SenseTime’s Project Galaxy: Powering China’s Domestic AI Ambitions

Introduction

At a glance, SenseTime, a prominent AI software company, has unveiled its ambitious “Galaxy Project,” an initiative designed to dramatically scale China’s domestic AI chip infrastructure. Teaming up with nearly 20 partners, this project aims to create a self-sufficient and robust ecosystem for AI computing power within the country, addressing growing demand and strategic national goals.

The Vision Behind Project Galaxy

Meanwhile, Yang Fan, SenseTime’s co-founder and president of its Large Device Business Group, articulated a comprehensive vision: a “closed loop” connecting chip-level innovation, strong ecosystem partnerships, and practical commercial deployment, all centered on domestically produced AI computing capabilities. This strategic timing aligns with three key trends: a surge in token demand across enterprises, industrial AI adoption catching up to consumer use cases, and the maturation of domestic chip commercialization, enabling the rapid establishment of intelligent computing centers powered by Chinese silicon.

Ambitious Growth Projections and Important Caveats

SenseTime has put forth impressive figures regarding its large-scale device platform. The company reports processing an average of 2.42 trillion tokens daily and forecasts a remarkable 25-fold increase to 10 trillion tokens per day by Q4 2026.

Furthermore, SenseTime claims significant cost-effectiveness benefits from its heterogeneous hybrid inference technology, citing an 85–152 percent increase in Model FLOPs Utilisation on mainstream domestic chips and 1.25x the inference cost-effectiveness of Nvidia‘s H-series parts. Compared to domestic homogeneous setups, they claim a 2.5x increase in token output at equivalent cost.

However, it’s crucial for potential buyers and stakeholders to note that these figures, including token throughput and cost-effectiveness claims, are currently self-reported by SenseTime and have not been independently verified. The gap between optimized test environments and real-world production scenarios often leads to a softening of such numbers, emphasizing the need for third-party benchmarking and future quarterly performance reports.

Tackling the Domestic Chip Challenge

A historical hurdle for domestic AI chips has been a fragmented software stack, often requiring significant rework to migrate models between different architectures. SenseTime states it has developed a full-stack adaptation layer – encompassing models, frameworks, operators, toolchains, and hardware – specifically to overcome this challenge. The goal is to enable seamless workload migration across various domestic chip vendors without extensive code rewrites.

For example, As examples, SenseTime highlights a three-fold reduction in prediction time for an AI4S long-sequence protein prediction task through fused operator optimization. For AIGC video generation, they claim a 93 percent multi-card parallel acceleration ratio on domestic chips running DiT models, alongside “zero-cost migration” for mainstream AI development tools. Again, these are compelling figures that will need to prove their robustness in diverse, real-world customer pipelines.

Innovations in Energy Efficiency

Recognizing the critical importance of data center efficiency, SenseTime has introduced “Tokens Per Watt” as a new benchmark. This metric aims to provide a more holistic measure of AI data center performance. Complementing this, a “Computing-Power Collaboration Agent” has been developed to manage resource scheduling, predict electricity prices, and optimize energy storage across an intricate eight-level data system with five decision chains.

That said, Through these innovations, SenseTime asserts an 80 percent increase in token output per unit of electricity cost, average power prices 10 percent below comparable regional data centers, and an impressive 96 percent accuracy in computing load prediction. These are significant claims that warrant close observation over future operational cycles, particularly as systems encounter genuine demand volatility and seasonal fluctuations.

A Powerful Ecosystem of Partners

The Galaxy Project boasts an impressive roster of nearly 20 partners, underscoring the collaborative nature of this initiative. This ecosystem includes leading domestic chip vendors such as Cambricon, Muxi, Hygon, Huawei Ascend, Moore Threads, Sunrise, and Biren Technology. Component partner Xizhi Technology and infrastructure firms like Silicon Motion, Qujing Technology, Zhongke Jiahe, Qingcheng Jizhi, Sophon Information, and Jiliu Technology are also integral.

Interestingly, “Domestic production is not simply about replacing individual chips, but rather a collaborative effort across the entire chain of China’s innovation capabilities, from chips and components to infrastructure and application scenarios,” Yang Fan stated.

