Telco for AI: Orchestrating the Global AI-Native Fabric

Enterprises spent the past decade moving workloads to centralized clouds. Now, AI is forcing a dramatic reversal.

By 2030, half of enterprise AI inference workloads are expected to run on edge devices or local endpoints, lowering reliance on cloud infrastructure while improving latency, traffic efficiency, and data governance (IDC). This multibillion-dollar infrastructure shift represents telecommunications operators’ biggest revenue opportunity since the mobile internet.

The stakes are enormous. Market projections put the AI infrastructure sector at $200 billion by 2028. But capturing this opportunity requires understanding why AI workloads need telecommunications infrastructure—and how operators can monetize that upper hand.

In this article, we’ll explore the architecture and technical foundations of telco AI and how operators can capture value. In Part 2, we’ll examine the specific, revenue-generating use cases operators can deploy today.

Why AI needs telco

Autonomous vehicles cannot wait 100 milliseconds for a cloud data center to decide whether to brake. Industrial robots rejecting defective parts cannot afford the latency of round-tripping sensor data to a hyperscaler. While centralized cloud platforms were essential for training large language models, inference-driven AI in the real world reverses the equation: real-time decisions require proximity and determinism, not just massive scale.

Telcos already own what hyperscalers are racing to assemble—thousands of distributed edge sites close to end users, high-capacity fiber connecting them, and decades of experience operating and automating large-scale, geographically dispersed systems.

But the opportunity for telcos is not simply a matter of deploying GPUs at cell towers.

Edge AI at scale is a coordination problem, not a hardware problem. The real value emerges when compute, data, and networks are orchestrated as a single system.

Consider a federated “Global AI Cloud,” where telcos manage AI workloads seamlessly across edge locations, core networks, data centers, and public clouds using open standards such as TM Forum Open APIs and Mplify LSO. Enterprises could provision AI compute with the same zero-touch automation that telcos already use for network services. In this model, workloads receive predictable performance, data remains sovereign, and operators monetize AI as a service—rather than competing solely on connectivity.

Seen this way, AI behaves like a distributed operating system, coordinating inference across factory floors, vehicles, data centers, and edge devices in real time. This shift introduces three infrastructure requirements where telcos hold clear structural advantages:

Latency & determinism: Real-time inference in factories or autonomous vehicles demands sub-50ms response times with zero jitter. Telcos guarantee these SLAs through network slices—dedicated virtual pathways providing predictable performance via orchestrated L2/L3 connectivity and edge computing.

Data gravity & sovereignty: Regulations like GDPR prohibit moving petabytes to central clouds. Telcos process data at the edge, moving algorithms to where data lives while maintaining compliance.

High-speed interconnects: LLM training requires 100Gbps+ connections for gradient synchronization across GPU clusters. Using Mplify LSO standards, telcos provision ultra-fast, lossless connections between providers, creating federated virtual supercomputers.

Meeting these requirements demands a specific global, performance-guaranteed and standardized architecture—one that telcos already manage.

The “Telco for AI” architecture

To support AI at scale, the AI-Native Fabric architecture acts as a unified orchestration platform that works across three layers: business integration that translates enterprise intent using TMF/Mplify standards; orchestration that places and manages AI workloads and clusters; and infrastructure that delivers high-performance networks and GPU compute.

Business integration layer (Northbound)

This top layer is essentially the storefront and business logic that makes AI infrastructure consumable without requiring customers to understand what’s happening underneath. For example, when a company orders “AI compute with 99.9% uptime,” it uses standards such as TM Forum and Mplify to translate that request into technical requirements, provision the necessary resources, and coordinate with partner networks if additional capacity is needed.

Orchestration layer

The orchestration layer turns business orders into running AI workloads. It maintains a real-time view of available edge capacity, GPU locations, and network bandwidth, making placement decisions based on latency and data proximity. Built on Kubernetes, it abstracts physical hardware into portable workloads deployed across cloud, edge, and on-premise environments, while continuously monitoring performance and automatically adjusting for congestion or resource imbalance.

