Chinese artificial intelligence startup DeepSeek is planning to deploy at least 160,000 high-end AI accelerators from Huawei Technologies at a massive data centre under construction in Inner Mongolia, according to reports citing people familiar with the matter. The strategic initiative marks a significant step in China’s broader industrial push to reduce its reliance on Nvidia Corp processors for developing and executing complex AI models.
The proposed infrastructure project would establish one of the largest publicly known computing clusters utilizing Huawei’s domestic silicon architecture. Market observers noted that the announcement coincided with a 1.4% premarket rise for Nvidia shares in New York, reflecting ongoing investor evaluation of shifts in the semiconductor supply chain.
Deployment Focus and Operational Scope
DeepSeek intends to utilize Huawei’s next-generation Ascend 950DT chips primarily to operate and run its artificial intelligence models rather than for initial model training. While Huawei originally positioned the 950DT processors for demanding training workloads, DeepSeek has opted to rely on Nvidia accelerators for the critical training phase of its frontier models, according to sources cited by Bloomberg.
The startup had previously explored training its infrastructure on Huawei silicon but maintained its dependency on established U.S. alternatives for heavy computational training tasks. The deployment of 160,000 Ascend processors at the Inner Mongolia hub represents an effort to segregate operational inference workloads onto domestic hardware while managing performance constraints.
Manufacturing Bottlenecks and Supply Constraints
Fulfilling the large-scale order presents substantial hurdles due to severe production constraints facing Huawei’s semiconductor supply chain. Industry analysts indicate that Huawei is projected to produce only a few hundred thousand Ascend 950DT chips during the current year, constrained by shortages of key components including high-end memory.
Because Huawei must balance allocation across multiple domestic clients while attempting to service limited overseas shipping channels, sources estimate that completing DeepSeek's order could require more than a year. The startup has reportedly sought assistance from central authorities in Beijing to accelerate manufacturing allocations and priority delivery schedules.
Infrastructure Scale and Capital Expansion
The Inner Mongolia facility is designed to operate at a gigawatt scale, which at full operational capacity would consume electricity equivalent to the needs of approximately 750,000 residential homes. The 160,000 Huawei processors would comprise only one segment of the massive data centre's total computing capacity.
To finance this extensive infrastructure buildout, DeepSeek has been engaged in active discussions to raise billions of dollars in new funding rounds. The company generated more than $70 million in revenue during the first seven months of 2026, bolstering its balance sheet as it expands its computational footprint.
Geopolitical Context and Competitive Landscape
A facility housing over 100,000 localized AI chips aligns DeepSeek's expansion goals with the scale of infrastructure deployed by major Western technology companies, although Chinese processors remain technically less powerful than Nvidia's leading-edge hardware. Washington's export controls prohibit the sale of advanced Nvidia accelerators to China without specific regulatory approval.
While Chinese government directives strongly encourage domestic AI laboratories to transition toward home-grown silicon, major domestic firms like Alibaba and Tencent continue to seek access to Nvidia hardware when permitted. Meanwhile, industry participants continue to monitor whether domestic alternatives can scale sufficiently to bridge the technological gap created by international trade restrictions.
