DeepSeek's production density reportedly hits ~$38 per GPU-hour at saturation
bookwormengr · x · 2026-10-21
The key figure in this compute-economics thread: DeepSeek's reported production density with DeepInfer plus speculative decoding (DSpark) and 96% KV-cache hit rate equals $38.00 per GPU-hour at high token saturation. The table shows three regimes: 8 concurrent requests → 2,600 tokens/s (325 tokens/s stream, interactive); 64-96 requests → 11,500 tokens/s (the $38/GPU-hr production baseline); 256 requests → 26,000 tokens/s (offline batch peak). Combined with the earlier sizing (1 GW ≈ 522K GPUs, 3.89B GPU-hours/year), this frames the theoretical revenue ceiling of large-scale inference fleets.
Related event: DeepSeek's Inference Density: $38 Revenue per GPU-Hour(3 posts)→
More from Infra
- Sizing a 1 GW datacenter: ~522,000 Blackwell GPUs fit inside — bookwormengr · 2026-10-21
- PyTorch details KTIR and Dataflow Scheduler behind IBM's Spyre accelerator backend — PyTorch · 2026-10-21
- Indie dev replaces $1k OpenAI bill with a YAML pipeline on spot GPUs and vLLM — cmm324 · 2026-10-21
- DeepSeek open-sources DeepInfer, the vLLM-based engine behind V4.1-Flash — The Decoder · 2026-10-21
- Robotics startup launches Deployments to serve fine-tuned HF checkpoints to robots in real time — arnie_hacker · 2026-10-21
- ByteDance, Alibaba, Tencent order 3M+ Nvidia RTX PRO 5500 cards worth $35B, Reuters reports — SumitGup · 2026-10-21