China's AI Chip Race: DeepSeek, Domestic Chips, and What It Means Globally
DeepSeek is building its own inference chip
According to reports from July 7, 2026, China's DeepSeek is developing its own AI chip, aimed at inference — the stage where trained models generate responses to user queries — rather than training. The move is aimed at reducing dependence on Nvidia and Huawei chips, which DeepSeek has relied on to train and run its globally popular models.
A trillion-parameter model trained entirely on domestic chips
In late June 2026, Chinese tech firm Meituan launched LongCat-2.0, a large language model it says is comparable in performance to Google's Gemini 3.1 Pro. Notably, Meituan describes it as the industry's first trillion-parameter model to complete end-to-end training and inference on a 50,000-chip domestic compute cluster — a significant milestone in reducing reliance on foreign chip supply chains.
Shanghai's WAIC 2026 showcased the full domestic AI stack
The 2026 World Artificial Intelligence Conference, opening July 17 in Shanghai, is highlighting China's 'silicon-to-agent' AI stack — domestic chips, models, and agentic AI devices — signaling a coordinated push toward AI self-sufficiency across the entire technology layer, not just at the model level.
Why this matters for businesses outside China
A credible second AI hardware ecosystem, developing in parallel to the Nvidia-centric global supply chain, has implications for global chip pricing, supply chain diversification, and how quickly competitive open-source-adjacent models emerge from Chinese labs. Businesses evaluating long-term AI infrastructure strategy should watch this space — a more diversified global chip supply could eventually ease the AI infrastructure cost pressure many enterprises are currently facing.
Need help navigating what this means for your business?
Talk to our team about what's changing in AI and how it affects your roadmap — free consultation.
Get in Touch