DeepSeek's own numbers reveal ~82% gross margin: 155T tokens/day per cluster at $38/GPU-hour
ssh4net · x · 2026-10-21
DeepSeek quietly published revenue figures in its open-sourced DeepInfer docs: a single cluster serves 155 trillion tokens per day, generating $38 in revenue per GPU-hour at off-peak pricing. Peak API rates are $0.30/M input and $1.20/M output tokens.
Yuchen Jin did the math: even assuming rented B300s at $7/GPU-hour, that's roughly 82% gross margin — other estimates run higher. Cheap AI isn't charity or "dumping"; serving a frontier-class model is highly profitable if your inference stack is good enough.
The author argues this undercuts the narrative from US labs that intelligence must be expensive, $200/month is fair, and usage limits exist because compute is scarce — partly true, partly a business model built on users not knowing what a token actually costs. DeepInfer supports both NVIDIA Blackwell and Huawei Ascend, though the specific GPUs weren't disclosed.
Related event: DeepSeek Discloses 155T Tokens/Day per Cluster at $38/GPU-Hour Revenue(6 posts)→
More from Venture
- Indie Site Case Study: 79 Visitors in 9 Hours Converted to 7 Orders — gefei55 · 2026-10-21
- Low-margin AI businesses can work, but margin is the purest signal of moat — curious_vii · 2026-10-21
- YouTube remains the largest US streaming platform, nearing 15% of TV viewing time — vitaliychiley · 2026-10-21
- Autonomous aviation startup Reliable Robotics raises $115M, topping $1B valuation — Sethwinterroth · 2026-10-21
- Personal AI agents could threaten dating apps that profit from keeping you single — VraserX · 2026-10-21
- "98% can't burn a $20 sub": viral rant argues the AI bubble bursts on unused compute — Linahuaa · 2026-10-21