Lumen Doubles Network-as-a-Service Customers to Meet Enterprise AI Infrastructure Demands

Enterprise AI workloads are reshaping network architecture at pace. Lumen Technologies, the global infrastructure provider serving Fortune 500 companies across North America and Europe, has announced a doubling of its Network-as-a-Service (NaaS) customer base in the past 18 months, driven largely by the infrastructure demands of artificial intelligence deployment at scale. For UK Chief AI Officers and technology leaders, this market signal carries strategic weight: it underscores the growing realization that AI isn't a software problem alone—it's a connectivity and infrastructure imperative.

The shift reflects a fundamental truth that CAIO teams must grapple with: building a production-grade AI capability—whether large language models, retrieval-augmented generation systems, or enterprise machine learning pipelines—requires rethinking network architecture from the ground up. Lumen's growth in NaaS adoption suggests that forward-thinking enterprises are moving beyond traditional telecom procurement and towards managed, consumption-based network infrastructure that can flex to meet the unpredictable bandwidth and latency demands of AI training, inference, and data movement.

This article explores what Lumen's expansion means for UK enterprises, how NaaS models align with AI governance frameworks, and the strategic questions CAIOs should be asking their infrastructure teams about network readiness for AI at scale.

The Infrastructure Reality Behind Enterprise AI Ambitions

Most organisations publishing AI strategies focus on governance, talent, and model selection. Few address the unglamorous question: can our network handle it?

Large language model inference at enterprise scale demands both consistent bandwidth and predictable latency. A single inference request routed through a multi-node inference cluster—whether running on-premises, hybrid cloud, or fully cloud-native—creates a complex choreography of network traffic. Add data movement for model fine-tuning, vector database synchronisation, or real-time feature engineering for ML pipelines, and the infrastructure requirements become non-trivial.

Traditional enterprise networks, provisioned on fixed annual contracts with three-to-five-year lock-ins, were never designed for this variability. A machine learning training job might consume 500 Mbps of sustained bandwidth for three weeks, then drop to background inference traffic. A sudden need to retrain a recommendation model in response to market conditions could spike requirements overnight. Legacy networks lack the elasticity to respond, forcing CAIOs into uncomfortable choices: over-provision and waste budget, or under-provision and risk production degradation.

Lumen's NaaS model—offering on-demand, metered network capacity managed as a service rather than as capital infrastructure—addresses this gap. The doubling of the customer base suggests enterprises are finally waking to the strategic importance of this problem.

For UK organisations, this is particularly pressing. The Department for Science, Innovation and Technology (DSIT) has explicitly positioned the UK as a frontier AI nation. That ambition depends on robust infrastructure. The UK AI Safety Institute, now embedded within DSIT, has begun publishing guidance on safe AI deployment—but safety includes infrastructure robustness. A network outage that interrupts model inference or training isn't just a service incident; it's a governance incident.

What Network-as-a-Service Actually Means for AI Operations

NaaS is not simply "the cloud but for networking." It's a fundamental rethinking of how enterprises procure and consume network capacity.

The Key Operational Differences

• Consumption-based pricing: Organisations pay for network bandwidth actually used, measured in real-time, rather than committing to fixed port speeds and overprovisioning for peak days that may never come.
• Managed service model: Lumen (or equivalents like Cisco, Equinix, or regional providers) takes responsibility for network design, provisioning, monitoring, and incident response—not your network team.
• Dynamic scaling: Network capacity can be increased or decreased on demand, typically within hours or days, without capex purchases or equipment installation.
• Software-defined networking: Many NaaS offerings integrate with SD-WAN platforms, allowing policy-driven traffic management that can prioritise AI workloads or isolate model training from production inference.
• Performance guarantees: Service-level agreements (SLAs) typically include latency, jitter, and packet loss thresholds—metrics that directly impact AI performance.

For AI operations, this is material. Consider a scenario: your organisation has deployed a large language model for customer service (a common use case). Inference traffic is steady but latency-sensitive—customers expect sub-200ms response times. Network congestion from unrelated traffic could degrade that experience. With NaaS and SD-WAN, you can define policies that guarantee a minimum bandwidth allocation and latency ceiling specifically for inference traffic, leaving other services to compete for residual capacity.

Or consider model training. Your data science team wants to fine-tune a foundation model on proprietary customer data. This requires moving terabytes of data from your data warehouse to a training cluster. With traditional fixed networks, this competes for the same capacity as your production applications. With NaaS, you can provision a temporary, dedicated corridor for data movement, complete at speed, then release the capacity once training is done. You're billed only for what you use.

Implications for AI Governance

Governance frameworks—whether your own internal AI steering committee or compliance with emerging UK AI regulation—increasingly require visibility into infrastructure dependencies. The UK government's AI Bill of Rights emphasises transparency and auditability. Network infrastructure isn't mentioned explicitly, but it underpins the reliability and trustworthiness that these principles demand.

