The emergence of large-scale AI infrastructure is fundamentally altering how networks function within and between data centers. Unlike traditional facilities where networks simply provide connectivity, AI compute clusters integrate networking as a core component of the system. Accelerators, storage systems and processing pipelines depend on multiple network layers operating in concert, and when workloads span multiple buildings, campuses or regions, the links between them become essential to overall performance.

This shift mirrors earlier moves by major cloud and content platforms toward building private infrastructure. Just as hyperscalers constructed dedicated cables and private backbones for content delivery, and later deployed edge caching closer to users, AI is now driving similar investment in compute interconnection. For network operators, AI infrastructure initially appears as requests for private fiber, dedicated wavelengths and data-center interconnection—often before any visible surge in public traffic.

The network becomes part of the machine

An AI factory comprises accelerators, storage, data pipelines and multiple separate networks all functioning as a unified system. Within a single campus, these networks transport model state, datasets and checkpoints. When compute spreads across several campuses, dedicated optical links enable those sites to operate as a larger resource pool.

Power availability emerges as the dominant factor determining where these systems can be constructed. The International Energy Agency characterizes conventional data centers at roughly 10–25 MW scale, while hyperscale AI-focused facilities operate at 100 MW or more, with individual projects now reaching gigawatt scale. Facilities of this magnitude cannot simply integrate into existing Internet hubs. Instead, their location must combine available electricity, cooling capacity, land and multiple independent fiber routes.

Google articulates this logic directly: transporting electricity proves more difficult than moving data across fiber, so the company positions data centers near energy sources and uses its network to aggregate AI compute across campuses. Microsoft similarly connects its Fairwater sites in Wisconsin and Atlanta through a dedicated optical AI wide-area network. Meta, presenting at RIPE 92, described how power and space constraints were already distributing AI infrastructure across broader metropolitan regions.

AI traffic is not one flow

The term 'AI traffic' encompasses multiple systems with different behaviors and visibility points. For operators, the practical consideration involves determining where each system becomes observable: within the factory, between campuses, or on the public Internet.

Inside the factory, accelerators exchange model state within a rack, cluster fabrics handle scale-out operations between racks, and storage supplies datasets and checkpoints to compute. The volumes are staggering: Google reports its Virgo fabric can connect up to 134,000 TPU 8t chips with as much as 47 Pb/s of non-blocking bisection bandwidth in a single fabric. Meta has documented checkpoint bursts that congested its data-center network and left GPUs idle. None of this appears on an Internet transit graph, yet this is where the network determines how much of the compute actually functions.

Between campuses, private interconnection and dedicated AI wide-area networks move checkpoints, replicas and distributed workloads over dark fiber or wavelengths. Meta indicates one pair of sites in its AI Backbone is twice the size of the global backbone it constructed during the preceding decade. For a fiber or data-center interconnection provider, this generates demand for private capacity with specific delivery windows.

On the public Internet, two flows reach operators directly. Inference sessions, where a small prompt returns streamed text, images, audio or video: Cisco measurements show median inference flows lasted about twice as long as non-AI web flows. And agents and crawlers, which read pages, call APIs and trigger other services: in one controlled Cisco task, an agent generated up to 450% more total traffic than a manual workflow completing the same task. These high request rates can create operational pressure even when bandwidth consumption remains modest.

PeeringDB reported degraded service from AI scrapers and introduced rate limits for unauthenticated web access. Fastly found that more than half of the AI-bot requests it classified reached origin, compared with less than 9% of human requests.

AI changes the backbone and the edge

Training requires large amounts of compute operating together, so access to power, cooling and connectivity strongly influences where clusters can be located. Inference—running a model to answer a request—may benefit from proximity to users when response time matters. However, cost, model availability and data-residency requirements also shape where that processing occurs.

This differs from the CDN caching model. A CDN can place reusable software, media and other objects near users, while an inference request normally requires model execution. The nearest cache, the service ingress and the GPUs running the model can exist in three different locations. AWS Bedrock cross-region inference, for example, can accept a request in one region and process it in another over the AWS network.

For this reason, an operator-hosted cache does not automatically become a GPU edge. At RIPE 92, Meta stated it did not expect forward-deployed appliances inside ISP networks to host power-hungry GPUs, and estimated that five to eight European locations could serve 99% of its current AI use cases.

Who builds the factory, and who connects it?

An AI factory may carry the name of a cloud or AI company, but multiple parties assemble it. The compute operator defines which sites must work together, how much capacity and path diversity they require, and how quickly the system must expand. Utilities supply power, developers build the facility, and fiber operators provide the routes and optical systems. Power availability, permits and existing fiber still constrain what can be built and where.

