Ask most AI infrastructure providers about networking infrastructure and the conversation turns quickly to throughput. It should. When GPU clusters stall waiting on data, deterministic connectivity is the foundation that delivers predictable, high-capacity performance that prevents network bottlenecks and keeps expensive accelerators fully fed.

A strong connectivity backbone is essential for fueling neocloud and AI infrastructure performance, but throughput is only part of the equation. As platforms evolve beyond raw compute to meet customer demands, new challenges of deployment timing and capacity uncertainty come into view. Network-as-a-Service (NaaS) models extend what the network can do, with on-demand services delivering the flexibility and agility needed to activate in minutes and adapt with changing demand.

Start with a deterministic foundation, then extend it

Dedicated routes help answer a specific question in the context of AI-native traffic flows: how do we deliver predictable, high-throughput performance between clusters, data centers and regions so the network never becomes the bottleneck? For AI infrastructure providers, high-capacity optical transport helps hold performance steady under the most demanding training workloads and is the foundation the rest of the network is built on.

Once that foundation is in place, providers begin to examine the next stages of platform growth. On-demand connectivity for AI infrastructure providers can extend the network to answer a different set of questions to meet end-user and tenant needs:

• Can we support unpredictable data ingress and egress associated with large training runs?
• How can we ensure secure access to public-facing models and APIs?
• How do we create a repeatable ecosystem for tenants and customers to access clouds and partners without defining new infrastructure for each use case?

The answer to these questions lies in network agility. While deterministic connectivity provides consistent performance, on-demand connectivity can help absorb uncertainty by adapting as requirements shift.

Where on-demand connectivity extends the network’s value

The following use cases put on-demand connectivity into practice for AI-infrastructure providers, exploring its application within changing environments and variable customer demands:

Scale capacity for large training and data staging

Challenge: Large training and data-staging jobs don’t consume bandwidth on a smooth curve. A single run can pull petabytes of data in then go quiet for hours. Dedicated transport gives operators the high-capacity backbone these workloads run on, but the variable ingress and egress spikes benefit from a way to adjust bandwidth rather than provisioning for every peak in advance.

Solution: Lumen® Intelligent Internet and Lumen® Ethernet Fabric Connect give operators programmable control to scale bandwidth up to 100 Gbps in minutes for high-volume ingest windows and back down when the run completes, all through Lumen Connect®. Changes can be made in near real-time so bandwidth spend aligns to actual demand, helping keep operational costs predictable.

Support and protect public models and APIs

Challenge: Not every workload can live on a private circuit. Public-facing models, customer APIs and tenant onboarding all depend on the internet to reach the users, developers and applications that consume them. However, each additional public-facing endpoint extends the attack surface.

Solution: Pairing elastic public connectivity with network-level security helps ensure applications and APIs stay available and protect against attack conditions. Intelligent Internet provides scalable internet capacity so spikes in usage don’t restrict users while Lumen Defender℠ adds threat detection and blocking informed by global network visibility from Black Lotus Labs® to filter malicious traffic before it impacts customer workloads.

Streamline connections to clouds and partners

Challenge: Tenants and end-users are increasingly reliant on multicloud environments. Training data sits in one provider, inference runs against another, and enterprise customers expect private, high-performance paths into the clouds they already use. Building these connections as one-offs increases topology complexity and isn’t scalable.

Solution: Lumen® Multi-Cloud Gateway delivers private, on-demand connectivity into major cloud providers without requiring dedicated infrastructure for each one. With Ethernet Fabric Connect as the underlying Layer-2 fabric, cloud connections can be managed as configuration changes rather than physical builds giving operators a single interconnection layer where multicloud access is scalable and on demand.

These use cases represent only a fraction of the utility an agile network can provide for AI infrastructure providers. With a strong, deterministic foundation in place to support predictable performance and consistent routing with dedicated capacity, flexible capacity becomes a powerful complement. On-demand connectivity is most valuable when bandwidth requirements fluctuate, connections need to be provisioned quickly, cloud and partner access changes frequently or demand changes faster than traditional provisioning can keep up with.

Build on a deterministic foundation and scale with on-demand connectivity

Whether the challenge is variable traffic patterns, growing user bases or expanding models, on-demand connectivity helps AI infrastructure providers adapt to changing demands and better serve the needs of tenants and end users.

Multiservice fabric ports provide the programmable foundation for agile networking. Intelligent internet delivers flexible, dedicated public connectivity so users can reach models and bandwidth can flex around training spikes. Ethernet Fabric Connect is the digital fabric for private connectivity and helps extend that fabric into cloud ecosystems with Multi-Cloud Gateway. Lumen Connect makes it easy to activate and manage connections in near real-time, giving AI infrastructure providers more control and network agility.

Extend your network with on-demand connectivity and save on your first three months. Explore the Lumen® Network-as-a-Service On-Demand offer and activate programmable connectivity in minutes. Offer ends 10/26; terms apply.*

_{*Offer available August 1 – October 30, 2026, with purchase of a Lumen® Fabric Port (minimum 1-year term agreement), and Lumen® Intelligent Internet or Ethernet Fabric Connect (EFC) services, in one eligible continental U.S. on-net or off-net location. Must activate service within 30 days of port installation. Service credit will be issued in fourth month after port billing commencement. 1 Gbps, 10 Gbps or 100 Gbps customers receive 100% service credit on Lumen® Intelligent Internet or EFC bandwidth charges. Offer excludes 1 Mbps Lumen® Intelligent Internet service. Taxes, fees and optional add-ons are not included. Customers must remain in good standing. Offer may terminate if customer changes service location. Early termination fees apply. Lumen may change, cancel or substitute offers and services, or vary by service area at its sole discretion without notice. Credit approval and deposit may be required. Offers may not be combined with other offers. Additional restrictions, eligibility requirements, terms and conditions apply; see Order Agreement and Order Addendum.}

_{This content is provided for informational purposes only and may require additional research and substantiation by the end user. In addition, the information is provided "as is" without any warranty or condition of any kind, either express or implied. Use of this information is at the end user's own risk. Lumen does not warrant that the information will meet the end user's requirements or that the implementation or usage of this information will result in the desired outcome of the end user. All third-party company and product or service names referenced in this article are for identification purposes only and do not imply endorsement or affiliation with Lumen. This document represents Lumen products and offerings as of the date of issue. Services not available everywhere. Lumen may change or cancel products and services or substitute similar products and services at its sole discretion without notice. © 2026 Lumen Technologies. All Rights Reserved.}