SenseTime’s Galaxy Project Aims to Accelerate Domestic AI Chip Production
In an era where cutting-edge technology shapes the landscape of AI, **SenseTime** has unveiled an ambitious initiative known as the **Galaxy Project**. Collaborating with nearly 20 partners, this project aims to elevate domestic AI chip infrastructure within China, marking a significant step in the country’s technological evolution.
During his keynote, titled ‘Intelligent Transformation and Symbiosis,’ **Yang Fan**, the co-founder and president of SenseTime’s Large Device Business Group, painted a visionary picture of a self-sustaining ecosystem that seamlessly integrates chip technology, strategic partnerships, and practical implementations for domestically produced AI computing power.
A New Era for AI Infrastructure
The **Galaxy Project** encompasses not just a bold vision, but tangible collaborations, including a space computing agreement with **Guoxing Aerospace** and partnerships with five research institutions, including the renowned **Shanghai Artificial Intelligence Laboratory**. These alliances focus on advancing scientific computing applications, reflecting a holistic approach to AI development.
Responding to Emerging Trends
Yang identified three pivotal trends fueling this initiative:
- Rising **token demand** across enterprise applications.
- Increasing adoption of industrial AI, now catching up with consumer applications.
- Matured commercial viability of domestic chips, paving the way for rapid deployment of intelligent computing centers based on Chinese technology.
Nonetheless, the extent to which SenseTime’s projections hold true is contingent upon independent verification, an essential factor for enterprises considering this infrastructure.
Evaluating Throughput Figures
SenseTime asserts that its large-scale device platform processes an astonishing **2.42 trillion tokens** daily, with ambitious plans to escalate this number to **10 trillion tokens** by the end of 2026. However, businesses should view this projection as a forecast rather than a verified result until concrete data emerges.
Cost-effectiveness claims also require scrutiny, as SenseTime reports significant enhancements in model efficiency on domestic chips. They claim an 85-152% increase in **Model FLOPs Utilisation**, alongside a competitive edge in inference costs compared to leading market players.
Overcoming the Multi-chip Challenge
Historically, domestic AI chips have grappled with a fragmented software landscape, where models often require reworking across different chip architectures. SenseTime claims to have addressed this issue by developing a comprehensive adaptation layer designed to facilitate seamless workload migration across various domestic chip vendors.
For instance, their efforts in optimizing long-sequence protein prediction workloads reportedly halved prediction times, while their video generation models achieved impressive parallel acceleration.
Redefining Energy Metrics
In addition to enhancing performance, SenseTime introduced a new metric called **Tokens Per Watt**, aimed at redefining efficiency in AI data centers. This innovative approach, combined with advanced resource scheduling technology, claims to elevate token output per electricity cost significantly, suggesting a more sustainable future for AI infrastructure.
A Stellar Roster of Partnerships
The Galaxy Project boasts an impressive lineup of partners, including prominent domestic chip manufacturers and infrastructure firms. SenseTime plans on establishing not only **token factories** but also a series of computing clusters, all while supporting up to **200 AI startups**.
As Yang articulated, the initiative transcends mere chip replacement; it symbolizes a synergistic effort across China’s innovation spectrum, promising transformative advancements in technology.
Looking Ahead: Space and Quantum Computing
Beyond immediate goals, SenseTime has also set ambitious sights on optical and quantum computing, collaborating with **Guoxing Aerospace** on a project that envisions a network of computing satellites by 2030. This venture aims to expand the reach of AI services, particularly in under-connected environments.
Expanding Physical Infrastructure
With facilities spanning from **Shanghai to Riyadh**, SenseTime is making strides toward becoming a dominant force in AI infrastructure. Their Shanghai data center, noted for its **5A** intelligent computing rating, is already processing over **20 trillion tokens** daily across multiple sectors.
The company’s efforts integrate research partnerships to bolster capabilities in life sciences and manufacturing. As this initiative unfolds, it represents not only a leap in domestic AI capabilities but also aligns with China’s broader **“Artificial Intelligence+”** policy initiatives.
As we anticipate the ambitious forecast of **10 trillion tokens** per day by the close of Q4 in 2026, the journey of the Galaxy Project promises to reshape the landscape of AI. Keep an eye on SenseTime’s developments, as they could signal a transformative era in technology.
Are you ready to embrace the future of AI? Join us in witnessing these exciting developments and become part of the conversation. Your next level of innovation awaits!
