Reuters/Investing.com sources report that DeepSeek — the Chinese AI lab that rattled US tech markets earlier this year with its cost-efficient R1 model — is now working on developing its own proprietary AI chip. The move, if confirmed, would follow a pattern seen at US hyperscalers like Google and Amazon, which built custom silicon to reduce dependence on Nvidia.

The development matters most for Nvidia, which dominates the AI accelerator market globally, and to a lesser extent for AMD and other GPU suppliers. DeepSeek has already demonstrated an unusual ability to train competitive models at a fraction of the compute cost of US peers; custom silicon would extend that edge further and reduce its reliance on whatever chips it can source under US export restrictions.

The second-order read is ambiguous. On one hand, a domestically-designed DeepSeek chip would reduce China's addressable demand for Nvidia H20s and other export-tier products — a modest but real revenue headwind. On the other hand, DeepSeek's chip ambitions likely face years of execution risk: tape-out, yield, and software stack challenges are enormous, especially without access to TSMC's most advanced nodes.

The bull case for Nvidia is that DeepSeek custom silicon is years away from being competitive at scale, and global AI capex outside China remains on an accelerating trajectory. The bear case is that this story reinforces a secular narrative — hyperscalers and now frontier labs all want off the Nvidia dependency — that compresses long-run pricing power.

Key things to watch: any follow-on reporting naming the foundry partner (SMIC vs TSMC access), the chip's target use case (training vs inference), and whether this accelerates US export control tightening on lower-tier Nvidia SKUs.