Realtime AI News
SenseTime Launches 'Galaxy Plan' with 20 Partners to Build 5 Exascale Domestic AI Computing Clusters
SenseTime kicked off the 'Galaxy Plan' at WAIC 2026, partnering with nearly 20 ecosystem players including Cambricon, Huawei Ascend, and Moore Threads to build five 10,000-GPU-scale domestic AI computing clusters. The company also announced a space computing partnership with Guoxing Aerospace and launched a science discovery platform with five top research institutions.
At WAIC 2026 on July 18, during SenseTime's Foundation Model Architecture Innovation and Ecosystem Cooperation Forum, co-founder Yang Fan delivered a keynote outlining SenseTime's "technology-ecosystem-business"闭环 strategy for domestic AI infrastructure. He announced the 'Galaxy Plan' in partnership with nearly 20 ecosystem partners, marking what the company calls the "first year of commercial-scale domestic AI infrastructure."
Yang noted that China's 15th Five-Year Plan period opens with transformative trends in AI infrastructure: explosive Token market growth, surging downstream demand, and accelerating commercialization of domestic chips. These factors have opened a window for large-scale domestic AI infrastructure deployment.
Galaxy Plan partners include leading domestic chipmakers Cambricon, MXHI, Hygon (HaiGuang), Huawei Ascend, Moore Threads, Sunrise, and BirenTech, as well as core component partner Xizhi Technology and infrastructure firms such as SiliconFlow, Trendjourney, and ZKAI (Zhongke Jiahe). SenseTime will jointly build five 10,000-GPU-scale domestic AI computing clusters under the initiative.
On the technology front, SenseTime has achieved 85-152% MFU improvements on mainstream domestic chips through heterogeneous hybrid inference, expanding Token output by 2.5x at the same cost. The company also introduced an electricity-computing collaborative Agent that uses a "compute x power x agent" model to reduce average electricity costs by 10% versus local IDCs while achieving 96% load forecasting accuracy.
SenseTime signed a strategic space computing agreement with Guoxing Aerospace to extend AI infrastructure from terrestrial GPU clusters to orbital satellite collaboration. By 2030, the partnership aims to build the "SenseTime Space Computing Constellation" with 1,000+ computing satellites and over 10,000 PetaFLOPS of total space-based compute power.
Additionally, SenseTime launched a science discovery platform in partnership with five top research institutions including Shanghai Artificial Intelligence Laboratory, Zhongguancun College, and Shanghai Jiao Tong University, creating an integrated research service system spanning compute infrastructure, platform tools, model capabilities, and scientific innovation.
On regional deployment, SenseTime has built the first 5A-grade computing center in Shanghai serving 20+ trillion Tokens daily. It is constructing Hong Kong's largest domestic AI computing center targeting 40,000P by 2030, and will deploy China's first overseas domestic computing cluster in Saudi Arabia.
Why it matters
The Galaxy Plan represents a pivotal shift from isolated domestic chip breakthroughs to systematic ecosystem collaboration in China's AI computing infrastructure, with space computing and science platforms signaling long-term strategic ambition.
Nearby Updates
All07/18, 14:52
IDC Pre-Releases First DAA Report at WAIC 2026: 2.2 Billion Active AI Agents by 2030
IDC has pre-released the industry's first DAA (Digital AI Agent) Research Report at WAIC 2026, forecasting that active AI agents will surge from 28.6 million in 2025 to 2.216 billion by 2030. The report introduces a standardized evaluation framework for measuring AI agent business value.
07/18, 14:50
Kimi K3 Tops BAAI's OpenCompass as World's #1 Open-Source Model
BAAI's latest OpenCompass leaderboard shows Moonshot AI's Kimi K3 has claimed the top spot globally among open-source models. The 2.8-trillion-parameter model, released July 16 at WAIC 2026, continues to draw industry attention with its million-token context window and native multimodal capabilities.
07/18, 15:20
Houmo AI Debuts M50 Inside Terminals at WAIC 2026, Powering Edge AI Computing
Houmo AI showcased multiple AI terminals powered by its M50 compute-in-memory chip at WAIC 2026, covering Agent Computers, AI PCs, and embodied robots. The M50 delivers 160 TOPS at 10W power consumption, capable of running 30B to 120B parameter models locally, marking a major step toward commercial-scale edge AI deployment.
07/18, 15:55
National AI Application Pilot Base (Finance Sector) Launches Phase Event
China's national AI application pilot base for the finance sector held its phased launch event, signaling progress toward standardized testing and verification of AI products in financial services. The facility aims to bridge the gap between AI R&D and commercial deployment in banking, insurance, and other financial domains.