DeepSeek plans 160,000-chip Huawei Ascend 950DT cluster in Inner Mongolia, Bloomberg reports
kimmonismus · x · 2026-09-04
Bloomberg reports DeepSeek plans to deploy at least 160,000 Huawei Ascend 950DT chips in Inner Mongolia — potentially one of the largest known Huawei AI clusters — to run its models. Training still relies on Nvidia for now, with no plan to switch.
Huawei's announced specs: 144GB HBM and 4TB/s bandwidth vs. Nvidia H200's 141GB and 4.8TB/s — close on memory capacity, lower bandwidth. Bloomberg describes the 950DT as broadly comparable to Nvidia's older Hopper generation, though real-world performance remains unproven.
More from Infra
- LFM2-350M NVFP4A16 hits 1.7M tok/s decode on a single RTX 5090 — AlpinDale · 2026-09-04
- Lightning AI ships 200x faster Drive persistence, starts swapping H100s for H200e — LightningAI · 2026-09-04
- LLM inference metrics explained: what TTFT, TPS, TPOT actually measure — abhijithneil · 2026-09-04
- LLM decoding explained: prefill reads in parallel, decode writes token by token — abhijithneil · 2026-09-04
- GPU Inference Explained: Memory Bandwidth, Not Compute, Caps Tokens Per Second — abhijithneil · 2026-09-04
- LLM inference 101: bandwidth ÷ weight bytes gives your throughput ceiling before any code — abhijithneil · 2026-09-04