Cisco identifies service provider opportunity in AI inference and agentic traffic
Cisco executives predict a rising opportunity for network access providers, such as cable operators and telcos, in the AI inferencing era, as AI shifts from data-center training to distributed inferencing and autonomous AI agents. Agentic AI traffic is expected to consume significantly more data than human-led actions, altering internet traffic patterns. Cisco stresses the need for service providers to proactively address these shifts and consider establishing their own inferencing clouds, potentially competing with or complementing existing GPU-centric 'neoclouds'.
Key Takeaways
- AI inference traffic is projected to grow from negligible levels today to 25% of total network volume by 2035.
- Autonomous AI agents generate approximately 450% more data traffic than human users and maintain flows for twice as long.
- Cisco is deploying network sensors to track real-time AI data flows, with plans to publish trend updates every six months.
- Network infrastructure provider Megaport is raising $594 million to build a dedicated inference cloud strategy across 20 U.S. cities.
- Coding platforms like Codex and Claude currently represent the highest source of enterprise-level AI traffic.
Why It Matters
The transition from AI training in centralized data centers to edge-based inference represents a structural opportunity for telcos and cable operators to monetize their distributed infrastructure. Immediate shifts in traffic symmetry — where AI prompts increase upstream loads — require operators to move beyond simple connectivity to host sovereign 'neoclouds.' For the streaming and video industry, this suggests a move toward local, latency-sensitive AI processing for real-time personalization. Market participants should monitor for large-scale GPU deployments in edge hubs as a signal of this shift.
Additional Context
The push toward edge-based AI has accelerated through several major operator partnerships in 2026. Per Light Reading and company announcements from March 2026, Comcast and Charter have initiated field trials using Nvidia GPUs to run latency-sensitive AI workloads directly in regional facilities. Comcast’s 'Personalized Advertising Agent' uses Decart AI video models to customize ad delivery at the household level, while Charter identifies its footprint of over 1,000 edge data centers as a key advantage for real-time inference requiring sub-10 millisecond response times.
Simultaneously, telcos are securing specialized hardware to compete in the inference market. Per Convergence Digest and PR Newswire in May 2025, Bell Canada established the 'Bell AI Fabric' in partnership with Groq, utilizing custom Language Processing Units (LPUs) across six hydro-powered sites. This sovereign AI strategy aims for 500 megawatts of compute capacity, starting with a 7MW facility in British Columbia. These projects highlight a broader industry trend where operators use their domestic real estate to offer domestically governed AI compute as a managed service.
Further solidifying the 'AI Grid' concept, AT&T and Cisco announced a collaboration in March 2026 to integrate Nvidia accelerated compute into AT&T’s IoT core. According to AT&T Business, this architecture is designed for 'physical AI' applications such as industrial automation and public safety, where processing must occur near the data source. These moves indicate that the industry's infrastructure focus is expanding from 5G transport to a unified intelligence layer that scales AI workloads across the network boundary.
Read full article at lightreading.com
Enjoy our coverage?
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source