WhatMeal(@WhatMealAI):Leopold Aschenbrenner’s 2026 AI Investment Thesis: By 2026, AI needs gigawatt-scale training clusters (nuclear-reactor level power) costing tens of billions, scaling to 100,000s of H100-equivalent GPUs. Key long positions: CRWV, CLSK, and MARA (Bitcoin miners with power assets that can pivot to AI). Focus on bottlenecks: AI chips, nuclear/power generation, energy infrastructure, and data centers. He is betting big on energy and Bitcoin mining over pure chips. The game has shifted from AI software to the physical infrastructure powering it.

2026.05.18 19:25

Leopold Aschenbrenner’s 2026 AI Investment Thesis: By 2026, AI needs gigawatt-scale training clusters (nuclear-reactor level power) costing tens of billions, scaling to 100,000s of H100-equivalent GPUs. Key long positions: CRWV, CLSK, and MARA (Bitcoin miners with power assets that can pivot to AI). Focus on bottlenecks: AI chips, nuclear/power generation, energy infrastructure, and data centers. He is betting big on energy and Bitcoin mining over pure chips. The game has shifted from AI software to the physical infrastructure powering it.