Table of Contents
The 2026 F5 State of Application Strategy report highlights a massive shift toward decentralized AI production
In sum – what we know:
- The inference explosion – 77% of enterprises now prioritize inference over training, with the average organization managing or evaluating seven different AI models simultaneously.
- A hybrid reality – Infrastructure is increasingly decentralized, with 86% of organizations distributing AI applications across on-premises, public cloud, and colocation sites to avoid provider lock-in.
- The telco opportunity – Telecommunications providers are positioned as ideal orchestration partners due to their experience in intelligent routing, latency reduction, and securing complex, multi-tenant networks.
F5 has released its 2026 State of Application Strategy (SOAS) Report, surveying hundreds of enterprise IT and security leaders worldwide. The headline finding is that AI inference has crossed the line from experimental workload to production reality, and enterprises are overwhelmingly choosing to run it themselves — rather than outsourcing it to hyperscalers.
What makes the report interesting isn’t just the scale of that shift, but what it implies about who gets to play a role in the infrastructure underneath it. F5 is pointedly positioning telecommunications companies as critical partners for enterprises wrestling with the messy, decentralized reality of AI workloads spread across clouds, colos, and on-prem data centers.
Enterprise AI infrastructure complexity
Some 77% of organizations now identify inference, not model training, as their dominant AI activity, and the average enterprise is running or actively evaluating seven AI models simultaneously.
Where those models live is the harder part. 93% of surveyed organizations operate in hybrid multicloud environments, and 86% distribute their applications across on-premises, public cloud, and colocation facilities. More than half (52%) are actively chaining or orchestrating multiple AI models together, meaning a single request may need to hop between models hosted in entirely different environments. Every inference request becomes a routing decision weighed against cost, accuracy, availability, latency, and geographic distribution. Multiply that across seven models and three or four infrastructure tiers, and the traffic management problem starts to look less like application delivery and more like running a small network.
Telcos in traffic management
Managing multi-model routing across distributed infrastructure isn’t a new problem in principle — it’s a close cousin of the CDN load balancing and failover work telcos have been doing for decades. The mechanics map over reasonably well: intelligent routing, performance optimization, and graceful failover across geographically distributed nodes.
Telcos also bring direct experience managing hybrid cloud and on-premises integrations across distributed data centers, which is precisely the topology enterprise AI is settling into. And there’s the issue of latency on top of that. Direct network integration via telcos offers intelligent edge placement that can reduce inference latency in ways cloud-only routing simply can’t match. Quality-of-Service capabilities, long a telco strength, let inference traffic be prioritized based on specific business requirements rather than treated as generic HTTP traffic. Whether enterprises actually route their AI through telcos at scale is another question, but the capability alignment is hard to argue with.
Security and governance
The security piece of the report is a little alarming. Over 90% of organizations say production-level agentic AI introduces significant new security challenges, including credential stuffing and difficulties auditing what AI agents actually do. New control points are emerging at the prompt and token layers — 29% of organizations identify prompts as a top delivery mechanism and control point, and 23% prioritize token layers for delivery and security. These are new surfaces, and the tooling to govern them is still catching up.
Enterprises are responding with what they know. 55% cite authentication management for AI services and API access as their primary security strategy, and observability across distributed models ranks second. Both are problems telcos have been solving at scale for a long time. Multi-tenant, multi-cloud security, authentication, and compliance infrastructure is basically table stakes for a carrier, and that native capability maps cleanly onto what enterprises are now trying to bolt onto their AI governance stacks.
A telco AI strategy
Telcos are shifting from pure connectivity providers into infrastructure orchestration partners for enterprise AI. That’s a different kind of business, and it plays to different strengths. Enterprises increasingly wary of locking themselves into a single public cloud provider for their AI stack are looking around for alternatives, and telcos are one of the few categories of player with the footprint, the network, and the operational discipline to credibly fill that role.
There’s also a feedback loop worth noting. 67% of organizations use AI to accelerate IT operations automation, and 66% use it to automatically adjust policies and configurations. AI is being used to manage the infrastructure that’s running AI, which only raises the stakes on reliability and observability at the infrastructure layer.
The competitive differentiator for telcos is providing reliable, secure, managed inference delivery infrastructure. Whether most carriers actually reorient themselves to capture that opportunity is a separate question. But F5’s report makes a reasonably convincing case that the opening is there.
