DeepSeek is reportedly preparing to install at least 160,000 of Huawei’s Ascend 950DT chips at a new gigawatt-scale data centre in Ulanqab, Inner Mongolia, according to people familiar with the plan. The facility is scheduled to partially come online by late 2027 or early 2028 and would form one of the largest known clusters of Chinese-made AI chips. Neither DeepSeek nor Huawei has commented publicly on the reported initiative.
A Strategic Split Between Training and Inference
Industry observers often compare AI training to a student studying a vast library for years, while inference resembles that student answering exam questions efficiently. Because of ongoing trade restrictions, DeepSeek is splitting these tasks between hardware suppliers. Huawei designed the Ascend 950DT for training, but DeepSeek does not plan to use it for that purpose; the heavy lifting of training remains on available Nvidia hardware while inference shifts to Huawei.
Facility Details and Energy Demands
Located about 350 kilometres north-west of Beijing, Ulanqab offers cheap wind and solar energy alongside an average temperature of 4.3 degrees Celsius for natural cooling. At full load, the one-gigawatt site will consume enough power for roughly 750,000 homes. DeepSeek has historically relied on rented computing, but the company raised over $7 billion in June 2026 to fund this physical infrastructure.
Huawei’s Ascend 950DT Timetable and Revenue Outlook
Huawei plans to launch the Ascend 950DT in the fourth quarter of 2026 with its own high-bandwidth memory. Shortages of that memory are expected to hold output to the low hundreds of thousands this year, meaning fulfilling DeepSeek’s order could take more than a year. Huawei still expects its AI chip revenue to rise from $7.5 billion in 2025 to approximately $12 billion, or about KES 1.55 trillion, in 2026.
Geopolitical Pressures Reshape Supply Chains
The United States and China have traded export restrictions for four years, and each round has changed what Chinese developers can buy. Washington first required licences for Nvidia A100 and H100 sales to China, then blocked slower China-specific H800s in 2023. By late 2025 Beijing had told ByteDance and Alibaba to halt Nvidia testing and directed state-funded data centres to use Chinese chips.
Efficiency, Scale, and Open Weights
China has developed three distinct workarounds to constrained chip access. Architectural efficiency allows DeepSeek V3 to activate only 37 billion of its 671 billion parameters per word, keeping its final training run under $6 million compared with more than $100 million for OpenAI’s GPT-4. Brute-force scale and open-weight distribution further extend Huawei hardware advantages, with Chinese open-weight models capturing 41 percent of Hugging Face downloads this spring and 61 percent of tokens on OpenRouter in May.
A Bifurcated Global AI Industry
The global AI industry has split into two broad camps. The United States frontier is built on closed weights and, at OpenAI, more than 100,000 Nvidia GPUs, helping drive Nvidia to a $96.2 billion quarter. China’s frontier relies on a hybrid model combining Nvidia for training where possible, Huawei for inference, and open weights for global distribution.
DeepSeek’s reported plan signals a significant attempt to build AI computing capacity outside the reach of United States export controls while scaling inference on domestic Chinese silicon. The Ulanqab facility could become a reference point for China’s wider push toward chip self-reliance. Success, however, depends heavily on Huawei’s ability to deliver the Ascend 950DT on schedule and in sufficient volume, making the fourth quarter launch a critical milestone.
Source: Bloomberg