NVIDIA is trying to find a balance between serving its large Chinese customer base while trying to keep its products effective for the Chinese market and complying with sweeping export restrictions imposed by the United States.
The key feature of the new solution is Groq’s Language Processing Unit (LPU). NVIDIA acquired the rights to the technology and certain company assets as early as December 2025. Unlike traditional accelerators, the Groq architecture relies on large amounts of fast SRAM memory and pre-planned operations by the compiler. This allows you to minimize latency and ensure that requests to your AI model are processed very quickly.
This development is particularly important given the controversial situation surrounding the supply of U.S. accelerators to China. Although the Trump administration allowed NVIDIA H200 exports, Beijing is now calling on local companies to refuse to buy them. The new chips must take into account current restrictions in the United States, including HBM volume requirements and the availability of advanced packaging technologies such as CoWoS.
Meanwhile, China is actively developing its own alternatives. Local manufacturers are experimenting with RISC-V processors, 3D memory architectures and near-memory computing, while SMIC has benefited from growing domestic demand. Against this backdrop, it seems that Nvidia is not only trying to get around these restrictions, but also trying to adapt its products to the new realities of the Chinese market – until it completely loses its position in the Chinese market.