Huawei CONNECT 2026: ThinkPro Agentic Computing Framework Open-Sourced with CNY 5B Ecosystem Investment
Huawei CONNECT 2026: A New Era of Agentic Computing
In September 2026, Huawei officially announced its entry into the "Agentic Computing" era at its annual flagship technology summit, HUAWEI CONNECT 2026. This strategic transformation marks Huawei's evolution from a traditional cloud computing and AI infrastructure provider toward a platform supporting large-scale autonomous AI agent deployment.
The summit's centerpiece announcement was ThinkPro (ThinkProcess) — a new framework serving as the atomic unit for agent thinking, execution, exploration, and evolution within the openEuler operating system. Simultaneously, Huawei announced a commitment to invest CNY 5 billion over the next three years to drive collaborative innovation among universities, research institutions, and industry partners.
ThinkPro: The Foundation of Agentic Computing
ThinkPro is the technical core of Huawei's agentic computing strategy. As the fundamental atomic unit within the openEuler ecosystem, ThinkPro provides developers with atomic-level APIs spanning both user space and kernel space, abstracting state management and hardware resources (CPU and memory).
Five-Dimensional Development Transformation
Huawei proposed a five-dimensional development transformation framework for SuperPoDs and SuperClusters to address the scaling challenges of the agentic AI era:
- Infrastructure: Moving from single-server to SuperPoD-level development, supporting larger-scale agent workloads
- Operating Systems: Transitioning from process-centric runtimes to ThinkPro-based agent development environments
- Programming Style: Shifting from SIMD-only to homogeneous SIMD + SIMT (Single Instruction, Multiple Threads) and Regbase programming
- Precision: Evolving from HiF8 toward low-precision HiF4 to improve computational efficiency
- Workflow: Transitioning from manual operator programming to agent-assisted generation, adaptation, and precision tuning
This framework addresses the core challenges facing agentic AI in large-scale deployment: massive-scale operations, long-horizon task execution, increased time-to-first-token (TTFT), and growing memory demands.
Open PTO ISA: Breaking Down Operator Portability Barriers
Huawei announced the opening of the PTO (Processor Tile Optimization) Instruction Set Architecture at the summit, comprising over 120 virtual instructions across eight categories. This tile-based architecture allows operator source code to be ported across different processor generations, supporting both Auto and Manual optimization modes.
The opening of the PTO ISA holds significant importance for the AI developer community:
- Reduced Migration Costs: Developers no longer need to rewrite operators for each generation of Ascend chips
- Accelerated Ecosystem Building: More third-party developers can develop high-performance operators based on a unified instruction set
- Facilitated Academic Collaboration: Universities and research institutions can more easily conduct cutting-edge research on the Ascend platform
Agentic Tool Suite: Full-Chain Support from Development to Deployment
Huawei open-sourced a suite of tools supporting AI-native application deployment, covering the complete lifecycle of agent development:
- CANNBot: Operator generation tool, automating the development of AI operators
- Model Agent: Model adaptation tool, helping migrate existing models to the Ascend platform
- MindStudio Agent: Precision tuning tool, optimizing model numerical precision on Ascend hardware
- Solution Agent: Industrial deployment tool, supporting AI solutions in real production environments
The open-sourcing of this tool suite significantly lowers the technical barriers for developers building and deploying AI agents on the Ascend platform.
Compute Resource Support: Shared Access to 100,000-NPU Scale
To support the development of the agentic computing ecosystem, Huawei announced that the Ascend community now provides access to shared, 100,000-NPU-scale compute resources. Additionally, Huawei launched a "100 NPU-Hour Program" to provide developers with a baseline compute allocation for training and inference experiments.
This initiative is particularly important for small and medium-sized AI startups and academic research institutions, enabling them to explore and validate agentic AI applications without bearing high hardware costs.
CNY 5 Billion Ecosystem Investment: A Signal of Long-Term Commitment
Huawei committed CNY 5 billion over the next three years to drive collaborative innovation among universities, research institutions, and industry partners. This investment plan covers:
- Academic Collaboration: Supporting leading universities in conducting cutting-edge AI research on the Ascend platform
- Industry Incubation: Providing technical and financial support for AI startups based on the Ascend ecosystem
- Standards Development: Participating in the development of agentic AI technical standards to promote industry normalization
Asia-Pacific Perspective: The Strategic Significance of the Ascend Ecosystem
For AI developers and enterprises in the Asia-Pacific region, the announcements at Huawei CONNECT 2026 carry important strategic significance. Against the backdrop of U.S. export controls restricting NVIDIA GPU exports to China, the Ascend platform has become the core compute infrastructure for China's AI industry.
The opening of the ThinkPro framework and PTO ISA provides AI developers in Asia-Pacific (particularly China, Hong Kong, and Southeast Asia) with a complete agentic AI development ecosystem, helping reduce dependence on Western AI infrastructure.
For AI enterprises in Hong Kong, the maturation of the Ascend ecosystem means more compute options and lower technical barriers, helping drive innovation and deployment of local AI applications.
Conclusion: Infrastructure Competition in the Agentic Computing Era
Huawei CONNECT 2026 marks the entry of global AI infrastructure competition into a new phase. The ThinkPro framework, PTO ISA opening, and CNY 5 billion ecosystem investment collectively constitute Huawei's strategic positioning in the agentic computing era.
As AI agents move from experimental applications to large-scale production deployment, competition in underlying infrastructure will become a key factor determining the AI industry landscape. Huawei's series of announcements demonstrates its long-term strategic resolve in this competition.
