VECTOR WIREAI INTELLIGENCE
UTC
Refresh Models Deals Regulatory Sources

DeepSeek bets next-model training on Huawei chips due Q4 2026 or Q1 2027

DeepSeek CEO Liang Wenfeng told investors that training on Huawei and domestic chips is a top priority, with chip deliveries expected Q4 2026 or Q1 2027.

DeepSeek CEO Liang Wenfeng told investors that training the company's next models on Huawei Technologies and other domestic accelerators is among DeepSeek's largest strategic priorities, according to reporting by the Information summarized on September 211,2. People familiar with the remarks said DeepSeek hopes to receive a fresh batch of Huawei Technologies training chips in the fourth quarter of 2026 or the first quarter of 2027.

The comments mark an elevation of domestic silicon from an inference-scaling story to a training-critical one. Coverage of the investor briefing frames the shift as a second pass at Ascend-class training after earlier experimentation, with Huawei Technologies deliveries described as the gating item for a larger domestic-chip training push. Liang Wenfeng's remarks position training success on domestic hardware as a must-succeed goal alongside continued access to whatever training infrastructure DeepSeek already operates.

Prior public discussion of DeepSeek's large Huawei Technologies chip orders had often emphasized inference-cluster scale. The investor briefing, as reported, shifts the emphasis: the remarks focus on the training stack DeepSeek wants for forthcoming model generations rather than on a consumer API release.

The timeline language is the actionable detail. Liang Wenfeng gave investors a Q4 2026 to Q1 2027 delivery window for the Huawei Technologies training chips. The secondary reports reviewed did not publish exact SKU names, cluster sizes, or software-stack commitments attached to that window.

Huawei Cloud announced it will begin commercial operations of its Ascend 950 series AI cluster cloud service in China on September 30 and expand to global markets on November 30[2]. ANALYSIS DeepSeek's training ambitions and Huawei Technologies' cloud commercialization timeline are converging in the same quarter, tightening the link between China's leading open-weight model lab and its leading domestic chip supplier.

The practical signal for model builders and infrastructure buyers is timeline, not specification: DeepSeek leadership is conditioning investors on a domestic-silicon training path whose start date depends on Huawei Technologies hardware arriving on schedule in late 2026 or early 2027.