SenseTime Co-Founder Foresees Five-Year Overhaul of China's AI Hardware Ecosystem

Deep News
Jul 21

At the recently concluded WAIC 2026 conference, Sensetime-W (HKEX: 00020) announced multiple strategic partnerships with entities including Guoxin Data, China Unicom, CATL, and Jinko Solar to foster collaborative development in AI infrastructure. The company's co-founder and head of its large-scale AI infrastructure division emphasized that domestic substitution over the next five years will extend far beyond just GPUs, representing a comprehensive restructuring of the entire upstream AI hardware ecosystem and creating new growth avenues for the industry.

The Shift to Agentic AI

As AI application models evolve from simple conversational interactions to complex agent-based tasks in the Agentic era, the explosive growth in token consumption is placing higher demands on AI infrastructure. This shift is simultaneously creating fresh opportunities for the development of domestic computing power. Sensetime-W's large-scale infrastructure is built around a full-stack "computing power + platform + services" capability, forming a system designed for the Agentic AI age.

Computing Power and Efficiency

At the computing layer, the infrastructure utilizes heterogeneous hybrid inference to boost the MFU of mainstream domestic chips by 85% to 152%, achieving an overall inference cost-performance ratio 1.25 times that of NVIDIA's H-series. The platform currently handles a daily token service volume of 2.42 trillion, with an expected 25-fold growth in annual service scale by 2026.

Regarding energy efficiency, the initiative is advancing intelligent computing centers from traditional IDC PUE management to a full-chain value management system centered on TPW (Tokens Per Watt). This approach directly links power consumption to token output, connecting the entire chain from power grids and plants to IDC cabinet services, computing cloud services, and model services, providing an efficiency metric more aligned with the essence of AI production. Through end-to-end optimization, Sensetime-W has increased token output per unit of electricity cost by 80%, achieving system-level computational efficiency gains.

Strategic Collaborations

During the conference, Sensetime-W entered a strategic partnership with Guoxin Data. The collaboration will focus on constructing a national integrated computing power network testbed, coordinating computing resources, developing scheduling platforms, and fostering the domestic AI chip ecosystem to jointly promote the national computing power network project.

Focusing on "computing-power synergy," the company also signed strategic agreements with partners including Jinko Solar, China Unicom Shanghai, Huaneng Shanghai, and CATL's Puquan Capital. These partnerships aim to build an integrated ecosystem for source-grid-load-storage synergy.

Supporting Innovation

Furthermore, Sensetime-W has jointly launched an AI technology innovation acceleration program with platforms like the Hong Kong Science Park and the Shanghai Jiao Tong University Industrial Technology Innovation Research Institute. This program is designed to provide AI startups with comprehensive support throughout their lifecycle, from R&D and computing power to market expansion and commercial implementation.

Disclaimer: Investing carries risk. This is not financial advice. The above content should not be regarded as an offer, recommendation, or solicitation on acquiring or disposing of any financial products, any associated discussions, comments, or posts by author or other users should not be considered as such either. It is solely for general information purpose only, which does not consider your own investment objectives, financial situations or needs. TTM assumes no responsibility or warranty for the accuracy and completeness of the information, investors should do their own research and may seek professional advice before investing.

Most Discussed

  1. 1
     
     
     
     
  2. 2
     
     
     
     
  3. 3
     
     
     
     
  4. 4
     
     
     
     
  5. 5
     
     
     
     
  6. 6
     
     
     
     
  7. 7
     
     
     
     
  8. 8
     
     
     
     
  9. 9
     
     
     
     
  10. 10