Ascend 950DT can roughly replace H100 for training but not for economical inference
teortaxesTex · x · 2026-11-18
Analyst account teortaxesTex assesses Huawei's Ascend 950DT: viable for training and some RL, and roughly a replacement for the Nvidia H100, but not economical for inference and inefficient for long agentic RL workloads. The key open question, he argues, is how many 950DTs will be produced by end of 2026, which will determine its real impact on China's AI compute landscape.
More from Infra
- Google donates Agent Substrate to CNCF: 500ms restores, 10x density — krishnan · 2026-11-18
- Report: Nvidia brokered Alibaba-Megaspeed deal, Alice Huang ties run deeper than known — GaryMarcus · 2026-11-18
- n8n's open-source agent Nod: secure remote access via private reverse proxy — victorialslocum · 2026-11-18
- pci_eerie: Open-Source PCIe Soft Stack for Xilinx 7-Series FPGAs as a Drop-in PCIE_2_1 Replacement — IgorCarron · 2026-11-18
- Hugging Face Nears 1 Exabyte of Monthly Data Distribution as Open Model Downloads Explode — andimarafioti · 2026-11-18
- NVIDIA, AMD, Intel, Google and Meta join SGLang Consortium as v1.0 ships — heyshrutimishra · 2026-11-18