Technology & AIAnalysis
China Launches DeepLink Computing Platform to Unify Heterogeneous AI Chips
Guoxin Shusuan unveils an integrated platform and service standards aimed at bridging fragmented domestic computing resources and supporting national network strategy.
Share
Photo by Matheus Bertelli on Pexels
The Brief
Guoxin Shusuan has officially launched its DeepLink Integrated Computing Power Network Platform alongside a new computing service cost standard at the WAIC Pujiang Ecological Forum. Operating as the commercialization vehicle for the Shanghai AI Laboratory's DeepLink open computing architecture, the platform unifies and dispatches heterogeneous domestic chips to overcome resource fragmentation. With a testbed deployed in the Beijing-Tianjin-Hebei region in late June 2026 and further expansion planned for the Yangtze River Delta, the initiative aligns with China's policy to integrate computing networks alongside national power and water grids.
Why it matters
Efficient scheduling and unified management of heterogeneous computing resources are vital for sustaining China's artificial intelligence ecosystem. By addressing the 'island effect' among disparate domestic chips, the DeepLink platform aims to maximize cluster performance and reduce AI infrastructure costs for small and medium-sized enterprises.
China context
In April 2026, Chinese policymakers formally elevated computing power networks into the nation's strategic 'Six Networks' framework alongside water management systems and new power grids. This policy push seeks to optimize cross-regional computing dispatch—such as converting western green electricity into computing capacity for eastern demand—while building an end-to-end, controllable domestic hardware ecosystem.
Editor's View
EDITOR'S VIEW — Analysis and inference, not factual reporting.
The introduction of the DeepLink platform reflects China's pragmatic response to hardware fragmentation across its domestic AI chip landscape. Rather than relying on a single dominant chip architecture, research institutions and state-backed entities are attempting to build software abstraction layers that stitch together diverse domestic processors into unified clusters. If successfully scaled beyond the initial Beijing-Tianjin-Hebei testbed, this approach could help ease software porting bottlenecks, though the long-term challenge remains maintaining low latency and high training efficiency across distributed nodes.
What to watch
- Operational performance and training efficiency metrics of the Beijing-Tianjin-Hebei testbed following its June 2026 launch.
- Expansion progress into key economic centers, particularly the upcoming nodes in the Yangtze River Delta.
- Industrial adoption and final publication of the group standard for smart computing service cost measurement.
Key Takeaways
- 1Guoxin Shusuan launched the DeepLink Integrated Computing Power Network Platform at the WAIC Pujiang Ecological Forum.
- 2The platform commercializes Shanghai AI Laboratory's DeepLink architecture to manage heterogeneous domestic chips.
- 3China incorporated computing networks into its strategic 'Six Networks' framework in April 2026.
- 4The Beijing-Tianjin-Hebei testbed launched on June 28, 2026, with Yangtze River Delta nodes to follow.
- 5Drafting began for a national group standard on smart computing service cost measurement and evaluation.
At the 2026 World Artificial Intelligence Conference (WAIC) Pujiang Ecological Forum, Chinese developers officially launched the 'Guoxin Shusuan · DeepLink Integrated Computing Power Network Platform' alongside a new computing service standard. The platform is designed to resolve resource fragmentation across domestic data centers through unified management and cross-regional scheduling of heterogeneous artificial intelligence chips.
The initiative follows the inclusion of computing power networks into China's national 'Six Networks' strategic framework in April 2026, alongside water supply systems and new power grids. Historically, localized data centers functioned as isolated computing 'islands,' resulting in regional imbalances where some facilities remained underutilized while others faced capacity shortages. Guoxin Shusuan—acting as the commercialization entity for the Shanghai AI Laboratory's DeepLink open computing architecture—serves as an underlying software layer to bridge these disparate systems. Chen Weiguo, chairman of Guoxin Shusuan, noted that the company utilizes heterogeneous adaptation and cross-domain training and inference technologies to achieve unified management and dispatching across mainstream domestic hardware.
Wang Hui, a scientist at the Shanghai AI Laboratory and chief technical advisor to Guoxin Shusuan, emphasized that DeepLink acts as a structural bridge between domestic hardware and AI applications. According to Wang, enabling collaborative training and inference across heterogeneous chips helps maximize cluster performance and accelerates the development of new token-based economic models.
To test the architecture in live environments, a Beijing-Tianjin-Hebei testbed officially went online on June 28, 2026, with plans to expand nodes into the Yangtze River Delta and other regions under a '1+3+N' ecosystem structure. Simultaneously, industry stakeholders initiated the drafting of a group standard titled 'Smart Computing Computing Service Cost Measurement and Evaluation' to establish unified cost and evaluation metrics across the sector.
Sources
- 国信数算一体化算力网平台发布--经济·科技--人民网 — People's Daily · 7/22/2026
- 163 — NetEase
- 163 — NetEase