurally expensive vs. solar.
• Electricity is <10% of AI inference cost, so hyperscalers can absorb 100x electricity price increases without meaningful impact on margins.
• GDP will undercount AGI’s value the same way it undercounts oil (1% of GDP but causes double-digit recessions if removed) — total energy use is a better civilization metric.
• AGI automating human labor targets a $60T global wage base; OpenAI’s $20B ARR is a rounding error compared to the lower bound.
• Handmer’s long-run vision: silicon wafer solar sails in space — one square meter simulating one human brain, no batteries needed, self-steering via integrated LCD panels.

2025-08-15 · Watch on YouTube