Infrastructure layer (Underlay)

At the foundation, the infrastructure layer configures the AI Fabric itself: GPU clusters and accelerators, high-speed parallel file systems, and purpose-built networks. These include 400 Gbps fabrics for GPU-to-GPU communication during training, alongside carrier-grade Ethernet and SD-WAN for reliable inference traffic. Distributed across edge locations, this layer enables AI workloads, such as route-optimization systems, to run just kilometers from where decisions are made, rather than in distant cloud regions, preserving latency and reinforcing telcos’ structural advantage.

Breaking the single provider ceiling

No single operator can provide all the infrastructure needed for AI at scale. Geographic reach, compute density, and regulatory constraints make isolated deployments impractical.

The fix? Industry standards and inter-provider federation.

Mplify LSO enables collaboration—Cantata for B2B marketplaces, Sonata for inter-provider federation, Legato/Presto for provisioning, and Interlude for assurance—while TM Forum Open Digital Architecture (ODA) productizes AI services from catalog to billing.

By adopting these standards, operators can dynamically negotiate, trade, and provision AI resources across networks in near real time, removing capacity ceilings caused by infrastructure gaps or regulatory constraints and enabling use cases otherwise out of reach.

Consider a simple scenario. If a customer orders 64 GPUs but a single operator has only 32, standards-compliant orchestration can provision the rest from partner networks. Kubernetes workloads are deployed, performance is monitored via unified telemetry, and settlement happens automatically—creating a seamless “global virtual supercomputer” for the customer.

The business case

Standards-based orchestration positions telcos as indispensable architects of the AI-Native Fabric, delivering measurable gains across operations, revenue, and scalability. By automating end-to-end lifecycles—from intent to infrastructure—telcos resolve AI’s core constraints, turning connectivity into high-margin intelligence platforms.

The following benefits illustrate how standards-based automation enables telecom providers to effectively capture the AI opportunity.

​Operational agility: weeks to minutes

Traditional AI deployments demand weeks of manual coordination across network, compute, and security teams. Intent-based orchestration automates order decomposition, K8s deployment, and network slice provisioning simultaneously. As a result, telcos can rapidly respond to spikes, like retraining models, reduce lead times by orders of magnitude, and enable experimentation at scale.

​AI-aware network fabric with SLAs

AI hypersensitivities like jitter or packet loss halt training; generic internet fails here. Orchestrators configure networks dynamically, guaranteeing sub-1ms latency and zero-loss paths. Customers buy deterministic channels, not best-effort pipes, and telcos offer premium SLAs (e.g., 99.999% uptime for gradients), backed by telemetry and closed-loop healing. This elevates networks from commodity to critical AI enablers.

​Monetization beyond connectivity

Standards unlock granular service models: GPU-hours, petabyte-sync fees, latency-guaranteed sessions. Edge sites become AI Points-of-Presence, monetizing 5G assets through bundled services. Federated AI Clouds allow operators to sell capacity wholesale to hyperscalers or peers, while sovereign AIaaS—blending local data control with global reach—creates high-margin revenue streams that dwarf traditional IP transit.

​Interoperability and ecosystem scale

Proprietary silos trap resources; open APIs and multi-domain orchestration federate vendors and providers seamlessly. Operators can “burst” workloads globally—for example, a London site dynamically leveraging Warsaw GPUs—scaling beyond local limits without lock-in.

​Competitive moat for telcos

Hyperscalers dominate compute but lack telco-grade determinism and edge density. By combining orchestration, fiber, MEC, and NaaS expertise, operators can deliver sub-10ms inference where hyperscalers cannot, creating a clear competitive advantage.

What’s next?

The technical architecture is clear. The business case is compelling. The question is execution: which specific use cases generate revenue today?

In Part 2 of this blog, covering telco for AI use cases, we’ll explore deployment-ready scenarios, showing how operators capture their share of the $200B+ AI infrastructure wave.

Watch this space and contact us directly if you have any questions.