The Megawatt
Thesis
How a 24-year-old turned $225M into $5.5B by betting on the one thing AI actually needs.
Leopold Aschenbrenner
He didn't read about
AI infrastructure.
He built it.
Working inside OpenAI's Superalignment team gave him direct sight of the actual compute requirements at the frontier. He didn't make a financial bet on a trend. He translated what he'd seen from the inside into a portfolio thesis.
Backed by
He sold every share of Nvidia and Broadcom. Then bought fuel cells, Bitcoin miners, and power companies.
While AI funds were stacking Nvidia at 30x revenue, Leopold built the exact opposite portfolio.
A 165-page essay.
AGI by 2027.
Written in 2024, while he was building his fund. Specific, dated, unflinching. Not a vague prediction — a roadmap with a delivery estimate.
You can order 100,000 GPUs and take delivery in six months. You cannot add 500 megawatts to the grid in six months.
The binding constraint on who builds AGI first is watts per rack.
AI compute scales
every 12–18 months.
Inference runs on top.
Continuously. For everyone.
on top
The power problem is not just about building AGI. It's about running it once you have it.
US electricity production
grew 5% in a decade.
The grid was not built for this. Transformer shortages. Switchgear backorders. Cancelled data centre projects. These are not teething problems — they are the first visible symptoms of a power system hitting a wall.
The bottleneck is not the models. Not the chips. Not the software. The bottleneck is whether civilisation can generate enough electricity to run the machines fast enough to matter.
Bloom Energy.
Power in 55 days.
Solid oxide fuel cells that generate electricity directly at the data centre site. No grid connection required. While competitors wait 2–3 years for grid interconnection, Bloom delivers a fully operational system in under two months. Oracle's 1.2 GW deal closed because Bloom beat its own schedule by a month.
Two curves.
One widening gap.
Leopold bet his entire net worth on the gap between those two curves. So far, the gap is only getting wider.
He also holds Core Scientific and other former Bitcoin miners — betting they have the plugs AI needs to grow.
Oracle. 2.8 gigawatts.
The thesis confirmed.
Oracle contracted 1.2 GW of Bloom fuel cells immediately, with a pipeline to 2.8 GW total. They chose Bloom because it delivered a fully operational system in 55 days — a month ahead of schedule. Grid interconnection would have taken three years. Oracle needed power yesterday.
He did not buy a power company because he liked the chart. He bought it because he ran the math on what AGI physically requires to exist — and concluded that electricity is the asset class of the decade.
His own timeline says the demand that drove this trade is just getting started.
