Huawei Predicts A 100,000-Fold Surge In AI Use
Huawei predicts that global AI token consumption will grow 100,000-fold by 2035, with autonomous agents generating over 90 per cent of the associated traffic, raising questions about the infrastructure, energy and oversight needed to turn that activity into useful work.
What Has Huawei Announced?
The Chinese technology company published two reports on 16 September, ahead of its Huawei Connect conference, setting out how it expects AI to develop and what would need to be built to support it.
Intelligent World 2035 focuses on the technology, while the Global Digitalization and Intelligence Index 2026, developed with Tsinghua University’s Institute of Economics, examines the economic potential. The latter forecasts more than US$27 trillion in cumulative AI economic value over five years, with annual investment in digital and intelligent infrastructure exceeding US$4 trillion by 2030.
It should be noted here that these are just projections, rather than measured outcomes, and Huawei has a commercial interest in their underlying argument because it supplies computing, networking, storage and power infrastructure. Its forecasts therefore offer insight into the future it is preparing to serve.
What Does Token Consumption Mean?
Tokens are the units into which AI systems divide information for processing, including words, parts of words and representations of other content. They cover what a model receives and generates, so they are not simply a count of the words someone sees in an answer.
For example, an AI assistant summarising a report processes the document and produces a response. An agent investigating a business problem might repeatedly read information, plan its next step, search databases and review results before delivering anything to its user.
That longer chain of activity explains Huawei’s expectation of much heavier demand. As David Wang, Huawei’s deputy chairman and rotating chairman, puts it, “Agentic AI is a key variable in this transformation.”
Why Would Agents Generate So Much Traffic?
Unlike a chatbot that mainly responds to individual questions, an agent can work through a task using software tools and information from different sources. Several agents may also divide the work between them, creating further exchanges as they compare findings or pass instructions.
There is already evidence that this approach requires more processing. For example, in a June 2025 account of its research system, Anthropic reported that agents typically used about four times as many tokens as chat interactions, while systems using multiple agents consumed about fifteen times as many.
Those figures describe Anthropic’s experience, rather than proving Huawei’s global forecast, but they still illustrate why widespread agent use could increase demand considerably. They also show why businesses need to assess whether the work completed justifies the resources consumed.
What Would Need To Change?
Huawei identifies ten areas requiring development, including computing clusters, memory, connectivity, chip design, autonomous systems and security. Its proposals include increasing computing cluster scale 100-fold and reducing the cost of completing an agent task 1,000-fold.
The connection between those ambitions matters because widespread automation becomes more affordable if each task costs substantially less. However, cheaper processing could also encourage businesses to run many more tasks, so falling unit costs would not necessarily mean lower overall spending.
Huawei also highlights the need for reliable memory and traceable information, recognising that agents need more than processing power. As Wang states, “The direction is clear, but bringing our vision to life demands concrete action.”
Can Energy Supplies Keep Pace?
More AI activity creates demands on electricity supplies and cooling, but a 100,000-fold increase in tokens would not automatically mean the same increase in energy consumption. The relationship depends on the models used, hardware efficiency and how effectively systems share resources.
The International Energy Agency’s 2025 Energy and AI report projected that global data centre electricity consumption would more than double to around 945 terawatt-hours by 2030. It also warned that grid constraints could delay projects, showing why infrastructure availability will influence how quickly AI expands.
What Does This Mean For Your Business?
For businesses, the useful starting point is the cost of completing a job successfully, including checking and correcting the result. An agent that processes thousands of documents may deliver considerable value, but one that repeatedly searches the same material or pursues irrelevant tasks can generate costs without improving the outcome. Trials should therefore measure accuracy, completion time and staff effort as well as usage.
Spending controls also become more important when software can keep working without another instruction. Businesses should establish limits on budgets, running time and access to paid services, with clear arrangements for stopping unsuccessful tasks or seeking human approval. Suppliers should explain how customers can see what agents have done and identify unexpected increases in consumption.
The US$27 trillion forecast describes a potential global opportunity, but individual organisations will benefit only where AI improves their own operations. Better data, suitable tasks and employees who can judge the results remain essential investments, whatever happens to token volumes. More automated activity becomes valuable when it produces dependable services, useful decisions or work that would otherwise be difficult to complete.
