Why smarter AI models could drive up compute prices 10x

Why smarter AI models could drive up compute prices 10x – Dwarkesh Podcast

Market views

Bullish

ANTHROPIC— Anthropic has achieved 10x year-over-year revenue growth for three consecutive years, ending last year at $9B and projected to reach $100-150B this year, demonstrating exceptional scaling and monetization of AI capabilities.
CLOUDGPU— Compute prices are rising significantly (40%+ since February) and will continue increasing as AI models become more valuable, with premium compute commanding 2x spot prices. As AI reaches human-level capabilities, H100 rental prices could reach 15x current levels.
AILABS— Leading AI labs have strong competitive moats due to economies of scale in model training and ability to outbid competitors for scarce compute resources. Labs with more efficient models can charge premium margins due to the Alkin Allen effect.

Bearish

OPENAI— OpenAI's compute allocation shifted from 25% inference in 2024 to potentially 50%+ now, signaling potential stagnation in model improvement and shift toward being a cloud provider rather than AGI builder, which is less compelling to investors.
INFRA— Compute supply growth is structurally constrained and unlikely to sustain even 3x year-over-year growth due to Moore's Law limitations, ASML EUV machine bottlenecks through 2030, and wafer allocation hitting ceiling at 86% by end of next year.

Not investment advice.