Huawei: AI Token Use to Surge 100,000x by 2035
▼ Summary
– Huawei released two reports ahead of Huawei Connect 2026, outlining ten strategic directions for the future of digital intelligence.
– The company forecasts a 100,000-fold increase in global token consumption by 2035, driven primarily by agentic AI rather than standard chatbots.
– Key technological shifts include moving from application-centric models to agent-centric systems that perceive, reason, and act autonomously.
– Huawei plans to scale computing clusters 100-fold while reducing agent task costs by 1,000-fold through innovations like SuperPoD systems and the Tau Scaling Law.
– Chinese policy is aligning with this transition, shifting industry focus from model training to inference and embodied intelligence.
“Visions are painted in words, but measured in deeds,” David Wang declared in the foreword to Huawei’s latest strategic forecast. Just prior to the opening of Huawei Connect 2026, the tech giant released two pivotal documents that outline the trajectory for the next decade. The first, titled Intelligent World 2035: Turning Vision into Action, serves as a roadmap for technological evolution. The second is the Global Digitalization and Intelligence Index 2026, a collaborative effort with the Institute of Economics at Tsinghua University that evaluates national progress across ninety countries.
While Huawei has published this series for three years, the current edition shifts focus from identifying megatrends to defining concrete architectural directions. “Agentic AI is a key variable in this transformation,” Wang noted. “The direction is clear, but bringing our vision to life demands concrete action.” This statement underscores the central premise of the report: the industry must move beyond theoretical models and build the physical infrastructure required to support them.
The Exponential Rise of Agentic AI
At the heart of Huawei’s projection is a staggering prediction regarding data consumption. The company forecasts that global annual token consumption will increase by a factor of 100,000 by 2035. Crucially, Huawei expects that agentic AI systems will generate more than 90% of this traffic. This figure highlights a fundamental distinction between traditional chatbots and autonomous agents. While a standard assistant responds only when prompted, an agent perceives its environment, reasons through problems, plans actions, utilizes external tools, and continuously learns. This autonomy creates a significantly heavier computational workload.
This forecast is not merely about software efficiency; it is a statement about energy and resource allocation. Each token produced requires power, memory bandwidth, and network capacity. Huawei describes a transition from an application-centric digital world to an agent-centric one, where software interacts with humans and tools continuously. This shift aligns with recent policy changes in China, where a state report in September indicated that the domestic AI industry is moving from training large models to focusing on inference, thereby shifting the primary compute burden.
Ten Strategic Directions and Infrastructure Needs
The report outlines ten specific directions that function less like abstract visions and more like a detailed product roadmap. The broadest argument posits that achieving Artificial General Intelligence (AGI) requires integrating language, embodied, and scientific intelligence. Consequently, Huawei envisions vehicles as mobile embodied agents and computing clusters scaling up by 100 times while the cost per agent task drops by 1,000 times. To achieve this efficiency, the industry must adopt SuperPoD-style systems, a hardware architecture Huawei heavily promoted during its recent conference.
Other directions address critical bottlenecks in current technology. Huawei advocates for storage and memory systems that offer causal accuracy and traceable provenance, supporting the launch of its context memory storage cluster in Shanghai. It also introduces the Tau Scaling Law, a chip design methodology unveiled in May that replaces geometric scaling with time scaling. This approach is vital for a company facing restrictions on accessing the newest lithography equipment. Additionally, the report calls for breaking through power and thermal limits in data centers and establishing robust security protocols for autonomous agents. An Agent OS is proposed, defined by a formula combining coordination, execution, memory, and connectivity, all driven by evolutionary improvement.
Economic Impact and Production Factors
The economic implications of this shift are substantial. The joint report with Tsinghua University predicts that cumulative AI economic value will exceed $27 trillion over the next five years. Global investment in digital and intelligent infrastructure is expected to grow at a compound annual rate of 19.14% surpassing $4 trillion by 2030. The index categorizes nations into three stages: Builder, Adopter, and Frontrunner.
A significant aspect of the report is its redefinition of core production factors for the digital economy. Huawei argues that data, ICT talent, and digital intelligent technologies are now the primary drivers of value. Furthermore, the integration of networking, computing, and storage is presented as essential for industrial upgrading. Given that Huawei manufactures all three components, this framing reinforces its position as a comprehensive solution provider in the emerging agent economy.
Interpreting Vendor Forecasts and European Context
It is important to view these projections with an understanding of vendor interests. A 100,000-fold increase in token consumption necessitates massive new infrastructure, which Huawei produces. Similar caution should apply to forecasts from American vendors, as methodologies behind such exponential growth claims often lack rigorous scrutiny. However, vendor reports are valuable for revealing strategic priorities and constraints. For Huawei, the emphasis on the Tau Scaling Law reflects sanctions limitations, while the focus on power and thermal management addresses grid constraints. The push for cluster scaling aims to compensate for per-chip performance gaps compared to competitors.
Europe currently faces a deficit in computing capacity, a concern highlighted by ECB President Christine Lagarde, who stated that Europe must build its own AI capacity to avoid being cut off from critical technology. Huawei is positioning itself to fill this gap, with its latest AI cluster service launching outside China on November 30 and its enterprise agent platform following on December 30. The report concludes with five commitments: taking a long-term approach, staying people-centric, upholding tech for good, remaining open, and thinking in systems. While these pledges are common among large vendors, Huawei distinguishes itself by publishing a decade-long forecast while simultaneously manufacturing most of the hardware required to realize it.
(Source: The Next Web)
