Tencent Cloud details embodied AI infra: 512 domestic chips training, 90%+ GPU utilization
新智元 · wechat · 2026-11-03
While NVIDIA expanded Isaac at GTC Berlin, Tencent's 2026 Global Digital Ecosystem Conference showcased cloud AI infrastructure for embodied intelligence and launched its AgentHosting brand.
Compute: Manifold AI boosted domestic-chip training 3x in two weeks on a 512-chip cluster; Tencent's HCC migration tools moved a VLA workload to domestic chips in two weeks (+30% throughput), HyperAcc adds 22%+ acceleration, HyperTrace cut training-stop debugging 91%. Storage: a three-layer stack (COS + Data Wanxiang + GooseFS/CFS Turbo) cut data processing time 35%, lifted GPU utilization above 90%, and shortened training 30%+ for lighthouse customers; RoboScience saved tens of millions with a unified data lake and tide-computing boosted labeling 5-10x. Network: in-house HCCL yielded 40% AllReduce bandwidth gains (+13% end-to-end for one customer), and AILink grew cross-region bandwidth 5x at half the unit cost.
Tencent argues data quality and volume, not algorithms or compute, will now separate embodied AI winners.
More from Embodied
- 'Beyond Validation Loss': best pretraining checkpoint may hurt downstream RL fine-tuning — SongShuran · 2026-11-04
- Liber-0 hits 36% zero-shot success on 60 robot tasks using 2.2M hours of human data — CyberRobooo · 2026-11-04
- DISCO distills tree search into policies, a CoRL 2026 Best Paper finalist for robot planning — yining_hong · 2026-11-04
- RoboColosseum Benchmark Finds Weak Cross-Embodiment Agreement in Robot Policies — huggingface · 2026-11-04
- DexSuite: open-source dexterous manipulation framework with 10 arms, 12 hands, 30 tasks — mangahomanga · 2026-11-04
- Fruit fly's 166,000-neuron circuit map is being used to drive robot behavior — CurieuxExplorer · 2026-11-04