China's second‑generation 7nm TPU to launch next year with performance targeting NVIDIA H100
Zhonghao Xinying CEO Yang Gong Yifan has confirmed that the company's second‑generation 7 nm TPU has successfully completed tape‑out and is now undergoing testing. The chip is scheduled to begin shipping in the second quarter of 2026, marking a major step forward for China's domestic AI‑accelerator ecosystem.
According to early industry leaks, the new TPU is designed for autonomous‑driving model training and large‑scale data‑center inference workloads. Single‑chip compute performance is expected to reach 400-800 TFLOPS, placing it directly against NVIDIA's H100 and Google's TPU v5p. Energy efficiency may be more than 30 percent better than comparable GPUs, and cost per unit of compute has reportedly been optimized, though these details have not yet been officially confirmed.
Yang Gong Yifan also stated that Zhonghao Xinying plans to maintain a rapid product‑iteration cycle of "one core per year, two stacks per year" to strengthen its competitiveness and improve market responsiveness.
Founded in 2019, Zhonghao Xinying focuses on high‑performance AI chips and computing clusters for large‑scale model training. Its first‑generation Chana TPU, built on a 12 nm process, entered mass production in 2023. The chip features a fully self‑developed instruction set and IP core, delivering up to 1.5x the performance of NVIDIA's A100 while reducing power consumption by 30 percent under similar large‑model workloads.
Zhonghao Xinying is a Chinese semiconductor company specializing in AI‑accelerator hardware and large‑scale computing platforms. Established in 2019, the company has rapidly expanded its R&D capabilities, developing its own instruction sets, IP cores, and training‑cluster architectures. Its products are designed for data centers, autonomous‑driving model training, and other high‑performance AI applications. With a focus on domestic supply‑chain independence, Zhonghao Xinying aims to become a key player in China's next‑generation AI‑computing ecosystem.
