Huawei’s 2035 AI-Agent Forecast and Ten Technology Directions

Huawei’s vision for 2035 is a digital world organized around AI agents that perceive, reason, use tools, and interact continuously. On September 16, 2026, the company released Intelligent World 2035: Turning Vision into Action alongside the Global Digitalization and Intelligence Index 2026 (GDII 2026). The first sets out ten technology directions; the second addresses pathways toward an intelligent economy. 1017

The forecast behind the roadmap

Huawei envisions 900 billion AI agents serving a global population of nine billion by 2035. It also projects that global annual token consumption will grow 100,000-fold, with agents responsible for more than 90% of AI token traffic. The 900-billion figure also appears in Huawei’s earlier Intelligent World 2035 outlook; it should not be read as a count of agents already deployed. 7916

Those estimates explain the emphasis on infrastructure. Agents that repeatedly reason, retrieve information, call tools, and communicate could place different demands on computing, storage, and networks than a single user prompt. The scale of those demands, however, depends on how agents are built and used; Huawei’s projections are a scenario to evaluate, not a measured requirement for every operator. 69

Huawei’s ten technology directions

1. Physical-world AI and AGI. Advance language, embodied, and scientific intelligence to support AI with broader capabilities beyond purely digital tasks. 8
2. Larger computing clusters. Pursue SuperPoD-style scaling, with ambitions to increase cluster scale 100-fold while reducing the cost of an agent task 1,000-fold. These are proposed targets, not demonstrated outcomes. 818
3. Trustworthy storage and memory. Develop systems that can turn data into usable knowledge and memory while preserving causal accuracy and traceability. 9
4. Intelligent connectivity. Bring AI into communications systems to support the connections and experiences an agent-centered world would require. 9
5. Chip design through Tau Scaling. Explore what Huawei calls a shift from geometric scaling toward time scaling, with a focus on latency and energy efficiency. 817
6. Agent OS. Build an agent-native foundation for coordination, execution, memory, and connectivity. 917
7. Device intelligence. Extend intelligent services across devices rather than confining them to one endpoint. 1017
8. Intelligent driving. Treat vehicles as a setting for embodied, increasingly autonomous AI. 410
9. AI data-center efficiency. Address the power and cooling needs associated with expanded AI infrastructure. 10
10. Agent security and privacy. Develop controls for software that can act with greater autonomy, rather than treating security as an afterthought. 110

Together, the directions describe an interdependent system: more capable agents would need affordable computation, persistent memory, responsive connections, energy to run the infrastructure, and safeguards around autonomous action. That is the strategy Huawei is proposing—not evidence that one architecture will fit every deployment. 810

How operators should use the roadmap

Start with workloads, not the headline agent count. Estimate traffic and latency under plausible levels of agent use; benchmark cost per completed task rather than cluster size alone; and check whether power, cooling, memory, and network constraints change the result. Evaluate how agents retain trustworthy information and how their access and actions can be controlled before committing to an architecture. Huawei’s 100-fold cluster-scale and 1,000-fold task-cost ambitions are useful questions for testing a plan, not ready-made purchasing specifications. 89