DeepSeek develops in-house AI chip to reduce reliance on Nvidia amid US export controls and cost pressures.
DeepSeek is developing its own AI chip designed for inference rather than training, according to Reuters, reducing its reliance on Nvidia and Huawei hardware. The chip effort began approximately a year ago and remains in early stages. The company has been speaking with external partners, including chip-design, foundry, and memory companies, and has increased hiring of chip-design engineers in recent months, conducting recruitment privately rather than through public job postings.
The strategic rationale is clear. DeepSeek currently uses both Nvidia and Huawei chips, but US export controls have severely limited China's access to Nvidia's most advanced GPUs. The company trained its R1 foundation model on Nvidia's H800, a China-market chip subsequently banned by Washington. It has since shifted toward Huawei's Ascend platform, adapting its models accordingly. Government pressure strongly favors Chinese suppliers in supply chains.
This move aligns DeepSeek with a broader trend among major AI model companies—OpenAI, Anthropic, Alibaba, and Baidu among them—seeking greater control over their hardware stack. The motivation is not primarily performance, where US suppliers retain an edge, but supply security and control.
Cost pressures are reshaping the market. DeepSeek's January 2025 model release briefly shifted global attention toward cheaper Chinese alternatives, with Chinese models' share of global web traffic rising from roughly 3% to 13% within two months. US labs responded with their own lower-cost offerings, but the signal was unmistakable: if performance is comparable, price determines choice. Among 70,000 US firms tracked by Ramp, DeepSeek adoption grew from 0.1% to 0.3% during the first five months of 2026—still minimal but economically significant at scale.
The immediate competitive threat may be to Huawei, which benefited from Nvidia's restricted position in China but faces potential erosion if Chinese hyperscalers develop proprietary AI processors.
Designing competitive AI inference silicon is expensive and time-consuming. Manufacturing presents additional obstacles, as Chinese chip designers remain constrained by limited access to leading-edge foundries and high-bandwidth memory, both affected by US restrictions. Yet the strategy signals that China's AI leaders increasingly view hardware as central to model strategy and do not expect supply-chain pressures to diminish.
Beyond China's boundaries, this represents a broader shift toward vertically integrated AI. Lower-cost Chinese models running on domestic silicon could prove attractive in cost-constrained regions of Africa and Asia, where frontier US AI faces adoption barriers around cost, cloud access, and sovereign control. Europe faces different dynamics: recent US restrictions on access to Anthropic's most advanced models made dependency visible. EU sovereign models remain overly specialized for defense and science. Trust and autonomy—not just capability—are becoming determining factors for international adoption.
The long-term value in China may not accrue primarily to chip exporters but to IP suppliers enabling regional players to build domestic silicon. Companies such as Imagination Technologies, which supports local Chinese GPU and accelerator development without being the finished-chip supplier, may increasingly capture value in this middle layer.
Ultimately, DeepSeek illustrates where the market is heading: AI becoming more vertically integrated, model companies seeking greater hardware control, chip companies expanding into AI value chains, and IP suppliers gaining importance as enabling infrastructure.