DeepSeek's token density yields ~$38/GPU-hr revenue vs $3.50 for raw IaaS, engineer estimates
bookwormengr · x · 2026-10-21
An engineer's revenue math on compute economics: bare-metal IaaS at $3.50/GPU-hr generates $13.6B per GW per year, while DeepSeek's reported production token density — running DeepInfer with speculative decoding (DSpark) and 96% KV-cache hit rate at high token saturation — is equivalent to $38.00 per GPU-hour. That works out to roughly $147.7B per GW per year; even with 33% extra slack for power inefficiencies, $100B per GW is a reasonable assumption. That's a >10x gap in revenue per unit of compute between renting raw GPUs and serving tokens.
Related event: DeepSeek's Inference Density: $38 Revenue per GPU-Hour(3 posts)→
More from Infra
- FastVideo open-sources MiniMax FastH3 Omni with training code, 12x faster on Blackwell — multimodalart · 2026-10-21
- Anthropic's draft IPO filing: $16.3B H1 revenue, $80B net loss, $750B compute plans — kimmonismus · 2026-10-21
- dphnAI's distributed inference network served 32B DeepSeek tokens in 24h, ~6% of Venice's daily volume — Ronangmi · 2026-10-21
- Jensen Huang: every company will have two factories, one real and one digital — round · 2026-10-21
- DeepSeek V4 called 'god-tier infra, 4th-tier ML': 7 tricks stacked, 20 patches to fix them — teortaxesTex · 2026-10-21
- HSBC says CoreWeave shares to drop 30%+ as ROIC of 8.9% lags borrowing costs — GaryMarcus · 2026-10-21