Nokia additionally estimates that AI might signify 30% of all international site visitors
In sum – what to know:
A brand new site visitors wave – In contrast to video or SaaS, AI generates compounding demand by consumer interactions, API calls, agent-to-agent workflows, and fixed information replication throughout clouds and areas.
Inference and agentic AI – Coaching is centralized and deliberate, however inference is real-time, latency-sensitive, and more and more distributed, making efficiency and time-to-first-token important community metrics.
A brand new forecast from Nokia means that extensive space community (WAN) site visitors might develop by as a lot as 700% by 2034, pushed largely by the speedy enlargement of synthetic intelligence workloads — significantly inference and agentic techniques which can be reshaping how information strikes throughout networks. Nokia additionally estimates that AI might signify 30% of all international site visitors.
Talking to the findings, Cameron Daniel, the CTO of network-as-a-service supplier Megaport, stated that whereas earlier progress waves, resembling video streaming, SaaS adoption, and cloud migration, added important site visitors on the edge, AI introduces site visitors that multiplies throughout each layer of digital infrastructure.
“AI provides site visitors in every single place,” he advised RCR Wi-fi Information. He pointed to the compounding impact of user-to-AI interactions, application-level mannequin API calls, autonomous AI brokers speaking with different techniques, and big replication of knowledge throughout areas and clouds.
He famous, as effectively, that in contrast to earlier web booms, this progress is not going to be evenly distributed. Some routes — particularly these linking main AI information heart hubs — might see site visitors will increase of 10x or extra, whereas different areas might expertise solely modest change. In actual fact, he shared that a number of AI corridors have already recorded several-hundred-percent progress over the previous 24 months.
Nokia’s evaluation factors to inference, not coaching, because the defining AI driver for WAN evolution. Daniel defined that coaching workloads are typically centralized batch jobs, with huge bandwidth necessities, however they’ll usually be deliberate prematurely, and latency exterior the coaching cluster is “typically tolerable.” Inference is completely different. It represents the real-time “use” of AI techniques — from chat interfaces to autonomous enterprise brokers — and it’s much more latency-sensitive.
Agentic workflows, specifically, increase the stakes. Nokia tasks agentic AI site visitors will develop at a 26% CAGR by 2034, driving a major rise in inference site visitors carried over the WAN. As AI brokers start speaking with different brokers, efficiency turns into tightly linked to each community latency and inference latency. “Which is why most mannequin suppliers monitor metrics like time-to-first-token,” added Daniel.
Moreover, as inference shifts nearer to customers, factories, campuses, and industrial environments, Nokia argues that networks should evolve past best-effort connectivity.
Edge inference requires low and constant latency, quick failover, and infrastructure that may assist service-level intent — particularly as AI turns into embedded in operational decision-making. Daniel stated: “Edge inference shifts the community necessities from finest effort to one thing nearer to infrastructure with service-level intent. Networks more and more must ship low and constant latency with quick failover, and as AI strikes nearer to customers and industrial websites, these necessities are simply going to amplify.”
If AI site visitors scales as projected, Danial warns that the primary strain factors will seemingly emerge in interconnect and information heart interconnect (DCI) hyperlinks, the place east-west site visitors progress will outpace conventional web demand. “Community operators and AI suppliers want to remain forward of this,” he cautioned.
Safety can also be turning into a important concern: “Shadow IT is an issue, and a rogue agent might do significant injury. DLP and different instruments are important to make sure information stays protected.”
Finally, Daniel argued crucial shift will not be site visitors quantity alone, however the rise of agent-to-agent communication — a brand new class of community conduct that would redefine WAN design over the subsequent decade.
