In its latest forecast released ahead of Huawei Connect 2026, Huawei predicts that global AI token consumption will surge by 100,000 times by 2035, driven largely by the rise of autonomous agentic AI. This projection underpins a shift from simple AI assistants to complex, continuously learning machines requiring vast computational resources.
- Agentic AI expected to generate 90% of token traffic by 2035
- Huawei outlines ten development directions, including embodied intelligence and power-efficient AI data centers
- Forecasts $27 trillion in AI-driven economic value and $4 trillion digital infrastructure investment by 2030
What happened
Huawei published two key reports shortly before the opening of Huawei Connect 2026. These reports, Intelligent World 2035 and the Global Digitalization and Intelligence Index 2026 (prepared with Tsinghua University), lay out the company’s AI and digital infrastructure vision for the coming decade and beyond. A major forecast within these documents is the exponential increase in AI token consumption, projected to reach 100,000 times the current level by 2035.
The reports distinguish between traditional AI assistants and more advanced agentic AI, which autonomously perceives, reasons, plans, accesses tools, and continuously learns. This form of AI demands significantly more computing, memory, and network resources, highlighting the scale of future infrastructure needs and Huawei’s push toward solutions capable of supporting this new AI paradigm.
Why it matters
The forecast signals a fundamental shift in how digital services will operate, moving from application-centric systems to agent-centric ecosystems where AI agents drive continuous interaction and decision-making. This evolution implies a massive increase in compute requirements and energy consumption, underscoring the importance of innovations in data center design, chip architecture, and energy management featured in Huawei’s report.
Huawei’s outlined ten strategic directions provide a clear roadmap for navigating this transformation, including expanding computing cluster scale, reducing agent-task costs, improving storage accuracy and security, and developing agent operating systems. These initiatives also reveal Huawei’s strategic intent to lead infrastructure development in the expanding AI economy, as they coincide with China's broader shift from AI model training focus to operational inference.
What to watch next
Attention should be paid to Huawei’s advancements and commercial rollout of technologies like SuperPoD systems for scalable computing, the Tau Scaling Law for chip design, and context memory storage clusters that underpin continuous agent operation. Monitoring their progress and adoption in AI data centers will offer insight into the practical feasibility of supporting extreme token growth.
Additionally, the evolution of global AI economic value and infrastructure investment, projected to exceed $27 trillion and $4 trillion respectively within the next decade, will be crucial to track, particularly how different countries categorized as Builders, Adopters, or Frontrunners adapt to the agentic AI landscape. Progress in power efficiency, security, and privacy protections for autonomous AI agents will also be key indicators of sustainable and responsible AI ecosystem development.