Telecom operators didn't miss the cloud. They got unbundled from it.
Via HPE
Oct 7, 2026
Telecom operators didn't miss the cloud. They were unbundled from its value. As AI emerges, the challenge is no longer participation, but capturing a defensible position in the AI value chain.
Telecom operators participated extensively in cloud growth. Yet, they often describe cloud as a missed opportunity. Why? Because cloud value was captured by hyperscalers. As AI reshapes the industry, the critical question is not whether operators will enable AI like they did in cloud. It is whether they can avoid being unbundled from the value it creates.
For years, telecom operators have looked at cloud as the opportunity that slipped away. It's an understandable conclusion, but it’s wrong. Telcos didn’t miss participating in the growth of cloud; they enabled it. They carried the traffic, built the on-ramps, lit the metro fiber, and financed the internet capacity, which made hyperscale economics possible.
What they missed was benefiting from the value that cloud provided to their end customers. As a result, hyperscalers built cloud platforms, which became central to enterprise IT strategies, and their revenue grew exponentially because of it. So, while it is tempting to say that telecom missed the cloud, it didn’t. Participation was never the problem. Proximity to value was. The customer’s problem moved up the stack, into applications, platforms, orchestration, and developer ecosystems, while the telecom industry stayed one layer below the things the customer valued most.
Participation was not the problem. Proximity to value was.
That is what unbundling looks like. It is not exclusion from growth. It is participating in the market while becoming increasingly disconnected from the value customers are willing to pay for. Over time, the remaining contribution becomes standardized, interchangeable, and ultimately commoditized. For telecom operators, connectivity has often been the remaining contribution.
The same standardization and openness that enable networks to interconnect, which are critical capabilities for internet growth, have become one of a telco operator’s greatest liabilities. Why? Because they make it easy for customers to replace one connection with another. Over time, connectivity became interchangeable. For twenty-five years, telecom has enabled multiple growth markets holding the one piece in the bundle that could be cleanly replaced and shopped based on price.
Why does this keep happening?
Consider streaming video. Operators were not spectators to that transition. Most large operators sold pay television which included content, the subscriber relationship, advertising, content recommendations, and delivery. With internet growth, the delivery became standardized and interchangeable, but the content and the subscriber relationship did not. Operators retained the standardized, commoditized half and watched the other half become Netflix.
What came next is important to recognize. Content providers and content delivery networks placed caching hardware inside operator networks, and for the better part of a decade operators supplied the rack, the power, the space, and the proximity that made the experience work. Very little of the resulting value was recorded on the operator's side of the ledger. Notice that this is the same asset list, the same business model being repeated for AI edge inferencing.
Cloud then repeated the pattern at larger scale. Operators carried the traffic, connected the enterprises, and expanded the backbone. Connectivity remained essential throughout, but it turns out that essential does not mean defensible. The control points moved to the application, the platform, and the orchestration layer, none of which the operator held.
In our Telecom TV panel discussion in February on the AI-native telco, Francis Haysom of Appledore Research stated the requirement clearly: operators must become unbundle-able from the problem the customer is trying to solve. Yet, the cycles we just explored indicate how difficult that is to achieve.
In each of the business cases referenced, participation was not the problem, value-capture was, and this is the problem the telecom industry needs to understand if it is going to participate more fully in the AI-native value chain.
The uncomfortable part
Appledore’s analysis of the twenty largest operators worldwide across a ten-year period, provided by Haysom on the panel, found capital and operating expenditure essentially flat. Measured against hyperscalers, or against other capital-intensive utilities, that is not the investment profile of an industry moving to seize an opportunity.
Flat investment against a standardized asset leaves only one competitive lever—price. That is the uncomfortable reality. By the time a market is competing primarily on price, customers have already decided that one provider's offering is effectively interchangeable with another's. Price competition is not what precedes unbundling. It is what unbundling feels like when you're journeying through it.
The opportunity is the AI bundle, not traffic
The AI stack is a bundle like any other, including models, data, compute, orchestration, applications, and the connectivity between them. It is worth asking which of these elements is the most standardized, the most interchangeable, and the easiest to re-source without redesigning anything around it. The answer has not changed.
