OMM Newsroom
A Chinese-made artificial intelligence chip delivered results comparable to Nvidia’s H200 in tests conducted by its manufacturer, a potential step forward in China’s efforts to reduce reliance on foreign technology.
Shanghai-based technology company Iluvatar CoreX said its new Tiangai 300 chip matched or even surpassed the Nvidia H200, one of the most powerful AI chips currently permitted for sale to buyers in China. The company demonstrated the chip to Chinese media on Wednesday, showing its ability to perform complex AI tasks, including generating images and videos from text prompts, AnewZ reported.
In an image-generation test, the Tiangai 300 completed all tasks and produced an average of 0.101 images per second. The Nvidia H200 completed 87.5% of the tasks and generated 0.086 images per second. However, the Chinese chip took slightly longer to complete some individual tasks.
Iluvatar said the results make the Tiangai 300 the first Chinese-made general-purpose AI chip to deliver performance comparable to the H200 in real-world tests.
Test results need independent verification
The company cited other tests in which it said the chip could outperform its U.S.-made rival. According to Iluvatar, a single Tiangai 300 card performed slightly better than the H200 overall in an independent test. A separate financial report showed a 15% advantage in certain AI computing tasks.
However, the results need further independent verification because the company has not disclosed the full testing methodology.
China seeks to boost domestic chip production
The comparison is particularly significant amid U.S. export restrictions on advanced technologies. Under current U.S. export rules, the H200 is the most advanced Nvidia AI chip permitted for sale in China. After a prolonged period of supply restrictions, Nvidia began shipping its processors to Chinese data centers in August.
With U.S. restrictions making it harder to access some of the world’s most powerful chips, China is accelerating efforts to develop advanced processors domestically. Matching the H200’s performance could mark an important step toward greater technological independence.
Iluvatar is already targeting chips beyond the H200 generation. The company plans to develop processors capable of competing with Nvidia’s next-generation Rubin platform within the next two years. Iluvatar has also attracted growing investor interest since its stock-market listing in Hong Kong.
The Tongyang TY1200 processor, designed for smaller-scale AI tasks, also outperformed Nvidia’s Thor T5000 by 20% in certain tests involving large language models.
Despite the promising results, significant challenges remain. Nvidia’s advantage extends beyond hardware: developers around the world widely use the company’s CUDA software platform. Chinese companies need to build similarly powerful software ecosystems and demonstrate that they can mass-produce advanced chips.
Although Iluvatar’s claims have not yet been fully verified independently, the announced results reflect China’s ambition to compete with global chipmakers and strengthen its domestic AI industry.
