DeepSeek's rumored $1B ARR could pay back 42 GB200 racks in under 10 months
teortaxesTex · x · 2026-10-20
teortaxesTex says DeepSeek allegedly hit $1B ARR as of Sep 24, and runs the math: at current production throughput and off-peak prices only, 42 GB200 NVL72 racks (3K GPUs) would suffice to serve it.
- At $18M per rack (an "absurd" smuggling markup), fleet payback lands around 10 months, which Liang Sheng reportedly deems fair. Attached charts model China-cost cases at $7M-$11M upfront per rack, giving 3.6-5.6 month cash payback vs 2.1 months at a $3.1M unrestricted-market rack price.
- Using the implied $38 of API revenue per inference-GPU-hour, a rack-equivalent yields $23.97M/year; after $500K/year cash operating costs per rack, central-case payback is 4.6 months.
- Sensitivity: at half the revenue productivity payback stretches to 9-10 months; in the US it could be as little as 2 months.
Related event: DeepSeek Reportedly Hits $1B ARR, Needs Only ~3,000 GPUs(4 posts)→
More from Venture
- Indie hacker GeFei's AI Video Maker hits $5,321 in 30 days, MRR $329 — gefei55 · 2026-10-20
- Indie dev's third public-revenue site TrustMRR earns $1,189 in 30 days — gefei55 · 2026-10-20
- AI agent given $25K grew a car rental business to $188K bookings in 20 days — JaynitMakwana · 2026-10-20
- Daytona, building an 'agentic cloud,' raises $60M at $1B valuation — jeremyakahn · 2026-10-20
- Postiz hits $275K MRR: tools that agents use are the fastest-growing software — alvelda · 2026-10-20
- ObsessionDB launches realtime data warehouse for agents, claiming 100X faster and 20X cheaper than Snowflake — ycombinator · 2026-10-20