DeepSeek reportedly to deploy 160,000 Huawei Ascend 950DT chips at 1GW Inner Mongolia datacenter
Bloomberg and multiple accounts report that DeepSeek plans to buy at least 160,000 Huawei Ascend 950DT chips for a new 1GW datacenter in Inner Mongolia—potentially the largest known Huawei AI chip cluster—to run inference of its own models, while training remains on Nvidia GPUs with no current plans to switch. The report is unconfirmed but widely viewed as a landmark move by a top Chinese model lab toward domestic compute under US export controls.
Confirmed
- Bloomberg reports DeepSeek plans to deploy at least 160,000 Huawei chips at the 1GW Inner Mongolia datacenter, potentially among the largest known Huawei AI clusters, running its own models
- The chip is the Ascend 950DT; @ns123abc calls it Hopper-class; @rohanpaulai notes it is designed for decode inference and training with 144GB memory and 4TB/s bandwidth
- @heypearlai points out the 950DT is Huawei's flagship training chip, yet DeepSeek will use it only for inference; @teortaxesTex relays analyst Poe Zhao and argues the 950DT outperforms the Ascend 910 series in both training and inference
- @Polymarket notes a prediction market on a Chinese company topping Chatbot Arena by end-2026 prices only a 9% chance
Unconfirmed
- The overall report remains unconfirmed, sourced from Bloomberg and leaks
- @Hesamation notes HBM and component shortages may push delivery beyond a year; DeepSeek reportedly wants more chips than Huawei can supply
Why it matters
- If realized, this would be one of the largest domestic AI chip deployments ever, marking large-scale commercial use of the Ascend stack for inference; @pstAsiatech sees it as DeepSeek betting heavily on Huawei compute under US chip restrictions
- The split of "Ascend for inference, Nvidia for training" reflects the remaining gap of domestic chips in high-end training workloads and real constraints on Huawei capacity and the HBM supply chain
2026-09-04 ~ 2026-09-06 · 15 related posts
Primary sources
- DeepSeek plans 160,000-chip Huawei Ascend 950DT cluster in Inner Mongolia, Bloomberg reports — kimmonismus ·
- DeepSeek clusters 160,000+ Huawei Ascend 950DT chips for inference, still trains on NVIDIA — heypearlai ·
- DeepSeek wants 160,000 Huawei Ascend chips for massive Inner Mongolia data center — Hesamation ·
- [source] DeepSeek wants 160,000 Huawei Ascend chips for massive Inner Mongolia data center — Hesamation · 2026-09-04
- Hesamation adds detail on DeepSeek's 160,000-chip Huawei order — Hesamation · 2026-09-04
- [source] DeepSeek plans 160,000-chip Huawei Ascend 950DT cluster in Inner Mongolia, Bloomberg reports — kimmonismus · 2026-09-04
- DeepSeek reportedly bought 160,000 Huawei Ascend 950DT chips for 1GW Inner Mongolia data center — ns123abc · 2026-09-04
- DeepSeek to deploy at least 160,000 Huawei top accelerators at Inner Mongolia data center — pstAsiatech · 2026-09-04
- DeepSeek plans 160,000-chip Huawei cluster at gigawatt Inner Mongolia site, Bloomberg says — rohanpaul_ai · 2026-09-05
- DeepSeek to deploy at least 160,000 next-gen Huawei AI chips at massive Inner Mongolia data center — Polymarket · 2026-09-05
- DeepSeek to deploy 160,000 Huawei next-gen AI chips in Inner Mongolia data center — Polymarket · 2026-09-05
- DeepSeek readies a 160,000-chip Huawei Ascend cluster for its Inner Mongolia data center — emmanuelvivier · 2026-09-05
- DeepSeek Orders 160,000 Huawei 950DT Chips for Inference Rollouts, Not Training — teortaxesTex · 2026-09-06
5 near-duplicate retellings: zephyr_z9 · kimmonismus · ns123abc · heypearlai · rohanpaul_ai