Chinese Startup Unveils Domestic AI Chip
Zhonghao Xinying, a Chinese artificial intelligence (AI) chip startup also known as CL Tech, has officially launched its proprietary Chana Tensor Processing Unit (TPU). The Hangzhou-based company's move marks a significant step in China's broader strategy to reduce its dependence on foreign semiconductor suppliers, particularly in the advanced AI chip sector dominated by companies like Nvidia. The Chana TPU entered mass production as early as 2023, positioning it as a direct challenger in the global AI hardware market.
Performance Claims and Strategic Independence
The Chana TPU is designed primarily for large-scale AI model training. According to Zhonghao Xinying, its flagship chip delivers up to '1.5 times the compute performance' of Nvidia's A100 tensor core graphics processing unit (GPU). Furthermore, the company claims the Chana TPU achieves this while 'cutting energy consumption by 30 per cent for equivalent large-model workloads' and reducing 'per-unit compute cost to 42 per cent of Nvidia's'. Another report indicates the chip is 50% faster than Nvidia's A100 and consumes up to 75% less power. These performance metrics, however, remain speculative without independent verification through industry-standard benchmarks like MLPerf.
A key aspect of Zhonghao Xinying's strategy is its emphasis on indigenous technology. The company asserts that the Chana TPU utilizes 'solely its own intellectual property for its basic design', avoiding reliance on Western companies, software stacks, or components for its development, design, or manufacturing. This approach aligns with China's national push for semiconductor self-sufficiency, intensified by ongoing geopolitical tensions and export controls.
Company Background and Financial Overview
Founded in 2018 by Yanggong Yifan, an electrical engineer trained at Stanford and the University of Michigan, Zhonghao Xinying benefits from leadership with significant experience in chip architecture. Yanggong previously contributed to the development of Google's TPU v2, v3, and v4, as well as working at Oracle. Co-founder and CTO Zheng Hanxun also brings experience from Oracle and Samsung Electronics' R&D center in Texas.
Financially, Zhonghao Xinying reported a revenue of 485 million yuan (approximately US$68.4 million) and a net profit of 81.3 million yuan in 2023, largely driven by Chana sales. Revenue increased to 598 million yuan in 2024, with net profit reaching 85.9 million yuan. However, the first half of 2025 saw a significant downturn, with revenue dropping to 102 million yuan and the company incurring a net loss of 144 million yuan. The company has also introduced Taize, a large-scale compute cluster capable of linking 1,024 Chana units to support training for trillion-parameter-class foundation models.
Future Outlook and Broader Implications
Zhonghao Xinying's manufacturing process is outsourced, though its foundry partners remain undisclosed. The company has announced plans to acquire the Shanghai-listed Tip Corporation and aims to go public by the end of 2026. This development is part of a broader trend in China, where significant state support and investment are accelerating efforts to develop and mass-produce homegrown AI chips, aiming for increased self-sufficiency in the semiconductor supply chain.
6 Comments
Donatello
The ambition to challenge Nvidia is commendable and could foster innovation, but relying on undisclosed outsourced manufacturing partners still leaves a dependency in the supply chain.
Leonardo
China's push for semiconductor self-sufficiency is understandable given geopolitical tensions, but the long-term impact on global tech collaboration could be detrimental.
Raphael
This is just another step towards tech decoupling and global instability.
Michelangelo
Unverified claims mean nothing. Show us the benchmarks, Zhonghao!
Donatello
It's impressive that a Chinese company is developing its own IP, yet the significant revenue drop and net loss in early 2025 raise serious questions about its sustainability.
Raphael
While the Chana TPU's performance claims are exciting for competition, independent verification is absolutely essential before we declare it an Nvidia killer.