SenseTime Coreware Turns Profitable with Domestic AI Chips
量子位 · wechat · 2026-07-19
During WAIC, SenseTime's Coreware disclosed progress in the commercial scaling of domestic computing power. The gross margin for domestic chip-related businesses has turned positive. Daily Token processing volume is expected to surge from 400 billion at the beginning of the year to 2.42 trillion by the end of July, with a year-end target of 10 trillion.
The article highlights its "heterogeneous hybrid inference" solution, which separates Prefill and Decode phases. High-end resources handle bandwidth bottlenecks while domestic chips manage throughput. Combined with operator rewriting, compiler adaptation, and scheduling optimizations, this approach boosts the domestic chips' MFU by 85%–152% and increases Token output per unit cost by 2.5 times.
Additionally, SenseTime integrated electricity into its joint scheduling, launching a "Compute-Power Synergy Agent." By predicting task power consumption patterns and leveraging off-peak pricing and energy storage, it improves Token output per unit of power cost by roughly 80%. The company is also partnering with various domestic chipmakers and research institutions to advance the domestic AI infrastructure ecosystem and scientific discovery platforms.
Related event: SenseTime's Domestic AI Compute Turns Profitable(2 posts)→
More from Infra
- AI Infra Meme: While Countries Need TW-Scale Power for Datacenters, People Argue About Lightbulbs — mgostIH · 2026-10-27
- Legora runs 2,000 parallel agent evals on Modal Sandboxes for legal AI — blelbach · 2026-10-27
- Cloud release velocity scores: AWS 95, Azure 90, Google Cloud 82 — DavidLinthicum · 2026-10-27
- Selling a 5090 to scale local inference: one business weighs 5 hardware paths — Anarchaotic · 2026-10-27
- AWS tutorial: fleet-wide edge CV model lifecycle with SageMaker SDK v3, ~$0.07 per run — AWS ML Blog · 2026-10-27
- Qwen3.8-27B hits 40 tok/s on Jetson Thor with INT4 + MTP, accuracy beats FP16 — No-Dragonfly6246 · 2026-10-27