End users do not normally connect to the factory itself. They reach an API or service edge. The factory's external links more likely connect other compute campuses, storage and data platforms, private clouds, enterprise customers and network providers.

The Microsoft–Lumen agreement illustrates this division of work. Microsoft selected Lumen to expand connectivity between its data centers through a custom private fabric combining access to existing fiber with construction on existing and new routes. In another model, Zayo reports an anchor customer enabled 8,000 route miles of new builds and overbuilds, while Zayo builds, owns and operates the routes.

The compute customer steers where the carrier invests without replacing it as builder and operator. What changes is the product: dark fiber, dedicated wavelengths, diverse site entrances, inter-campus paths, restoration commitments and capacity delivered in phases as the factory grows.

An anchor customer can make a new route possible, but it can also leave the carrier dependent on one project's timetable. If that project is delayed or reduced, can the route serve other customers, and can later construction phases be deferred?

Public routing shows the edge, not the factory

An operator's public ASN does not reveal the full infrastructure of its AI factory. BGP, the RIPE Database, PeeringDB and RIPE Atlas can show its public edge: announced prefixes, registered resources, declared presence and selected paths. They do not expose accelerator fabrics, private data-center interconnection, storage traffic or workload placement. A traceroute can reach the service front door without revealing which campus or accelerator handled the request.

CoreWeave's AS33425 represents part of its customer and interconnection surface, while Direct Connect carries private customer links into virtual private clouds without crossing the public Internet. Lambda lists AS11588 and AS398090, but separately documents its private network and 400 Gb/s InfiniBand fabric.

Public traffic measurements tell a different story from the scale of AI infrastructure investment. In one mobile network studied by Ericsson, 21 identified generative-AI applications accounted for 0.06% of bytes during one week in April 2025. AppLogic reported AI traffic growth of 90% on fixed networks and 80% on mobile, while still describing it as a small share of total traffic. Interviews with public IP operators likewise found no massive AI-driven surge.

Both observations are correct. Public AI traffic grows rapidly from a small base, and the largest transfers never leave the private fabrics.

Traffic to an 'AI company ASN' tells operators little about the workload; the same cloud network can carry an AI API, storage, ordinary software services and unrelated customer workloads. Direction, flow duration, request rate and the type of interconnection can provide additional context, but do not necessarily identify traffic as AI-related.

Firebird: one regional example

Firebird AI's Hrazdan project was an early signal when first mentioned. Firebird announced the operational launch of the factory in August 2026.

Firebird reports that Team Telecom Armenia's network can provide the project with capacity of up to 1 Tbps and scale further. The operator separately reports four diversified international directions from Armenia. These are published network-capability statements, not measurements of used traffic.

For an AI factory, connectivity has to be planned alongside power and compute. Capacity needs to grow with each phase, with diverse routes to help maintain service when a link fails. That means designing for large data transfers and disruptions from the outset, as well as for everyday demand.

Arpine Chalkian, COO, Firebird AI

Firebird operates the compute platform, while a regional operator supplies its external connectivity.

AI Factory France: connecting national supercomputers

In France, AI Factory France provides researchers and companies access to national supercomputers. In October 2026, its coordinator GENCI and the Bristol Centre for Supercomputing announced plans to connect France's Jean Zay and Adastra with the UK's Isambard-AI for federated learning, with technical support from NVIDIA. Each site would train on its own data and exchange model updates rather than transfer the underlying datasets. The connection is targeted for the end of 2026 and would let separately operated computing centers contribute to a shared training process.

What operators should prepare for

For access and CDN operators, the useful starting point is service behavior: which automated requests reach origins, where they create load, and whether enterprise customers need private cloud connectivity. Request rate, errors and response time can expose pressure that a traffic-volume graph misses.

For fiber, DCI and hosting providers, three questions should be settled before capacity is committed:

  • What must be transferred, and by when? Establish dataset and checkpoint sizes, frequency and completion windows; averages alone do not define the link requirement.
  • What must continue operating after a route failure, and how quickly must the remaining services recover? Agree the required capacity on the surviving path and check for shared ducts and site entrances.
  • When is the next capacity step? Align wavelength or fiber delivery with power and compute commissioning, and agree the lead time for expansion.

Conclusion

Much of AI factory activity takes place on private networks, so public traffic figures alone will not tell operators what connectivity these systems need. Companies investing in compute increasingly shape network development around their sites and workloads. A location far from an established Internet hub can attract substantial investment if power and connectivity are available. For fiber and DCI providers, this means planning routes, capacity and delivery dates alongside the customer's power and compute deployment.