NVIDIA Denies China-Specific Groq LPU Report As China AI Chip Battle Intensifies

NVIDIA has denied a report claiming it plans to launch a China-specific language processing unit (LPU) by the end of 2026, even as the chipmaker expands its Groq-powered inference portfolio and navigates changing US semiconductor export rules.

 

TL;DR

 

• NVIDIA denied a report that it plans to ship a China-specific LPU by the end of 2026.
• The reported processor was said to use technology licensed from Groq and target AI inference workloads.
• NVIDIA already has Groq-based inference hardware in its wider product roadmap.
• Meanwhile, Chinese foundry SMIC is benefiting from strong domestic AI-related semiconductor demand.

 

Reuters reported that NVIDIA rejected claims from The Information that the company was preparing small-batch shipments of an LPU specifically designed for Chinese customers.

“The reporting in The Information on Nvidia's LPU is incorrect. We have no LPU sales in the China market today, and no China-specific LPU product in our roadmap,” an NVIDIA spokesperson said.

The Information had reportedly cited two NVIDIA employees and claimed that several customers had already placed orders for the processor, with shipments expected before the end of the year.

The reported chip was said to draw on technology NVIDIA gained through its non-exclusive licensing agreement with Groq, which the company previously said was intended to accelerate AI inference at global scale.

 

NVIDIA Already Has Groq LPUs In Its AI Inference Strategy

 

While NVIDIA denies developing a China-specific LPU, Groq technology is already an important part of its broader inference plans.

At GTC 2026, NVIDIA unveiled the NVIDIA Vera Rubin platform, which combines Rubin GPUs, Vera CPUs, networking technology and the NVIDIA Groq 3 LPU for large-scale AI workloads.

The company also introduced the NVIDIA Groq 3 LPX inference accelerator, designed to work alongside Vera Rubin systems for workloads requiring fast token generation and predictable latency.

Each LPX rack contains 256 interconnected Groq 3 LPU accelerators. NVIDIA says each LPU provides 500MB of SRAM, 150 TB/s of SRAM bandwidth and 2.5 TB/s of scale-up bandwidth.

The company describes the architecture as being built for low-latency and high-throughput AI inference workloads, particularly increasingly demanding agentic AI systems.

NVIDIA CEO Jensen Huang also said during the company's GTC keynote that the Groq chip was already in production and that shipments were planned for the second half of 2026.

However, NVIDIA's latest denial makes an important distinction: its global Groq LPU roadmap does not currently include a product designed specifically for China.

 

US Export Rules Continue To Shape NVIDIA's China Strategy

 

The development comes as NVIDIA continues navigating US restrictions governing the sale of advanced processors to China.

In January 2026, the US Department of Commerce's Bureau of Industry and Security changed its license review policy for semiconductor exports to China, allowing applications involving NVIDIA H200, AMD MI325X and comparable products to be considered on a case-by-case basis when specific security conditions are met.

Those requirements include compliance procedures, customer screening and independent testing of eligible products in the United States.

The policy means NVIDIA H200 exports can receive individual license reviews rather than facing an automatic blanket prohibition, although approval remains subject to US government conditions.

At the same time, broader US controls on advanced computing chips continue to restrict certain high-performance semiconductor products, technologies and related activities involving China.

This regulatory environment has made NVIDIA's ability to serve Chinese AI customers increasingly dependent on which processors meet current export thresholds and which customers receive approval.

SMIC Sees AI Demand Push Revenue Above $3 Billion

 

NVIDIA's restricted position also comes as China's domestic semiconductor industry continues expanding.

Semiconductor Manufacturing International Corporation, better known as SMIC, reported second-quarter 2026 revenue of approximately $3.006 billion, up 20% sequentially. Its profit attributable to the company reached approximately $479 million, while capacity utilization rose to 93.7%.

SMIC's official second-quarter results filing also showed that the company benefited from higher wafer shipments and pricing during the quarter.

Management attributed much of the shipment growth to surging demand for chips supporting AI applications and customers pulling orders forward. The company added around 8,000 units of 12-inch-equivalent monthly capacity during the quarter.

SMIC expects third-quarter revenue to grow another 2% to 4% sequentially and has said it plans to accelerate the ramp-up of additional production capacity.

That makes the wider competitive picture increasingly important for NVIDIA. Even though the company says a China-specific LPU is not currently on its roadmap, inference remains central to its product strategy, while export restrictions and stronger Chinese semiconductor suppliers continue reshaping how NVIDIA can compete in one of the world's largest AI markets.