SenseTime's DaJiangZao Achieves Profitable Domestic Compute Scale, Processing 2.42T Daily Tokens
智东西 · wechat · 2026-07-23
At WAIC 2026, SenseTime's DaJiangZao announced that its domestic AI compute has achieved profitable scale, with daily Token services reaching 2.42 trillion, projected to hit 10 trillion by year-end. Operating a 'Token refinery' model, the infrastructure shifts focus from pure compute stacking to stable Token delivery and energy efficiency.
Technologically, SenseTime boosted mainstream domestic chip utilization via heterogeneous mixed inference clusters, achieving 1.25 times the cost-performance of Nvidia's H-series. They also released a compute-power collaborative Agent, optimizing scheduling to increase Token output per unit of electricity cost by 80%.
Furthermore, SenseTime is partnering with nearly 20 domestic chipmakers to build smart compute centers and expand globally. They are also exploring optical, quantum, and space computing, planning a constellation of 1,000 compute satellites by 2030.
More from Infra
- 80% of the DIY LLM inference hype posters have already quit — it's brutally hard systems work — abhijithneil · 2026-09-11
- Hugging Face's Ultra Scale Playbook: a free book on training LLMs on GPU clusters — mdancho84 · 2026-09-11
- Is inference latency becoming the biggest bottleneck for production AI agents? — Euphoric_Sea632 · 2026-09-11
- LLM Serving Metrics Thread: Why TPOT and Uptime Make or Break User Experience — abhijithneil · 2026-09-11
- PlanetScale launches sharded Postgres: 768 servers acting as one, 1PB scale — dhruv2038 · 2026-09-11
- Can a 7900 XTX 24GB run Qwen locally? Reddit seeks ROCm tok/s benchmarks — thenomadexplorerlife · 2026-09-11