A managed NaaS provider typically offers:

• Detailed traffic analytics and logs, useful for auditing data movement compliance
• Isolation and segmentation capabilities, supporting data residency and security policies
• Consistent monitoring and alerting, reducing the risk of silent failures in AI pipelines
• Vendor support for regulatory questions (e.g., where data transits, encryption standards, incident response)

These aren't luxuries—they're increasingly table-stakes for enterprises handling sensitive data in regulated sectors (financial services, healthcare, public sector).

Why Lumen's Customer Growth Matters Now

Lumen's announcement of doubled NaaS customers is a market validation signal, but it's worth parsing what it actually signals.

Market Maturation and Adoption Pressure

The NaaS market has existed for several years, but adoption was gradual. Early adopters were primarily cloud-native enterprises and telecommunications companies. The doubling Lumen has experienced suggests the wave is now reaching mainstream Fortune 500 and mid-market enterprises—precisely the organisations where CAIOs operate.

This acceleration is directly tied to AI. Lumen has publicly attributed much of the growth to customers building out infrastructure for large language models and generative AI workloads. That's not surprising: enterprises that have deployed ChatGPT-style tools, or are evaluating them, quickly hit the realisation that network capacity becomes a constraint.

In the UK specifically, this timing aligns with the government's push to establish the country as an AI leader. The AI Framing Paper released by DSIT in 2023 called for a pro-innovation, principles-based approach to AI governance. One implicit requirement is that UK enterprises can compete on infrastructure. If your network can't support the same AI workloads as a US competitor, you're at a disadvantage.

The Competitive Landscape

Lumen isn't alone in this space. Competitors offering NaaS or adjacent services include:

• Cisco Meraki: SD-WAN and managed networking, increasingly bundled with infrastructure visibility tools useful for AI operations
• Equinix: Edge computing and network interconnection services, particularly relevant for distributed AI workloads
• Vapourcomm and other regional UK providers: Offering managed network services with UK data residency guarantees, important for regulated sectors
• Hyperscalers (AWS, Azure, Google Cloud): Increasingly offering managed network services as extensions of cloud offerings (e.g., AWS Direct Connect with managed bandwidth, Azure ExpressRoute)

Lumen's scale—operating one of the world's largest backbone networks—is an advantage. So is its focus on enterprise relationships. But the broader trend is what matters: NaaS is moving from niche to mainstream, and enterprises that haven't yet grappled with network architecture for AI are now being forced to do so.

What the Numbers Tell Us

Lumen has not disclosed absolute customer numbers, but industry analysts have noted that NaaS adoption across all vendors is accelerating. Gartner's WAN Edge Infrastructure research highlights that consumption-based models are gaining share from traditional fixed-capacity networking, driven by organisations seeking cost flexibility and operational agility.

For CAIOs, the implication is clear: if your enterprise hasn't yet evaluated NaaS options or refreshed network strategy in the context of AI, you're increasingly behind the curve. Lumen's customer growth is a canary in the coal mine—a signal that this conversation has moved from "nice to have" to "strategic imperative."

Network Architecture Questions CAIOs Should Be Asking Now

The doubling of Lumen's NaaS customers raises a set of strategic questions that CAIOs should be addressing with their infrastructure and network teams.

Is Your Network Optimised for AI Workloads?

This requires honest assessment:

• Can your current network provision additional bandwidth within weeks (not months)? If not, you're constrained.
• Do you have real-time visibility into bandwidth consumption by application? If not, you can't optimise or cost-account accurately.
• Can you isolate or prioritise traffic for latency-sensitive workloads like inference? If not, performance becomes unpredictable.
• What's your typical network utilisation rate? Sustained rates above 70% suggest you're running hot and should consider expansion.

These questions should prompt a conversation with your Chief Information Officer or VP of Infrastructure about the adequacy of your current network architecture.

What Are Your Data Movement Costs?

One often-overlooked aspect of AI infrastructure is the cost of data movement. Training a large model, or fine-tuning one on proprietary datasets, involves moving terabytes of data. If that data is sourced from an on-premises data warehouse and fed into cloud-based training infrastructure (or vice versa), the bandwidth costs add up quickly.

NaaS models with direct interconnection to cloud providers (e.g., Lumen's partnership with hyperscalers, or direct peering arrangements) can significantly reduce these costs compared to egress charges from public internet. Have you modelled this? For organisations planning to invest heavily in model training, these costs can be material.

How Will You Handle Distributed AI Operations?

