500MW of compute could serve $75B of inference at DeepSeek prices, $300B at Anthropic's

bookwormengr · x · 2026-11-18

Responding to a claim that "Anthropic might only need 500 MW of inference next year," bookwormengr runs the inference economics: even at 500MW of compute, you could serve $75B worth of inference at DeepSeek Flash off-peak prices, or $300B at Anthropic's prices—the gap comes purely from pricing.

To the pushback that models keep getting bigger and the assumption breaks: the math doesn't change, because larger models command higher token rates, so inference revenue scales alongside compute demand.

Related event: Debate Erupts Over Claim That Anthropic Needs Only 500MW for Inference(3 posts)→

Original post →

More from Venture

Venture channel →