1. On January 29, Alibaba Group’s chip subsidiary T-Head announced Zhenwu 810E, a high-end artificial intelligence (AI) chip. This confirms industry speculation that Alibaba is developing its own processing power unit (PPU). PPUs are highly integrated circuits specifically designed for large-scale language model (LLM) inference, with a focus on achieving high computational performance while maintaining low energy consumption.[para. 1][para. 2]
2. The Zhenwu 810E chip incorporates both AI training and inference capabilities and is recognized as a PPU previously introduced on CCTV broadcast in September 2025. Hardware-wise, its performance falls between Nvidia’s China-improved chips, the A800 and H20. The Zhenwu 810E comes with 96GB of HBM2e memory, matching Nvidia’s H20 and beating the A800’s 80GB. Interconnect bandwidth is 700GB/s, more than the A800’s 400GB/s but slightly less than the H20. This chip uses a PCIe 5.0 interface, which is comparable to the H20 and exceeds the A800.[para. 3][para. 4]
3. Demand for Zhenwu 810E exceeded supply last year, establishing itself as one of the most shipped domestic AI accelerators. Large deployments with more than 10,000 chips per cluster serve more than 400 customers. Notable customers include China National Grid Corporation, Chinese Academy of Sciences, and XPeng Inc., demonstrating considerable market penetration and confidence in the product.[para. 5]
4. This milestone highlights the full emergence of Alibaba’s “AI Golden Triangle” consisting of Tongyi Lab, Alibaba Cloud, and T-Head. Alibaba is currently the second technology company in the world after Google to have fully integrated in-house capabilities covering LLM, cloud computing, and AI chip manufacturing.[para. 6]
5. According to domestic reports, the total number of Zhenwu PPUs shipped has reached hundreds of thousands of units. This amount exceeded that of Cambricon Technologies (688256.SH), which was previously one of China’s leading AI chip makers. Many domestic competitors such as Moore Threads (688795.SH), Metax Integrated Circuits (688802.SH), and Shanghai Biren Technology, despite having respectable financial indicators at the time of listing, face challenges in ensuring sustainable growth in the absence of steady and continuous demand from application scenarios.[para. 7][para. 8]
6. T-Head’s rapid success is closely related to Alibaba’s strong position as one of China’s top Internet companies. Alibaba Cloud serves as one of the country’s largest cloud service providers and generates vast AI computing needs, from e-commerce recommendations to intelligent customer service, providing T-Head with a reliable deployment base. T-Head was founded in 2018 after Alibaba acquired Hangzhou C-Sky Microsystems and integrated it with the company’s DAMO Academy chip team.[para. 9][para. 10]
7. Since its founding, T-Head has launched a series of successful chips, including Xuantie 910 processor, Hanguang 800 AI inference chip, and Zhenyue 510 controller. All of this is effectively deployed within Alibaba Cloud, facilitating continuous improvement and performance optimization through access to a vast user base and diverse AI applications.[para. 11]
8. A key success factor is the deep synergy between Alibaba Cloud’s full-stack technological capabilities and T-Head’s hardware, resulting in technical and commercial spillover effects and accelerated commercialization. The tight ecosystem integration seen at Alibaba is mirrored by other Chinese tech giants such as Huawei and Baidu, which benefit from the tight integration between their chips and software and cloud services.[para. 12][para. 13]
9. Internationally, the development of Google’s Tensor Processing Unit (TPU) illustrates a similar model. This is about embedding chips across cloud services and products, leveraging huge internal demand for scale and integration. T-Head follows suit, packaging chips within Alibaba’s products to create tangible commercial value from data, traffic, and close access to end users.[para. 14][para. 15]
10. Alibaba’s semiconductor approach focuses on real-world scenarios and application-driven innovation, setting it apart from traditional chipmakers, according to company executive Qi Xiaoning. The competitive environment for this sector in China increasingly prioritizes ecosystem coordination and integration over cutting-edge process nodes and theoretical peak performance. Industry experience shows that large-scale AI chip deployment depends more on stable application scenarios than raw technical capabilities.[para. 16][para. 17]
AI generated, for reference only
