While American AI labs debate whether to pull the brakes on their most dangerous models, Huawei is quietly drawing up the blueprints for an agent-saturated future — one where artificial intelligence agents will generate 100,000 times more tokens than today and account for over 90 percent of all AI traffic by 2035.
The predictions came courtesy of Huawei's newly released Intelligent World 2035: Turning Vision Into Action report, unveiled at an online event on September 16 under the banner "Act now. Create the future." The company also published the companion Global Digitalization and Intelligence Index (GDII) 2026, which estimated AI will generate more than $27 trillion in cumulative economic value over the next five years alone.
The numbers are staggering. Huawei forecasts that roughly 900 billion AI agents will be operating worldwide by 2035 — approximately 100 agents for every person on the planet. That's not a projection of helper tools or chatbots. These are autonomous systems that continuously perceive their surroundings, reason, make decisions, and call external tools without human prompting.
Why Tokens Matter
To understand the magnitude of Huawei's prediction, it helps to understand what a "token" actually is. In AI systems, tokens are the fundamental units of data that models process — whether taking in a user's prompt or generating a response. A single token can be as short as a single character or as long as an entire word, depending on the model's tokenizer.
Today's AI interactions are largely application-centric: you type a question, the model processes maybe a few thousand tokens, and spits out an answer. But Huawei's vision is fundamentally different. An AI agent doesn't just respond — it observes, reasons, plans, executes, and calls external tools. Every action generates tokens. Every observation consumes them. Every decision cycle multiplies the load.
"Token use would rise sharply because AI agents, unlike existing application-centered AI usage, continuously perceive their surroundings, carry out reasoning and decision-making, and call external tools," Huawei wrote in the report.
The implication? We're moving from an era where humans occasionally prompt AI to one where billions of autonomous agents are constantly conversing with each other, with cloud services, and with the physical world — generating tokens at a scale that makes today's ChatGPT usage look like a trickle.
The 100,000-Fold Surge
Huawei's central forecast is blunt: global annual token consumption in 2035 will be 100,000 times higher than current levels, with AI agents responsible for more than 90 percent of that traffic. For context, China already processes 140 trillion AI tokens daily — a figure that stunned observers when it was reported earlier this year. Huawei is projecting something exponentially larger.
This isn't just about bigger models. It's about a fundamentally different internet architecture. Huawei explicitly noted that an agent-centered internet environment would expand computing and network demand in a qualitatively different way from the existing application-centered internet. The shift from human-initiated queries to agent-driven continuous operations changes everything about how infrastructure needs to be designed.
The report outlined ten core technology directions to support this agent-saturated future:
- Expanding computing cluster scale by 100 times
- Cutting the cost per agent task to one-thousandth of today's levels
- Evolution from simple chip performance improvements to "SuperPoD" computing systems
- Advances in storage and memory
- Agent operating systems (OS)
- Intelligent connectivity
- Chip design
- Intelligent driving systems
- Power and cooling for AI data centers
- Security and privacy protection
The emphasis on "SuperPoD" computing systems is particularly noteworthy. Huawei has been pushing this concept since its Ascend 960 launch — a massively parallel computing architecture designed to overcome the limitations of individual chip performance and instead scale across thousands of interconnected processors.
A Contrasting Philosophy
Perhaps the most interesting aspect of Huawei's report is how it positions itself against the ongoing debate in the United States about AI safety and the pace of development. While prominent American researchers and tech executives have argued that advanced AI development should slow — citing hard-to-predict agent behavior, autonomous control of physical equipment, and resource security concerns — Huawei has taken the opposite approach.
Rather than calling for caution, Huawei is focusing on developing technologies to secure safety and controllability on the premise that AI agents will spread anyway. It's a philosophy of adaptation rather than restraint: if the agent future is inevitable, the priority should be building the infrastructure and safeguards to manage it.
This isn't abstract theorizing. Huawei is already deploying AI agents in production. In April, the company launched an autonomous platform for security operations centers (SOC), designed to detect and respond to cyber threats using monitoring and response agents. The system operates without constant human oversight — exactly the kind of continuous, autonomous operation Huawei predicts will become the norm.
Leadership Vision
Huawei's senior executives have been unequivocal about the strategic importance of AI. Guo Ping, the company's chairman of the board of directors, called AI "Huawei's biggest opportunity" and stated that the company's computing infrastructure should play a role in China similar to Nvidia's role in the United States — a remarkably direct claim about Huawei's ambitions in the AI infrastructure space.
David Wang, Huawei's vice chairman and rotating chairman, emphasized that agent-based AI would be "a key element of future change," while acknowledging that "while the direction is clear, concrete execution is needed to make it reality."
The leadership's focus on agents rather than just large language models signals a deliberate strategic pivot. While Western competitors are primarily competing on model capability — who can build the smartest single model — Huawei is betting on the ecosystem of agents that will emerge around those models and the infrastructure needed to support them.
From Software to Physical World
Huawei also positioned AI's expansion beyond software into the physical world as a key step toward artificial general intelligence (AGI). The report explained that AI interacting with the physical world must be able to perceive, judge, and act in its environment to evolve into AGI capable of performing complex tasks.
This aligns with Huawei's broader investments in intelligent driving, robotics, and industrial automation — areas where the physical and digital realms converge. The company's earlier announcement of the Ascend 960 SuperPoD was specifically designed to support these kinds of compute-intensive, physically-interactive workloads.
The GDII 2026 report analyzed the digital and intelligence level of 90 countries, providing a framework for understanding which nations are positioned to lead in the agent-driven future and which risk being left behind.
🔥 Hot Takes
1. Huawei is playing 4D chess while the West is arguing about whether to take the pieces off the board. While American AI labs are locked in existential debates about whether to continue training more powerful models, Huawei is already building the infrastructure for a world with 900 billion autonomous agents. This isn't just about being faster — it's about having a completely different relationship with the technology's trajectory.
2. The "100,000-fold" token prediction is either wildly optimistic or deliberately conservative. If Huawei's own Ascend chip roadmap and SuperPoD architecture deliver even half of what they promise, the infrastructure to support that level of token consumption becomes not just possible but economically viable. But if the compute efficiency gains don't materialize, we're looking at an energy and cooling crisis that makes today's data center boom look modest.
3. The agent-vs-application distinction is the real story here, and nobody's talking about it enough. Most coverage of AI growth focuses on model size or parameter counts. But the shift from human-initiated queries to continuously operating agents changes the fundamental economics of AI deployment. It's not just more tokens — it's a different category of computational demand that requires entirely new architecture, new security paradigms, and new business models.
The Bottom Line
Huawei's Intelligent World 2035 report is more than a forecasting exercise — it's a strategic roadmap that reveals exactly where the company intends to compete. While American AI companies race to build smarter models, Huawei is betting that the real opportunity lies in the infrastructure, connectivity, and security systems that will support a world dominated by autonomous agents.
The 100,000-fold token projection may sound like hype, but it's grounded in a specific technological vision: one where agents don't just respond to human prompts but continuously operate, interact, and make decisions on behalf of their human counterparts. Whether Huawei has the chips, the capital, and the timing to deliver on that vision remains to be seen — but the fact that they're making the bet at all says something about where they believe the industry is heading.
In a year defined by American anxiety about AI safety, Huawei's message is refreshingly pragmatic: the future is coming, and the question isn't whether to build for it but how to build it best.