DeepSeek hires in Inner Mongolia for V4 on Huawei chips

DeepSeek posted openings for server maintenance engineers and delivery managers in Ulanqab, Inner Mongolia as it prepares a late-April launch of V4, the company’s first major model designed to run on Huawei processors rather than U.S.-made chips. The on-site job listings are the Hangzhou firm’s first public recruitment for physical computing infrastructure ahead of the rollout and follow two postponements from an initial February target.

V4 uses a mixture-of-experts architecture with roughly 1 trillion parameters in total. For any single task the model allocates about 32 billion to 37 billion parameters to limit operating costs as the overall model scales.

Moving a large model to different accelerator hardware requires substantial engineering changes. “That transition can slow development cycles and introduce performance trade-offs, especially for V4, a model expected to be state-of-the-art,” warned Wei Sun, principal AI analyst at Counterpoint Research.

The launch comes as several U.S. AI firms said on April 6 they would share intelligence through the Frontier Model Forum to respond to efforts to copy models. Anthropic reported tracking about 16 million interactions originating from three Chinese firms across roughly 24,000 fake accounts and described those exchanges as attempts to extract model outputs for training. OpenAI accused DeepSeek in a February 12 memo to the House Select Committee on China of copying its models through obfuscated methods.

DeepSeek had earlier been expected to use restricted Nvidia accelerators without disclosing technical identifiers; the company’s shift to Huawei processors represents reliance on domestically produced silicon. The company has not published performance comparisons between V4 on Huawei hardware and versions trained on Nvidia accelerators.

China tightened export controls on advanced U.S. chips in 2022 and chip output fell 9.8% that year. Industry forecasts project domestic AI chips will capture about 50% of China’s AI chip market by 2026. Chinese semiconductor equipment suppliers increased their share of the domestic market from 25% to 35% between 2024 and 2025. Beijing has directed about $150 billion to chip development, while the U.S. CHIPS and Science Act authorized $52.7 billion in subsidies and investments.

DeepSeek’s earlier R1 reasoning model gained attention in January 2024. The company’s lower-cost tools and services are widely used in China, Southeast Asia and the Middle East.