Key ideas from Dylan Patel

• Big tech's $600 billion combined CapEx includes deposits for turbines in 2028-29 and data center construction for 2027 - Dylan
• Anthropic needs 5+ gigawatts by year-end to serve $60 billion projected revenue growth, requiring $40 billion compute spend - Dylan
• A single gigawatt of AI chips requires 3.5 EUV tools, with ASML producing only 70 tools this year, scaling to 100 by 2030
• Memory prices have tripled, forcing smartphone volumes down 40% as DRAM gets reallocated from consumer devices to AI chips
• H100 rental prices have inflected from $1.40 to $2.40+ per hour as AI labs crowd out other compute buyers - Dylan
• By 2030, 700 EUV tools could support 200 gigawatts of AI capacity, making Sam Altman's 52 gigawatts annually feasible at 25% market share
• Power constraints are solvable through behind-the-meter gas turbines, reciprocating engines, and other alternatives beyond traditional grid capacity
• China could achieve semiconductor supply chain dominance by 2035 if AI timelines extend, but fast takeoff scenarios favor Western infrastructure advantages