DeepSeek's DSec sandbox platform serves 3M sandboxes daily for agent RL training; Liang Wenfeng co-authors
teortaxesTex · x · 2026-09-25
A new DeepSeek paper details DSec, the elastic compute sandbox platform behind the company's agent RL training — with founder Liang Wenfeng among 130+ co-authors.
Key numbers:
- 3 million sandboxes served per day, peak concurrency over 380,000, creation speeds above 5,000/sec
- One production unit: 160 CPU nodes, 30,000 cores, 250TB memory, petabyte-scale images
- A single training job can spin up 32,000 sandboxes at once
All RL training and evaluation from V3.2 through V4.1 ran on DSec. The takeaway: DeepSeek is signaling an infrastructure moat behind its agent push, not just model weights — scale most labs won't replicate.
More from Infra
- Altman says OpenAI aims to replace Nvidia in the medium to long term — zephyr_z9 · 2026-11-11
- Chinese University Bought Nvidia H200 Chips Despite US Blacklist — pstAsiatech · 2026-11-11
- Baidu's ERNIE Evolve tops public bests on NanoGPT Speedrun, brings RSI to real inference optimization — xiaohu · 2026-11-11
- Baidu says RSI auto-optimized ERNIE inference, 6.24x decode speedup on Kunxin P800 — xiaohu · 2026-11-11
- Nvidia's supply commitments jump from $119B to $279B in one quarter, locking up 2027-28 memory — tengyanAI · 2026-11-11
- Train 27B models on a 24GB GPU: ternary optimizer cuts memory to 1 byte/weight — oguzeldereli · 2026-11-11