Via dwglogo.com
The chipmaker is building a Groq-licensed language processing unit tailored to clear US export controls while competing in China's booming inference market.
Nvidia is preparing to ship a new AI chip specifically engineered for the Chinese market by the end of this year, according to a report from The Information. The chip, based on language processing unit technology licensed from Groq, represents Nvidia’s latest attempt to thread the needle between US export restrictions and the massive demand for AI hardware in China.
Chinese companies have already placed orders for the chip, though one significant hurdle remains: Chinese regulators haven’t approved it for import yet.
What Nvidia is actually building #
The new chip isn’t another GPU. It’s an LPU variant, a type of processor optimized for AI inference workloads rather than training. Nvidia designed this LPU to work alongside its existing GPUs, not replace them. The chip targets use cases like chatbot interactions and other deployed AI services.
Nvidia has already shipped compliant chips to China before. The H20, a deliberately hobbled version of its more powerful H-series processors, was designed to meet US export control thresholds while still being useful enough for Chinese buyers to want it. This new LPU follows the same playbook: build something that satisfies Washington’s restrictions while delivering enough performance to justify a purchase order from Beijing.
The export control maze #
The Biden administration tightened export controls on advanced AI chips significantly, limiting the computational power of processors that could be sold to Chinese entities. The Trump administration has since adjusted the approach, moving toward case-by-case approvals for certain advanced chips rather than blanket restrictions. Limited exports of the H200, a more capable chip than the H20, have been permitted under this framework.
Even if the new LPU clears US export controls, Chinese authorities still need to approve its import. Nvidia could find itself in the unusual position of having a product that both governments are scrutinizing from opposite directions.
Why inference is the real prize #
Nvidia’s dominance in AI training hardware is well established. But inference, the process of running trained models at scale for end users, is a different and arguably larger long-term market. Training happens once (or periodically). Inference happens millions of times per day.
Chinese companies like Baidu are building their own inference capabilities, creating competitive pressure that Nvidia can’t ignore. Shipping an inference-optimized chip into China isn’t just about revenue. It’s about preventing Chinese firms from developing entirely domestic inference stacks that cut Nvidia out of the equation permanently.
The planned shipment timeline of year-end 2026 gives Nvidia roughly four months to secure the necessary approvals.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our