SenseTime’s plan outlines the construction of one “token factory,” five computing clusters at what it calls “10,000-calorie” scale, joint work across ten technology directions, and support for 200 AI startups.

Beyond Earth: A Glimpse into Space Computing

However, Looking further into the future, SenseTime is venturing into advanced computing frontiers, exploring optical and quantum computing for data center efficiency and AI optimization. Most strikingly, the company has partnered with satellite manufacturer Guoxing Aerospace to build the “SenseTime Space Computing Constellation.”

This ambitious plan envisions a first satellite launch in 2026, scaling up to thousands of computing satellites with tens of thousands of petabytes of computing capacity by 2030. Yang Fan highlighted the strategic value of space-based computing beyond raw power, seeing it as a means to extend Chinese AI services to weak-network environments, such as maritime operations and disaster response, and to bolster China’s AI exports internationally. This 2030 target is groundbreaking, with no direct precedent for such a large-scale satellite computing deployment.

Expanding the AI Footprint: Global and Local

Meanwhile, On the ground, SenseTime is rapidly expanding its physical infrastructure. Its Shanghai facility, touted as China’s first “5A” intelligent computing level data center, reportedly handles over 20 trillion tokens daily across more than 20 industries. A new site in Yancheng has launched with an initial 3,000 petaflops capacity, focusing on energy, manufacturing, and low-altitude economy applications.

In Hong Kong, SenseTime is developing what it describes as the territory’s largest domestic intelligent computing center, aiming for 40,000 petaflops by 2030. Furthermore, the company plans China’s first overseas domestic computing cluster in Saudi Arabia, establishing a full-stack domestic computing base for the Middle East.

In practical terms, On the research front, SenseTime’s collaboration with institutions like the Shanghai AI Laboratory and Shanghai Jiao Tong University’s AI school seeks to build a shared platform for compute, tooling, and model capabilities in fields like life sciences, materials science, and manufacturing. This aligns with China’s broader “Artificial Intelligence+” policy push, with AI for Science identified as a key driver for basic research innovation.

Conclusion: A Vision to Watch

SenseTime’s Project Galaxy represents a monumental undertaking to bolster China’s domestic AI computing power and reduce reliance on foreign technology. While the ambitious targets and reported efficiencies are impressive, the lack of independent verification for many of the claims means that the true impact and success of this initiative will unfold over the coming quarters and years.

The forecast of 10 trillion tokens per day by Q4 2026 will serve as a critical benchmark to track against the company’s future performance reports. This project is undoubtedly a significant move in the global AI landscape, positioning SenseTime and China at the forefront of domestic AI infrastructure development.

Expert Perspective

From an industry angle, the clearest signal around China domestic AI chips is how it may influence domestic. The story reads less like a one-day spike and more like a marker of broader movement.

The next phase will depend on how quickly teams, regulators, or customers react. In practice, that gives China domestic AI chips room to reshape expectations across sensetime over the near term.

For readers focused on practical impact, the best next step is to watch what changes around chip once attention turns into execution.

Frequently Asked Questions

Why does China domestic AI chips matter right now?

IntroductionAt a glance, SenseTime, a prominent AI software company, has unveiled its ambitious “Galaxy Project,” an initiative designed to dramatically scale China’s domestic AI chip infrastructure.

What broader change could China domestic AI chips signal?

Teaming up with nearly 20 partners, this project aims to create a self-sufficient and robust ecosystem for AI computing power within the country, addressing growing demand and strategic national goals.The Vision Behind Project GalaxyMeanwhile, Yang Fan, SenseTime’s co-founder and president of its Large Device Business Group, articulated a comprehensive vision: a “closed loop” connecting chip-level innovation, strong ecosystem partnerships, and practical commercial deployment, all centered on domestically produced AI computing capabilities.

What should the market watch next around China domestic AI chips?

This strategic timing aligns with three key trends: a surge in token demand across enterprises, industrial AI adoption catching up to consumer use cases, and the maturation of domestic chip commercialization, enabling the rapid establishment of intelligent computing centers powered by Chinese silicon.Ambitious Growth Projections and Important CaveatsSenseTime has put forth impressive figures regarding its large-scale device platform.

Source: https://www.artificialintelligence-news.com/news/sensetimes-galaxy-project-targets-domestic-ai-chip-scale-up/