Huawei forecasts AI agents to manage 90% of traffic by 2035

Huawei turns the “agentic internet” into an infrastructure forecast

Huawei has moved the idea of an agent-driven internet from a marketing phrase into a measurable infrastructure forecast. In a report unveiled on September 16, 2026, the company said autonomous AI agents are expected to represent more than 90% of global AI token traffic by 2035 [1]. The projection appeared in Huawei’s “Intelligent World 2035: Turning Vision into Action” report, released during an online launch event under the theme “Act Today. Shape Tomorrow” [1].

The number is striking because it reframes traffic growth around agents rather than people. Huawei is not simply predicting more chatbot queries or heavier consumer use of generative AI. It is describing a future in which software agents continuously observe environments, reason over data, make decisions, use tools and communicate with other systems, generating a volume of machine activity that dwarfs today’s human-led interactions [2]. Reuters reported the same forecast as a bet that autonomous agents will dominate global AI token traffic, with the change requiring far larger computing capacity and new controls for independent software [2].

That distinction matters. In current reporting, Huawei’s “traffic” claim is specifically tied to AI token traffic, not every packet moving across the public internet [1]. A token is the unit of information an AI model processes when it reads, reasons or generates output; as agents break tasks into repeated observations, planning steps, tool calls and responses, token consumption can expand much faster than conventional web browsing [3]. The forecast therefore points to a new pressure point for digital infrastructure: not just bandwidth, but compute, memory, orchestration, latency, energy and governance.

The 2035 scenario: 900 billion agents and token demand exploding

Huawei’s 2035 vision is built around scale. The company says annual global token consumption could increase 100,000-fold by 2035, with traffic from agents accounting for over 90% of that total [1]. It also builds on a previous Huawei projection that as many as 900 billion AI agents could be active globally by 2035, roughly 100 times the projected human population [2].

Those figures describe a world in which AI agents are not isolated assistants but a basic operating layer of the digital economy. A person might still initiate a request, but a chain of agents could search data, negotiate with services, check compliance rules, book resources, generate code, inspect sensor feeds or coordinate robots. Each step would create model calls and token flows. The user sees one outcome; the network sees a dense mesh of automated reasoning.

Cryptopolitan’s coverage emphasized that Huawei’s projection implies a major increase in computing and networking capacity to support an agent-driven AI economy [3]. Huawei’s own launch statement places the forecast among 10 priority directions for the intelligent world, including AGI, computing clusters, storage and memory, intelligent connectivity, chip design, an Agent OS, device intelligence, intelligent driving, AI data center power and cooling, and security and privacy [1]. In other words, Huawei is treating agent traffic as a system-wide transformation rather than a software feature.

Why this forecast is bigger than telecom bandwidth

For telecom operators, the obvious implication is traffic growth. But Huawei’s framing is more complex than “more data over the same pipes.” Agentic services can require low latency, predictable compute access, reliable identity and context-aware quality of service. If agents are making decisions in logistics, manufacturing, finance, customer service or autonomous vehicles, the network must support task completion, not merely content delivery.

This is where Huawei’s forecast connects to the broader shift from app-centric internet use to multi-agent collaboration. In the app era, networks mainly transported human commands and media streams. In the agent era, agents may communicate with models, databases, APIs, devices and other agents before a human sees a result. The bottleneck may move between access networks, cloud inference clusters, edge compute nodes, memory systems and security gateways.

Huawei’s report also suggests that physical-world AI will be central to the path toward more capable systems, including AGI [2]. That means token traffic would not be confined to office productivity or online search. Cameras, industrial sensors, robots, vehicles and smart devices could feed agent systems that interpret, plan and act. The more agents are embedded in operations, the more token flows become part of the infrastructure of work.

The practical result is a new planning horizon for carriers and cloud providers. Capacity forecasts based on video streaming and mobile apps may understate future demand if large numbers of agents run persistently in the background. Agent traffic is also bursty in different ways: a single complex task can trigger many sub-tasks, model calls and verification loops. That can increase pressure on scheduling systems, inference availability and cross-domain coordination.

Security becomes a core design requirement

The same autonomy that makes AI agents useful also makes them risky. Huawei’s September 16 report identifies security and privacy protection for autonomous agents as one of the technologies needed to support the agentic era [1]. Reuters noted that Huawei’s forecast comes as U.S. researchers and executives debate whether frontier AI development should slow after rogue-agent incidents and warnings that future systems able to operate hardware or acquire resources could become difficult to control [2].

Huawei’s position, as reflected in the new roadmap, is not a call to stop the build-out. It assumes rapid expansion and focuses on technical safeguards, infrastructure readiness and controllability [2]. Investing.com’s report said Huawei has already launched an autonomous security-operations platform using sensing and enforcement agents, and has also published proposals for AI firewalls designed to protect enterprise AI systems [4].

That architecture question is central. If agents account for most AI token traffic, security can no longer be an outer layer added after deployment. It has to sit inside identity systems, permissions, model access, tool invocation, data governance, audit trails and network policy. Agents must be able to prove what they are, what they are allowed to do, which data they can access and when a human or higher-level controller must intervene.

The risk is not limited to malicious AI. Poorly scoped agents could loop, overspend, leak data, misread instructions or cascade errors across interconnected services. A future of hundreds of billions of agents would require automated security agents too, but that creates a recursive governance problem: who watches the watchers, and how are automated defenses constrained?

