A new Chinese chip developer is making waves in the AI hardware sector. Hangzhou-based CL Tech has developed the "Chana" General-Purpose Tensor Processing Unit (GPTPU), claiming it delivers 1.5x the computing performance of Nvidia's A100 while consuming 30% less power and reducing unit computing costs by 42%.
This breakthrough stems from fully independent R&D. The "Chana" chip, mass-produced as early as 2023, is based on a self-developed TPU architecture with no foreign IP in its instruction set, ensuring complete technical autonomy. Unlike general-purpose GPUs like the A100, the "Chana" is specifically designed for AI/ML workloads, reportedly offering 3-5x performance gains under equivalent manufacturing processes, making it suitable for training and inference of AIGC models with hundreds of billions of parameters.
The founder's background is key: after studies at the University of Michigan and Stanford, he worked on high-performance chip design at Oracle and later joined Google's core TPU team, contributing to the architecture development from TPU v2 to v4. He returned to China in 2018 to found CL Tech.
The chip has seen real-world deployment, supporting research at Zhejiang University and being integrated into data centers for Shenzhen Unicom and Tianjin Mobile. In 2023, the company reported RMB 4.85 billion in revenue with a net profit of RMB 813.264 million, a rarity for a startup in this capital-intensive field.
ICgoodFind: CL Tech's progress demonstrates China's growing capability in high-performance AI chips, offering a viable alternative in a critical market segment.