And several parties are already working to ensure it does not change. Hyperscalers are extending inference toward their own edge. Secure-access-service-edge (SASE) vendors are attempting to recreate premium performance over lower-cost transport through path selection, link aggregation, and error correction. Model and application providers are working to make premium, deterministic delivery unnecessary rather than pay for it. Different approaches, same objective: keep the network interchangeable so the value can be captured somewhere above it.
None of that makes AI a poor opportunity. It makes AI an opportunity with a known failure mode, which is a considerably more useful thing to know.
What does not unbundle
The previous post in this series argued that AI's distinguishing characteristic is not traffic volume but the demand for a guarantee measured against the workload rather than the link. The unbundling lens explains why that distinction is important commercially.
As I mentioned above, standardized solutions are often sourced on price. An underwritten one can only be replaced by a party willing to assume the same exposure, and most of the AI value chain is conspicuously unwilling to accept latency liability for someone else's production workload. A guarantee is a liability, and liabilities are not commodities, because no competitor can offer the customer a cheaper version of someone else's accountability.
The same logic applies to operational burden. Enterprises are increasingly deploying compute into factories, retail stores, and other distributed environments. What they do not want is the responsibility that comes with it: securing, patching, refreshing, managing, and supporting infrastructure across hundreds or thousands of locations. The challenge is not acquiring the hardware, it is operating it at scale. A less expensive server may reduce capital costs, but it does little to reduce the ongoing workload of securing, managing, and supporting it.
The solution, which could become part of an AI bundle, is a managed service, which offloads the burden—the support of the infrastructure—from the enterprise. And because the service isn’t open or standardized, it becomes more difficult to commoditize. When tied to connectivity, it becomes an upsell opportunity, which provides revenue growth.
It is worth noting that most telcos already have the assets and resources required to offer AI bundles: last-mile reach, metro fiber, real estate, power, sovereign position, and long-standing government and enterprise relationships.
Where to stand
The panelists in our Telecom TV webinar were unanimous that this is not an all-or-nothing proposition. No operator needs to own every layer of the AI stack, and no hyperscaler owns every layer either. What is required is a defensible position somewhere in the chain through which the telco operator can deliver a uniquely valued capability, whether that is deterministic networking, interconnect capacity, sovereign AI, edge inferencing, or GPU as a Service (GPUaaS). The specific choice matters less than making it deliberately rather than arriving there by default.
The operators most exposed are not the ones that choose wrong. They are the ones still waiting for the question to resolve itself.
The panelists were equally clear about method as well. They emphasized that telcos should not wait for the architecture to settle. Instead, telcos should choose a workload, choose a market, build the minimum viable version, and be prepared to fail quickly enough to learn from it. The operators most exposed are not the ones that choose wrong. They are the ones still waiting for the question to resolve itself.
The bottom line
Participation in the AI build-out is close to guaranteed. Value capture is not, and the industry has now run this experiment often enough to know the difference.
There is a test worth applying, and it does not require a strategy off-site: take the five largest enterprise relationships in the book and ask, for each, what the customer would actually lose by moving to a competitor. If the honest answer is a procurement cycle and a few milliseconds, that account is already unbundled. The contract has simply not caught up yet.
The alternative is easier than it may seem. Choose one workload, in one market, and underwrite its outcome rather than its link, not the entire architecture. One commitment the customer cannot buy more cheaply elsewhere, for the straightforward reason that no one else is willing to sign it. Test it, work out the bugs, then replicate it.
Operators did not miss the cloud. They were unbundled from it. Whether that outcome is written again about AI is a decision for telecom operators to make, rather than a highly accurate forecast.
This is the second post in a six-part series on the AI-native telco. The first post argued that AI is changing the economics of connectivity by shifting value toward guaranteed outcomes rather than best-effort transport. This post examined a familiar lesson from the cloud era: participation does not guarantee value capture. Next, we'll explore why agentic AI is unlikely to eliminate operational silos on its own, and why data models, context, and governance may prove more important than intelligence itself. Future posts will look at AI cloud interconnection, inference hosting, GPUaaS/CPUaaS, and infrastructure services, all focused on a single question: how can telcos secure a defensible position in the AI value chain instead of being unbundled from it?
Learn more at The AI-native telco: Capturing revenue opportunities in the AI value chain
By Author:
Steve Daigle,
Global Head, Telco Systems Engineers, HPE
Twitter handle: @sjdaig, X (Twitter)
LinkedIn account: https://www.linkedin.com/in/stevejdaigle/