DeepSeek has rolled out a preview of its V4 artificial intelligence model built to run on Huawei chips, a move that highlights China’s growing ability to develop advanced AI systems outside the U.S. semiconductor ecosystem, and one that is likely to keep competitors, particularly in the United States, on edge.
The new model marks a shift from DeepSeek’s earlier dependence on processors from Nvidia, though the company did not specify the exact chips used for training. Instead, it emphasized close integration with Huawei’s Ascend AI systems, which are central to Beijing’s push to reduce reliance on foreign chip technology.
Huawei said the collaboration ensures broad compatibility. “The entire Ascend supernode product line now supports the DeepSeek V4 series models,” it said, signaling that the model is designed to run across a wide range of domestic high-performance computing infrastructure.
Register for Tekedia Mini-MBA edition 20 (June 8 – Sept 5, 2026).
Register for Tekedia AI in Business Masterclass.
Join Tekedia Capital Syndicate and co-invest in great global startups.
Register for Tekedia AI Lab.
The V4 is being released in preview form, allowing DeepSeek to gather user feedback before final deployment. It also includes a lower-cost “flash” version, reinforcing the company’s approach of undercutting rivals on pricing while maintaining competitive performance, a combination that helped drive its rapid rise last year.
According to DeepSeek, the pro version of V4 outperforms other open-source models on world-knowledge benchmarks, trailing only Gemini-Pro-3.1, a closed-source system from Google. That positioning, if sustained, strengthens DeepSeek’s role as a leading player in the open-source segment, where accessibility and cost efficiency are becoming decisive factors.
The release comes at a delicate moment in U.S.-China relations. A day earlier, Washington accused China of acquiring intellectual property from American AI labs “on an industrial scale,” intensifying scrutiny of companies like DeepSeek ahead of a planned summit between the two countries.
DeepSeek has been a focal point in that debate. U.S. officials have alleged it circumvented export controls to obtain advanced Nvidia chips, while OpenAI and Anthropic have said it may have improperly “distilled” their proprietary models. DeepSeek has acknowledged using Nvidia hardware but has not confirmed whether those chips were restricted, and it has said its earlier V3 model relied on naturally collected web data rather than synthetic outputs from competing systems.
China has rejected the accusations. The Chinese Embassy in Washington called them “baseless,” adding that the country places importance on protecting intellectual property.
Beyond the political friction, the V4 launch illustrates a deeper shift in how AI systems are being built. With U.S. export controls limiting access to cutting-edge chips since 2022, Chinese firms have been forced to redesign their technology stacks, pairing domestic hardware with increasingly optimized software. The collaboration between DeepSeek and Huawei reflects that adjustment, narrowing performance gaps through tighter integration rather than relying solely on raw computing power.
This approach is beginning to show results. While Huawei’s chips are still often seen as less advanced than Nvidia’s top-tier offerings, improvements in software efficiency and model design are helping offset hardware limitations. That dynamic is central to China’s effort to build a self-sustaining AI ecosystem.
DeepSeek’s rise has already unsettled the competitive industry. Its earlier models demonstrated that high-performing AI systems could be developed and deployed at significantly lower cost, challenging assumptions about the scale of investment required to compete. Each successive release has forced rivals to reassess pricing, efficiency, and deployment strategies.
The V4 preview is likely to have a similar effect. Like previous launches, it raises the bar for performance within the open-source segment while reinforcing cost pressure across the industry. For U.S.-based developers, it adds urgency to an already intense race, particularly as competition extends beyond model capability to include infrastructure, cost, and global accessibility.
The immediate market reaction in China reflected that pressure. Shares of Zhipu AI fell 9%, while MiniMax dropped 7%, suggesting investors see DeepSeek’s latest model as a competitive threat to domestic peers as well.
Interest in DeepSeek itself is growing. The company, backed by High-Flyer Capital Management, is reportedly seeking funding at a valuation above $20 billion, with Alibaba and Tencent said to be exploring potential stakes.
The broader implication is that the AI race is becoming more distributed. Rather than a single dominant ecosystem, parallel development tracks are emerging, shaped by geopolitical constraints and differing approaches to cost, openness, and infrastructure. DeepSeek’s latest release does not settle the contest, but it reinforces a pattern. Each time the company introduces a new model, it shifts expectations around what is possible with fewer resources and alternative hardware.
That pattern is likely to keep competitors, especially in the United States, recalibrating their own strategies, as the gap between the two ecosystems narrows in both capability and ambition.