Huawei’s strategic subtext: compute, chips and China’s AI stack

The forecast is also a strategic statement from Huawei. The company is positioning itself not only as a telecom vendor but as a provider of the compute, network, storage, power and security layers needed for an agent-heavy AI economy. Reuters described the report as calling for huge increases in computing power alongside new systems to keep independent software secure and controllable [2].

This matters because AI infrastructure has become a geopolitical and industrial priority. Huawei is operating in a market shaped by U.S. export controls, China’s push for domestic AI hardware and global demand for alternatives to Nvidia-centered stacks. Investing.com reported that Guo Ping, chairman of Huawei’s supervisory board, told employees in remarks released on September 15, 2026, that Huawei views AI as its “biggest opportunity” and wants its computing infrastructure to become China’s equivalent of Nvidia [4].

The 2035 agent forecast supports that ambition. If token consumption rises 100,000-fold, the market does not only need better models. It needs inference chips, high-speed interconnects, storage systems, data-center power delivery, cooling, network intelligence and security controls. Huawei’s 10 priority directions read like a map of where it wants customers, carriers and governments to invest [1].

What to watch next

The key question is whether Huawei’s 90% figure becomes an industry planning benchmark or remains a vendor vision. Forecasts a decade out are inherently uncertain, especially in AI, where model efficiency, regulation, hardware constraints and business adoption can change quickly. Smaller, more efficient models could reduce token load per task; at the same time, cheaper inference could encourage far more agent activity.

Another question is how “agent traffic” will be measured. Huawei’s current framing uses global AI token traffic, a model-centric metric [1]. Telecom operators may prefer bandwidth, sessions, latency classes or quality-of-service indicators. Cloud providers may focus on inference requests, GPU hours, memory bandwidth or energy per token. Policymakers may focus on accountability, privacy and systemic risk.

Still, the significance of Huawei’s forecast is clear. By saying agents could generate more than 90% of AI token traffic by 2035, Huawei is arguing that the next internet will be automated first and human-facing second [1]. The winners in that environment will not simply be the companies with the best chatbot interface. They will be the firms that can make agent activity fast, cheap, secure, observable and governable at planetary scale.

Developments

1. TechCrunch Disrupt 2026 to Address AI Agents in Team Collaborationua.news · Sep 17, 2026, 4:29 AM UTC · 7/10
2. TechCrunch Disrupt 2026 Focuses on Teams with AI AgentsUA.NEWS · Sep 17, 2026, 4:29 AM UTC · 7/10
3. Huawei predicts AI agents will handle 90% of token traffic by 2035digitimes · Sep 17, 2026, 2:53 AM UTC · 8/10
4. Huawei predicts AI agents will account for 90% of token traffic by 2035digitimes · Sep 17, 2026, 2:53 AM UTC · 8/10
5. Huawei Forecasts AI Agents to Dominate Global Token Traffic by 2035Investing.com · Sep 16, 2026, 4:43 PM UTC · 9/10
6. Huawei forecasts AI agents to dominate global token traffic by 2035Investing.com · Sep 16, 2026, 4:43 PM UTC · 8/10
7. Gartner sees physical AI and autonomous agents reshaping connected systems by 2030IoT Business News · Sep 16, 2026, 2:00 PM UTC · 9/10
8. Gartner forecasts major impact of physical AI and autonomous agents on systems by 2030iotbusinessnews.com · Sep 16, 2026, 2:00 PM UTC · 8/10
9. Huawei predicts 90% of token traffic will be from autonomous AI agents by 2035Межа. Новини України. · Sep 16, 2026, 1:08 PM UTC · 9/10
10. Huawei predicts AI agents will handle 90% of token traffic by 2035Межа. Новини України. · Sep 16, 2026, 1:08 PM UTC · 8/10
11. Huawei predicts billions of AI agents will dominate traffic by 2035Investing.com · Sep 16, 2026, 1:04 PM UTC · 7/10
12. Huawei forecasts billions of AI agents to dominate traffic by 2035Reuters · Sep 16, 2026, 12:59 PM UTC · 9/10
13. China's Huawei forecasts billions of agents will dominate AI traffic by 2035 - Yahoo! Finance CanadaYahoo! Finance Canada · Sep 16, 2026, 12:53 PM UTC · 7/10
14. Huawei forecasts billions of AI agents to dominate 2035 trafficThe Lufkin Daily News · Sep 16, 2026, 12:53 PM UTC · 8/10
15. Huawei Forecasts Billions of AI Agents to Dominate Traffic by 2035marketscreener.com · Sep 16, 2026, 12:52 PM UTC · 8/10
16. AI Agent Statistics 2026 Show Widespread Adoption and Mature Capabilitiesmarkets.businessinsider.com · Sep 16, 2026, 12:20 PM UTC · 8/10
17. AI Agent Statistics 2026: Verified Data at SourceTradingView · Sep 16, 2026, 9:57 AM UTC · 8/10
18. AI's Capabilities and Threats in Biosecurity: A 2026 ReviewArXiv - Artificial Intelligence · Sep 16, 2026, 4:00 AM UTC · 10/10

Sources from the last 72 hours

1. [1]Huawei's latest report proposes 10 key directions to Intelligent World 2035Sep 15, 2026, 4:00 PM UTC
2. [2]China’s Huawei forecasts billions of agents will dominate AI traffic by 2035Sep 16, 2026, 8:53 AM UTC
3. [3]Huawei says AI agents will drive over 90% of AI traffic by 2035Sep 16, 2026, 12:00 AM UTC
4. [4]Huawei predicts AI agents to dominate global token traffic by 2035Sep 16, 2026, 1:33 PM UTC

AI-generated article based on recent web research, then preserved as a dated editorial snapshot.