The US government has given a new clarification that export restrictions on advanced AI chips apply not only to companies operating in China but also to Chinese-owned firms and subsidiaries located overseas. The move closes a potential loophole that may have allowed Chinese companies to access restricted AI hardware through operations in countries outside mainland China.

Why the US Made This Move

The clarification follows concerns in Washington that Chinese firms could obtain advanced AI processors through overseas subsidiaries despite existing export controls. Some policymakers argued that uncertainty created after the cancellation of the Biden-era AI Diffusion Framework may have opened a pathway for such purchases. The new guidance makes it clear that Chinese-headquartered companies remain subject to licensing requirements regardless of where they operate.

Pressure Grows on China’s AI Industry

The tighter enforcement is expected to make access to advanced US AI hardware more difficult for Chinese companies. NVIDIA’s most powerful Blackwell AI chips remain restricted, while purchases of other advanced processors continue to face regulatory scrutiny. As a result, Chinese technology firms are being pushed to rely more heavily on domestic alternatives.

China Accelerates Its Chip Strategy

In response to years of US restrictions, China has been rapidly expanding its domestic AI semiconductor industry. Companies including Huawei, Moore Threads, Biren Technology, Cambricon, and Alibaba have increased investments in AI processors designed for both training and deploying large AI models. Rather than focusing solely on replacing Nvidia products, many firms are now building complete AI hardware and software ecosystems.

Domestic Alternatives Gain Importance

Industry analysts expect Chinese AI companies to increasingly adopt locally developed accelerators as access to foreign chips becomes more limited. Huawei’s Ascend series and Alibaba’s custom AI processors are among the products attracting attention as China works to reduce dependence on US technology.

What to Expect Next

The latest US action is unlikely to slow China’s AI ambitions. Instead, it may further accelerate investment in domestic chips, software platforms, and AI infrastructure. The competition between the US and China is increasingly shifting from individual products to entire technology ecosystems. In the coming years, the key question may no longer be whether China can replace Nvidia, but how quickly it can build a self-sufficient AI industry capable of competing on a global scale.

Read More:

• China vs US in Robotics: Nvidia CEO Reveals the Real Power Shift
• Huawei aims to launch lithography equivalent to TSMC’s 1.4nm by 2031 in a major defiance of U.S. sanctions
• OnePlus Ace 6 Ultra vs OnePlus Ace 6: Are the Upgrades Really Worth the Extra Money?

The post US Expands AI Chip Restrictions Worldwide: What It Means for China appeared first on Gizmochina.

Chinese GPU maker Moore Threads made a major announcement this week at its MUSA 2025 Developer Conference. The company unveiled its new Huagang (or “Flowerpot”) architecture, which will power two new GPUs launching next year.

According to Moore Threads, the architecture features redesigned compute units that boost compute density by 50 percent and improve energy efficiency by about 10 percent. It also introduces a new instruction set, adds support for asynchronous programming, and improves thread scheduling efficiency, core ingredients for modern graphics and AI workloads.

Moore Threads Lushan and Huashan GPU

One of the GPUs built on this architecture is Lushan, which replaces Moore Threads’ aging MTT S80 and S90 graphics cards.

Moore Threads claims substantial performance gains across the board, including up to 15× better performance in AAA games, 50× stronger ray tracing, and 64× higher AI computing performance. 

The company also promises significant improvements in geometry processing and texture fill rates, along with a roughly fourfold increase in memory capacity from 16GB to 64GB.  

Beyond gaming, Moore Threads designed Lushan to handle professional workloads such as CAD and CAE software. The GPU also features a new “UniTE” unified rendering architecture with a dedicated AI hardware block.

The second GPU, Huashan, targets AI computing. The chip apparently uses a dual-chiplet design with 9 HBM modules, and Moore Threads directly compared it to NVIDIA’s Hopper and Blackwell GPUs.

The company claims Huashan’s floating-point performance (supports both FP4 and FP64) is close to NVIDIA’s Blackwell B200, with comparable total bandwidth and even stronger memory access capabilities. The GPU is scalable to over 100,000 units via MTLink 4.0 interconnect at 1314 GB/s.

Moore Threads says it plans to launch the first Lushan-based consumer graphics card in 2026, with Huashan products arriving around the same time.

For more daily updates, please visit our News Section.

Tech enthusiast? Get the latest news first! Follow our Telegram channel and subscribe to our free newsletter for your daily tech fix!

(Via)

The post Moore Threads announces a new GPU architecture that will power upcoming gaming and AI compute GPUs appeared first on Gizmochina.