As AI deployment matures, many enterprises are moving beyond centralised models. You might have:

• Inference clusters running at the edge (near customers or in regional data centres) with periodic model updates syncing from a central training hub
• Federated learning setups where model training happens across multiple locations
• A hybrid cloud architecture with different components running in different providers

Each of these topologies has different network requirements. NaaS providers increasingly offer tools and services to optimise connectivity across these distributed topologies, but you need to plan for it. Have you mapped your future AI architecture and its network implications?

What Does Your AI Governance Framework Require From Infrastructure?

If your organisation has published an AI governance framework—whether informed by the UK AI Safety Institute's guidance, the ICO's guidance on AI and personal data, or your own internal principles—does your infrastructure strategy align with it?

For example, if your framework requires that all customer data processing happens within the UK for data residency reasons, does your network architecture enforce this? If your framework requires audit trails of all data movement for compliance, can your current infrastructure provide these logs?

Governance that isn't backed by infrastructure design is governance theater. NaaS providers increasingly offer capabilities (segmentation, logging, traffic steering) that make governance enforcement easier, but only if they're designed into your architecture from the outset.

What UK Enterprises Should Be Doing Now

The market signal from Lumen's growth is clear. Here's a practical agenda for CAIOs:

Immediate Actions (Next 4 Weeks)

• Commission a network audit. Have an independent consultant (or your existing infrastructure partner) assess your current network's readiness for AI workloads. This should include bandwidth, latency, scalability, and cost metrics.
• Define your AI infrastructure roadmap. What AI capabilities are you planning to deploy over the next 12-24 months? What are the network implications?
• Engage your CIO/VP Infrastructure. Ensure network strategy is part of your AI governance and planning, not treated as an afterthought.

Medium-term (Next 3-6 Months)

• Evaluate NaaS options. Request proposals from at least three providers (Lumen, regional providers, hyperscaler offerings). Evaluate not just on cost but on capability, SLAs, and support for your specific AI topology.
• Align with governance frameworks. Map any NaaS solution against your internal AI governance requirements and regulatory obligations (GDPR, FCA guidance, ICO AI guidance, etc.).
• Pilot if appropriate. Consider a proof-of-concept, perhaps with a non-critical AI workload, to validate assumptions about performance, cost, and operational integration.

Long-term (6-12 Months)

• Migrate or integrate. If a NaaS solution proves compelling, plan the transition from your current network architecture. This is typically not a rip-and-replace; it's usually a phased integration.
• Build operational discipline. As network becomes consumption-based and dynamic, ensure your teams have the tools and training to manage it. This includes usage monitoring, cost attribution, and performance optimisation.
• Iterate based on learning. AI workloads will evolve. Your network strategy should too. Build in regular reviews (quarterly or twice yearly) to assess whether your infrastructure is keeping pace with evolving AI ambitions.

The Broader Context: UK AI Infrastructure Ambitions

Lumen's customer growth sits within a broader context of UK infrastructure investment in AI. The government's AI sector deal, announced by DSIT, includes commitments to support infrastructure for AI research and deployment. The Alan Turing Institute, the UK's national institute for data science and AI, is actively researching infrastructure questions including privacy-preserving distributed learning and efficient model deployment.

For commercial enterprises, this creates both opportunity and obligation. The opportunity is clear: infrastructure providers and cloud companies are investing in capabilities to support UK AI ambitions. The obligation is less obvious but equally important: UK enterprises that adopt sophisticated AI practices have a responsibility to demonstrate that they're doing so responsibly and sustainably.

Network infrastructure, though unglamorous, is part of that responsibility. A well-designed, well-monitored network that can support sophisticated AI workloads efficiently is foundational to trustworthy AI operations. Lumen's growth suggests that forward-thinking enterprises are recognising this.

Conclusion: Network Architecture as Competitive Advantage

For decades, network infrastructure was treated as a cost centre—necessary, but not strategic. The rise of cloud computing challenged that slightly, forcing CAOs to think about WAN architecture. But enterprise AI is forcing a genuine rethink.

Organisations that optimise network architecture for AI—whether through NaaS adoption, SD-WAN deployment, or edge computing integration—are positioning themselves to move faster, reduce costs, and operate more reliably. Those that don't risk being constrained by infrastructure just as their AI ambitions accelerate.

Lumen's doubling of NaaS customers isn't a random market fluctuation. It's a signal that this transition is underway. For UK CAIOs and technology leaders, the question isn't whether to engage with this shift, but how quickly you can do so.

The infrastructure question is no longer a sideshow. It's central to AI strategy. Act accordingly.

Related articles on CAIO Weekly:

• Building Governance Frameworks for Enterprise AI: A CAIO's Playbook
• UK AI Safety Institute Releases First Safety Guidance: What CAIOs Need to Know
• Infrastructure Planning for Large Language Models: Cost, Latency, and Scalability