ZeroHedge's GPU ROIC math uses wrong throughput, off by 8-10x
zephyr_z9 · x · 2026-09-27
- ZeroHedge posted a GPU rental ROIC calculation that @zephyrz9 calls a top-tier sleight of hand.
- The flaw: it assumes 2,700 output tokens/s per GPU, a figure for tiny cheap models like DeepSeek v4.1 Flash.
- V4.1-class models can reach 8-10x higher output token throughput per GPU on a GB300, meaning the calculation badly understates inference economics.
More from Infra
- RAM prices up 500%: buy local AI hardware now, says analyst roundup — julianharris · 2026-09-27
- UK grid power shortfall could delay a major AI project by years — rvp · 2026-09-27
- FT: OpenAI projects negative free cash flow of $278B through 2030 on compute spending — Beth_Kindig · 2026-09-27
- IIT Delhi says it has built India's first indigenously designed micro-GPU — rvp · 2026-09-27
- Musk: China Will Solve Its Compute, Lithography and Chipmaking Constraints in 2-3 Years — haider1 · 2026-09-27
- Dev argues Vercel is 'unjustifiable' now that agents can safely drive Cloudflare — generativist · 2026-09-27