[News] DeepSeek Reportedly Expands Huawei Chips to AI Training; New Supply Could Arrive in 4Q26

DeepSeek is reportedly looking to increase its use of Chinese-made AI chips. According to Commercial Times, citing The Information, founder Liang Wenfeng recently told investors at a closed-door meeting that the company plans to expand its use of domestic chips, including Huawei’s, for AI model training. Huawei could deliver a new batch of AI training chips as early as 4Q26. While the scale of Huawei’s supply remains uncertain, Liang said he believes its chips could catch up with NVIDIA’s within a few years.

As noted by Chosun Biz, Chinese AI companies have previously used domestic chips, including Huawei’s, mainly for inference, which requires less computing power, while relying largely on NVIDIA’s advanced chips for large-scale model training. Huawei also revealed in April that its chips had been partially used to train DeepSeek’s lightweight V4-Flash model, the report adds.

DeepSeek is currently training a 2-trillion-parameter model and plans to eventually develop an 8-trillion-parameter model. Larger models can learn more information and complex patterns but also require substantially more computing resources and AI chips, Chosun Biz notes. For comparison, Alibaba’s Qwen 3.8 Max has 2.4 trillion parameters, while Moonshot AI’s Kimi K3 has 2.8 trillion.

DeepSeek’s Compute Push Puts Huawei Chips in Focus

Liang estimated that training a model comparable in scale to OpenAI’s largest model under development would require around 50,000 NVIDIA GB300 chips or roughly 200,000 Huawei Ascend 950 chips, Chosun Biz highlights.

Notably, Huawei’s AI chip supply reportedly remains constrained amid strong domestic demand. According to Tom’s Hardware, the company plans to offer its next-generation Ascend 900-series AI accelerators only in China rather than internationally. The report adds that SMIC’s limited production capacity for 7nm- and 6nm-class nodes is restricting Huawei’s AI hardware supply.

Despite these supply constraints, Huawei is moving ahead with an accelerated product roadmap. Securities Times notes that the company has brought forward the Ascend 960DT to 1Q27, three quarters earlier than previously planned, while the Ascend 960PR is scheduled for 3Q27.

Earlier this month, Bloomberg reported that DeepSeek planned to deploy at least 160,000 Huawei Ascend 950DT chips at a data center in Inner Mongolia. At the time, the chips were expected to be used primarily to run AI models rather than for training.

DeepSeek is investing approximately RMB 30 billion to expand its computing capacity. The company is also finalizing a second investment round worth RMB 50 billion, targeting a valuation of RMB 500 billion, according to Chosun Biz.

Read more

• [News] Huawei Speeds Up AI Chip Roadmap, Reportedly Pulls Ascend 960DT Forward Three Quarters to 1Q27
• [News] China AI Model Developers Ramp Up In-house Chips: DeepSeek, Zhipu Explore Custom Silicon

(Photo credit: Huawei)