DeepSeek Plans to Deploy 160,000+ Huawei Ascend 950DT Chips at Inner Mongolia Data Center
Summary
- • DeepSeek plans 160K+ Huawei Ascend 950DT chip deployment at new Inner Mongolia data center
- • Cluster would be one of the largest Huawei AI chip deployments ever assembled
- • Chips designated for inference only — DeepSeek still uses Nvidia hardware for model training
- • Full delivery may take 1+ year due to Huawei memory component supply constraints
Details
160,000+ Huawei Ascend 950DT Chips Planned
DeepSeek plans to deploy at least 160,000 Huawei Ascend 950DT AI accelerators — Huawei's top chip — creating one of the largest known Huawei chip clusters ever assembled.
Data Center Located Near Ulanqab, Inner Mongolia
The new facility is sited near Ulanqab (Wulanchabu), Inner Mongolia, a region favored for data centers due to its cooler climate and lower electricity costs.
Inference Only — Nvidia Still Used for Model Training
The 160,000+ Ascend 950DT chips are earmarked for AI inference (model serving), not training; DeepSeek continues relying on Nvidia GPUs for core model training workloads, reflecting a split hardware strategy.
Delivery Could Exceed One Year on Memory Supply Constraints
Huawei's 950DT production is constrained by high-bandwidth memory component shortages; total 2026 output is estimated in the low hundreds of thousands, meaning full order fulfillment could take more than a year.
DeepSeek in Talks to Raise Billions for Infrastructure
DeepSeek is in talks to raise billions of dollars to fund its broader infrastructure expansion, of which the Inner Mongolia data center is a key component.
China's Push to Scale AI Infrastructure on Domestic Chips
The deployment signals China's accelerating effort to build large-scale AI infrastructure using domestic chips as U.S. export controls block new high-end Nvidia GPU imports to Chinese entities.
Details Compiled via Grok Live Web Research
Additional details sourced from Bloomberg, Moneycontrol, and BlockBeats coverage of the Bloomberg-originated report, compiled via Grok Research.
Each row covers a distinct aspect of DeepSeek's planned Huawei chip deployment and China's AI infrastructure buildout.
What This Means
DeepSeek's plan to deploy more than 160,000 Huawei Ascend 950DT chips is a landmark test of whether China can build large-scale AI inference infrastructure independent of Nvidia hardware. The split strategy — Huawei chips for inference, Nvidia for training — reveals that domestic Chinese accelerators are competitive enough for serving models at scale but have not yet displaced Nvidia for the demanding workloads of frontier model training. Supply constraints at Huawei, driven by high-bandwidth memory shortages, mean the buildout will be slower than DeepSeek's ambitions, with the timeline of more than a year for full delivery reflecting real limits in China's AI chip supply chain. For the broader AI industry, this is a concrete signal that China's AI infrastructure will continue scaling on domestic silicon — with consequences for both U.S. export control strategy and Huawei's standing as a global AI hardware player.
Sentiment
Cautious optimism on China's self-reliance push, tempered by skepticism over Huawei supply constraints
“DeepSeek purchased 160,000 Huawei ascend 950DT chips for new 1 gigawatt data center in inner Mongolia. 950DT are Hopper-class so only for inference. TRAINING stays on NVIDIA”
“deepseek just ordered the biggest known huawei ai pile. 160k ascend 950dt for a gigawatt box in inner mongolia. for running models, not training — they still lean on nvidia for that part. huawei's fab can't keep up anyway. fulfillment might take over a year.”
“$DeepSeek wants 160,000 Huawei chips for a new data center in Inner Mongolia. That’s a real shift away from Nvidia on the inference side. They still train on Nvidia, though. China is building its own stack because it has to. Whether Huawei can actually deliver that many chips on time is the part I don’t buy yet.”
“Huawei is capacity constrained on its latest Ascend 950DT AI chips for now. But DeepSeek and other top Chinese AI labs are preparing for domestic AI chip production to ramp up significantly before too long.”
Responding to Bloomberg reporting on the order
Split
~60/40 split: progress on domestic inference infrastructure vs doubts on Huawei's ability to deliver at scale within a year