NVIDIA is trying to thread the needle when it comes to its gigantic customer base in China, attempting to balance the efficacy of its China-specific offerings with the restrictions imposed under sweeping US export controls.

Now, at a time when Beijing is actively discouraging enterprises from importing the NVIDIA H200 GPUs that the Trump administration has specifically authorized for use within China, NVIDIA is apparently trying a different route to provide value to its Chinese customer base while remaining within the bounds of US export controls, and this narrow route apparently centers on Groq's LPUs.

The Information is now reporting that NVIDIA is plotting a China comeback by designing a new inference-geared chip for China. Apparently, the new chip will be ready by the end of the year, and will sport Groq's Language Processing Units (LPUs).

For the benefit of those who might not be aware, NVIDIA entered into a technology licensing and and asset/talent transfer agreement with Groq in December 2025. Groq's LPUs cluster hundreds or even thousands of specialized chips together, where each individual chip contains giant blocks of Matrix Multiply (MXM) and Vector (VXM) units as well as around 230MB of blazing-fast SRAM. Also, AI model weights are hard-baked directly into the SRAM, completely bypassing the concept of a memory cache.

Crucially, the LPU has no branch predictors or hardware schedulers. Instead, the Groq compiler plans every single calculation down to the exact nanosecond, ensuring that relevant data arrives from the SRAM for processing in a continuous, meticulously planned operational cadence, resulting in extremely fast inferencing.

According to The Information, NVIDIA's new chip will comply with US export controls, which mainly delineate how much HBM a specific China-focused AI accelerator can support, and also tend to withhold advanced packaging technologies such as CoWoS.

Meanwhile, the US export controls are translating into good business for China's SMIC, which is the Asian giant's only 7nm-class foundry, replete with a utilization rate of 93.7 percent, latest quarterly revenue of $3.01 billion (up 36.1 percent year-over-year), and net profit nearly tripling to $479.2 million.

Also, Chinese enterprises are adopting creative workarounds to deal with US export controls. For instance, Alibaba unveiled the XuanTie C950 in March 2026, marketing the chip as a RISC-V-based offering for edge AI. Unlike typical ASICs, the XuanTie C950 does not rely on GPUs for AI workloads. Instead, the chip is basically a server-grade 64-bit RISC-V processor, replete with 64 compute cores located on a single piece of silicon, with clock frequencies that are scalable up to 3.20GHz, and where multiple clusters - 8 cores per cluster - are linked together natively using high-speed AMBA CHI fabrics. What's more, to handle demanding AI workloads, matrix and vector acceleration engines are embedded directly into the chip, eliminating the need for GPUs.

At the other end of the spectrum, DFSX's DF1000 chip is fabricated using the mature 14nm process, but sports a novel 3D near-memory compute architecture, where memory is stacked directly on top of the compute layer and connected via 3D wafer-level hybrid bonding, which melds the copper pathways of the two layers, completely eliminating microbumps or wires, and acts like tens of millions of ultra-fast vertical elevators instead of a single horizontal highway.

What's more, the DF2000 chip , which is expected to debut in Q4 2026 on the 14nm node, takes things a step further. Instead of stacking just a single memory layer on top of the compute layer, the DF2000 combines multiple stacked memory-compute towers next to each other on a base foundation in a technique that DFSX calls 3.5D Infinity Chiplet layout. The chip essentially replaces basic storage layers with a custom-engineered 3D Dynamic Random-Access Memory (DRAM) layout, dramatically increasing the quantum of temporary data that can be held directly inside its structure.

Finally, do note that a secretive China-based company, called Shanghai Aishegna , has now entered the DUV lithography business, with plans to produce 5 DUV machines this year, and another 20 in the next year. Against this rapidly evolving backdrop, it seems NVIDIA has felt the need for a new approach to try to retain its footprint within China.

NVIDIA has termed The Information's report on LPU sales in China as "incorrect." We note, however, that much of NVIDIA's statement focuses on the current status of its LPU-related plans, leaving some room for future maneuverability.

Follow Wccftech on Google to get more of our news coverage in your feeds.