DeepSeek Releases V4 Models That Run on Huawei Chips for Inference
BEIJING — Chinese AI company DeepSeek has released two open-source models in its V4 series that use Huawei’s domestically produced AI chips for inference, according to AI Business (https://aibusiness.com/generative-ai/deepseek-v4-models-could-change-global-ai-race).
The V4 models are open-weight and designed to operate at low cost. The release marks another entry from DeepSeek, which has emerged as one of the most closely watched AI labs outside the United States.
The use of Huawei-manufactured chips for inference is notable given ongoing US efforts to restrict China’s access to advanced semiconductor technology. Washington has imposed successive rounds of export controls aimed at limiting Beijing’s ability to develop cutting-edge AI systems. DeepSeek’s latest release raises questions about whether those restrictions can limit China’s AI progress, as the company continues to develop high-performing systems without access to Nvidia or other US-manufactured chips.
DeepSeek’s ability to ship competitive models on domestically produced chips suggests that chip-focused export controls may face limits as a standalone strategy for maintaining US advantages in AI development.
Competitive pressure on US labs
The V4 release adds to competitive pressure on leading US AI companies — including OpenAI, Anthropic and Google DeepMind — which have invested billions in proprietary model development. DeepSeek’s open-source approach, combined with its low-cost inference capabilities, presents a different business model that could attract developers and enterprises seeking alternatives to expensive API access.
The company’s earlier models demonstrated performance comparable to leading Western systems at a fraction of the reported training cost.
Policy implications
The deployment of Huawei’s AI chips in production-grade models is likely to reignite debate in Washington over the scope and enforcement of technology export controls. US policymakers have relied on chip restrictions as a primary tool for slowing China’s AI progress.
The release also raises questions for the broader open-source AI community. With both V4 models available as open-weight releases, researchers and companies worldwide will be able to build on DeepSeek’s work, potentially accelerating AI development outside the US-dominated ecosystem.
Neither the US Commerce Department nor major US AI labs had publicly commented on the V4 release at the time of publication.
