On August 20, 2026, Nvidia denied a report by The Information claiming it plans to ship a Groq-licensed language processing unit (LPU) to China by the end of 2026. Nvidia stated it has no China-specific LPU on its roadmap. The report highlights Nvidia's struggle to navigate strict U.S. export controls while competing with domestic rivals like Huawei Technologies, which is preparing its Ascend 950DT chip amid severe computing capacity shortages in China.
Nvidia denies China LPU report
- ▪Nvidia on August 20, 2026 denied a report by The Information claiming that Nvidia plans to begin shipping a language processing unit tailored for Chinese customers by the end of 2026.
- ▪The Information reported on August 20, 2026 that Nvidia planned to begin small-batch shipments of an artificial-intelligence chip designed for Chinese customers by the end of 2026, with several customers already placing orders.
- ▪An Nvidia spokesperson stated on August 20, 2026 that Nvidia has no language processing unit sales in China and no China-specific language processing unit product on its roadmap.
- ▪In March 2026, Nvidia Chief Executive Jensen Huang denied a previous report that Nvidia was preparing a China-compatible version of AI chips, calling the report totally false.
Groq LPU technology details
- ▪The reported language processing unit was designed as a decode co-processor for Nvidia's Vera Rubin platform, which cannot be sold in China due to United States export restrictions.
- ▪The reported Nvidia chip was a version of Nvidia's language processing unit, developed using technology licensed from startup Groq, designed to work alongside graphics processing units to speed up AI chatbot responses.
- ▪The Groq 3 language processing unit is built on Samsung's 4nm process with 512 megabytes of SRAM per die and no High Bandwidth Memory, with shipments scheduled to begin in Q3 2026.
US export control compliance
- ▪ByteDance and Tencent each recently took delivery of approximately 10,000 Nvidia H200 chips, representing the first meaningful volume of Nvidia accelerators to enter mainland China since December 2025.
- ▪The United States in May 2026 approved sales of Nvidia's H200 AI chips to a limited group of Chinese firms, including Alibaba, Tencent, and ByteDance, with deliveries beginning recently.
- ▪U.S. export thresholds for China are set on compute density and bandwidth, allowing an SRAM-only decode accelerator with no High Bandwidth Memory stack to be exported without a cut-down SKU.
- ▪The Information reported that Nvidia modified software to allow its language processing unit to operate with processors available in China after its next-generation Vera Rubin AI system became restricted under U.S. export controls.
China AI chip shortage
- ▪Chinese artificial-intelligence companies face severe computing bottlenecks, leading Moonshot AI to freeze new sign-ups and DeepSeek to raise developer fees.
- ▪It remains unclear whether Beijing will allow Nvidia's language processing unit orders to proceed, as Chinese officials previously blocked purchases of Nvidia's H20 chip in 2025.
Huawei competitive pressure
- ▪Huawei Technologies is preparing to launch its Ascend 950DT chip, optimized for decode and training, in Q4 2026, while its prefill-focused 950PR chip has been in production since April 2026.
- ▪Nvidia Chief Executive Jensen Huang stated in May 2026 that Nvidia had largely conceded China's artificial-intelligence chip market to Huawei Technologies, which has emerged as Nvidia's main domestic rival.
Story comments
Loading comments…