China’s Moore Threads upgrades AI data center solution amid US restrictions

Chinese GPU startup Moore Threads has announced an upgrade to its KUAE AI data center solution, boosting the number of GPUs connectable in a single computing cluster to 10,000.

In a statement, Moore Threads described this development as a new benchmark for domestic GPU technology, aiming to achieve a significant leap in computing power in China’s AI sector.

The announcement comes amid ongoing US restrictions on China, which have significantly affected Moore Threads. In 2023, the four-year-old startup was added to the US Entities List, largely preventing it from obtaining American technologies without a special license.

The latest update represents a tenfold increase from previously announced support clusters of up to a thousand units, The Register reported, indicating efforts to scale up.

Aiming to expand in the AI market

Despite trade restrictions, Moore Threads has set global ambitions in the AI accelerator market. Last year, the company introduced its MTT S4000 chips, featuring 128 Tensor Cores, 48GB of VRAM, and a high-speed memory bandwidth of 768GBps.

Analysts point out that although Moore Threads is still behind the likes of Nvidia and AMD, the latest announcement indicates the options available to Chinese tech companies.

“While there is still a significant performance gap between Moore Threads’ offerings and those of global GPU giants, this solution enhancement provides a promising alternative for China’s burgeoning semiconductor market despite ongoing US sanctions,” said Charlie Dai, VP and principal analyst at Forrester. “Strategic partnerships among Chinese tech companies are crucial for building a local AI ecosystem and achieving technology self-reliance amid geopolitical frictions.”

Domestic and emerging markets to benefit

Being on the US Entity List restricts Moore Threads’ access to advanced process technologies. Overcoming this challenge could be difficult, but analysts suggest that the company won’t be out of options for growth.

Galen Zeng, senior research manager for semiconductors research at IDC Asia Pacific pointed out that Moore Threads can cater to the needs of the China domestic market in terms of technology and market strategy, providing special support in areas such as language processing and region-specific algorithm optimization.

“Additionally, Moore Threads can more flexibly offer customized solutions, deeply integrating and optimizing hardware and software according to the specific needs of China customers, which is uncommon among large international competitors,” Zeng added.

If they are able to develop comparable solutions, Moore Threads could also target emerging markets at a lower cost, according to Pareekh Jain, CEO of Pareekh Consulting. This is relevant because one of the biggest concerns for AI companies is the cost and availability of GPUs.

“If Moore Threads can achieve success, there is a huge domestic market that can help them build scale,” Jain said. “And if their performance claims hold true, they could further compete in many emerging markets. This could be similar to how Huawei succeeded in telecom. Western markets might not go for Chinese products, but many emerging markets in regions like Africa, Latin America, and Southeast Asia, could see them as an affordable alternative.”

More GPU news:

• Startups pursue GPU alternatives for AI
• Nvidia launches Blackwell GPU architecture
• AMD updates Instinct data center GPU line