The Biggest Bottlenecks For AI: Energy & Cooling

a16z Growth GPs David George & Jen Kha on why energy/cooling are the real AI bottlenecks, private market dynamics, and where AI business models are actually sticky

• Big tech annualized capex run-rate ~$400B, mostly AI infra — George says even that number is too conservative.
• AI input costs dropped 99%+ (100x) in 2 years; frontier model capability doubles every 7 months — faster than Moore’s Law.
• ChatGPT hit 365B searches in 2 years vs Google’s 11 years (5.5× faster), now ~1–2B MAU but only 30–40M paying.
• Energy is bottleneck now; cooling is the underappreciated next bottleneck after energy gets solved via nuclear/gas.
• XAI stood up the world’s largest data center in ¼ the normal time by buying every backup generator in a multi-state region.
• Consumer stickiness > B2B for AI: developer API usage switches on a better model API call; David’s parents in Kentucky won’t switch from ChatGPT.
• Stickiest AI apps: medical scribe, customer support, high-end financial analysis — because workflow rules + brand voice get embedded.
• Task-based pricing (monetizing human-task replacement) only clearly working in customer support; George is low-conviction it becomes universal in 5 years.
• Private unicorn market cap hit $3.5T (7× from $500B a decade ago); only 5% of public software companies forecast 25%+ growth — high-growth tech is now almost entirely private.
• a16z Growth was first outside money in xAI beyond Elon; 80% of growth investments have a pre-existing early-stage relationship.

Guests: David George (a16z General Partner, Growth Fund), Jen Kha (a16z Head of Investor Relations) · 2026-01-26 · Watch on YouTube