0

AI infrastructure investment continues to accelerate; the next big question is how to turn this capex supercycle into repeatable, monetizable business outcomes — and what role will telecoms play?

The telecoms industry is quite capable of deploying and operating technology infrastructure at enormous scale. The more persistent problem is turning new capabilities into products that enterprises can readily buy and use. In the context of the AI economy, operators have vital assets that could help bridge the shift from training AI models to the adoption of inference as a key part of consumer behavior and enterprise workflows. 

At the Mobile Future Forward event earlier this month in Seattle, event host Chetan Sharma put it succinctly. “Deployment is not diffusion,” he said. The functionality of a network feature or the impact of a piloted AI application are not necessarily indicators of adoption pace. He suggested measuring the time between the first deployment and the hundredth as a more accurate bellwether; that timeline would speak more directly to developing technologies that actually fit a market gap or need. 

As AI investment continues to rise, so do the stakes. Sharma contrasted the pace of data center spending with wireless capital expenditure, and questioned when this infrastructure spending would translate into a corresponding application economy. For operators, the tension is between building out capacity ahead of expected AI-driven demand, and what real consumers ultimately do with AI — and how willing to pay for that. 

Lessons learned from 5G

Ask any telecoms industry veteran whether 5G was a success or a failure and you’ll likely hear some version of the response, “Well, it’s complicated.” As compared to the hype cycle that ushered 5G into our hands and homes, 5G did not deliver a nebulous global transformation of industry, nor did it help operators broadly right the ship financially. But it did generally make networks perform better and reach further. In consumer-focused markets like the US, it brought us fixed wireless access, the lasting impact of which may have less to do with home broadband and more to do with the economics of a home broadband market historically dominated by cable companies. In other markets, particularly China, 5G did usher in industrial overhauls resulting in smarter, more efficient cities, factories, logistics hubs, hospitals, ports, and so on. 

The point is that two things can be true: 5G can be viewed as both a disappointment and a success depending on how you look at it. The next point is that there are different advantages that accrue when a particular company or country is better at technological innovation or better at diffusing technology. 

“Is the technology coupled with the market?” is how Sharma framed it. And the answer depends on a lot more than how well that technology performs. Buyers have different risk tolerance. Enterprises have to integrate new tech into potentially brittle workflows and persuade workers to use what they’ve purchased. These points of friction explain why a successful deployment may still be an isolated one. 

AI presents a sharper version of the same problem. It can demonstrate value quickly, yet the path from an impressive application to a repeatable operating model runs through data, governance, workflow changes and accountability for results. More infrastructure may increase what is possible without resolving what an enterprise can practically adopt.

From bespoke adoption to repeatable outcomes

Cisco’s Masum Mir drew a distinction between adoption and diffusion at the forum. Enterprises can commission bespoke AI systems and get meaningful results. Diffusion begins when the outcome becomes easy for others to consume without rebuilding the solution each time.

For telecom, that changes the commercial question. Mir characterized AI as a demand shock for communications infrastructure, but simply satisfying that demand would leave operators selling more capacity into somebody else’s application. His challenge was to focus on industries and the outcomes they need. As he put it, “What is the business outcome we can deliver to the customers and industries that we serve?”

Consider an enterprise trying to reduce the time required to unload a shipping container. It does not set out to purchase a specified number of gigabytes or network minutes. It wants a faster, more reliable process. Connectivity may be essential to that result, alongside devices, applications, data and AI. The service becomes easier to buy when someone can assemble those components, stand behind their performance and show that the process improved.

Operators have assets that could help. Mir pointed to their enterprise relationships, reputation for trust and experience delivering service-level commitments. They can manage connectivity, protect data and enforce policies at scale. But those strengths do not automatically become an AI product. The operator has to apply them to a defined customer problem and work with partners that understand the application and the industry.

That requires a different form of experimentation. Internal AI projects can improve operations, and those gains are useful. But Mir warned against letting fragmented efficiency projects consume the entire opportunity. “If we stop at just automating, we are not going to get to the other end of creating net new value,” he said. His prescription was to dedicate resources to new businesses and measure their commercial success.

The measure should extend beyond the number of trials. How long does it take to bring a second customer on board? Which integrations must be repeated each time? Can the operator and its partners commit to an outcome? Does the offer solve a problem that an enterprise budget owner recognizes? Those questions reveal whether an idea is spreading or merely being demonstrated.

A better starting point for 6G

This distinction should shape 6G while there is still time to make architectural choices. AI-native networks, sensing, distributed compute and more capable devices may create important new possibilities. A feature list alone cannot establish demand for them.

Sharma urged the industry to seek real-time information from the market even when the evidence is incomplete. That means testing proposed capabilities against buyer needs now, finding where adoption gets stuck, and changing the product or partnership before the next network generation is fully specified. Mir made the timing plain. “I think it’s super, super important now to not wait for the next G.”

The opportunity for telecoms is to make network capabilities part of outcomes that customers can trust and repeatedly obtain. Deployment proves an operator can deliver the technology. Diffusion proves the market has found a use for it. For AI — and eventually for 6G — the distance between those two achievements may determine how much value the industry captures.