AI-RAN Meets the Spreadsheet

by Brian Newman

The AI-Telecom Brief

There is a familiar pattern in telecom. A new technology arrives wrapped in superlatives, the vendor ecosystem rallies, and the trade press declares an inflection point. Then the capital plans get drawn up, and the economics assert themselves. The 3G spectrum auctions taught this lesson once. The early fiber-to-the-home builds taught it again. AI-RAN is now entering that same second phase, and the most telling signals are coming from the companies with the most to gain from its success.

Nvidia is no longer content to sell GPUs for the servers that sit behind the radio. The company is now developing an offer that goes directly into the radios themselves, a 6G radio unit chip. Jensen Huang has framed the software-defined base station as an "iPhone moment," a programmable platform capable of generating its own applications through rApps, xApps and dApps. The logic is to turn the radio into a developer surface rather than a fixed-function appliance.

The momentum behind that vision is real. Through 2026 Nvidia and Nokia announced AI-RAN collaborations with operators including T-Mobile US, SoftBank and Indosat, taking the technology outdoors and over the air, and Nvidia joined the Linux Foundation's Open CU DU ecosystem to push open-source RAN software. The architecture is converging, the alliances are forming, and the demonstrations are working. The question is not whether the technology functions. The question is who can afford to deploy it at scale, and when.

THE COUNTER-SIGNAL

The same vendors selling the future are beginning to flag its cost. Ericsson and Nokia have warned that higher chip prices arrive at a difficult moment, with radio access network margins already under pressure, a weaker dollar eroding Ericsson's heavily US-weighted revenue, and both companies having already cut tens of thousands of jobs to protect profitability. These are not skeptics on the sidelines. They are the firms whose order books depend on AI-RAN succeeding, and they are managing expectations on their own product economics.

Operators are saying the quiet part aloud as well. Vodafone Idea's chief technology officer has publicly acknowledged that AI-driven RAN holds tremendous potential, but that high costs remain a significant barrier to widespread adoption. The result is a market that looks expansive in narrative and narrow in absolute terms. One market estimate puts AI-RAN at roughly $150 million in 2024, growing toward $816 million by 2032 at a compound rate near 24 percent.

That is a healthy growth curve attached to a small base. It describes an emerging category, not a deployed one. 2026 is shaping up as a transition year, heavy on pilots and early commercial wins, but short of mass deployment of fully AI-native radio.

THE ENERGY PARADOX

The cost conversation has a second layer that rarely makes the headlines, and it cuts in two directions at once. Telecom is already one of the most power-hungry industries in operation. Operators consume an estimated 1 to 2 percent of global electricity, a figure that climbs as 5G expands and mobile traffic grows. Adding GPU-class compute into the radio layer increases that draw at exactly the moment when energy prices are under strain. Residential electricity prices in the United States have risen more than 36 percent since 2020 and are forecast to keep climbing through 2027, with AI data center demand cited as a contributing pressure.

The paradox is that the same intelligence driving consumption is also the most credible tool for reducing it. Early AI-driven energy pilots in the RAN have shown savings in the 6 to 10 percent range, with some vendors claiming up to 25 percent with no impact on customer experience, through dynamic power scaling, predictive maintenance and the elimination of unnecessary truck rolls. That tension defines the business case. AI-RAN justifies its own power and silicon bill only if the efficiency it unlocks outruns the load it adds. Operators that can prove that net-positive equation will deploy. Those that cannot will wait.

WHAT THE MONEY ACTUALLY SAYS

Budgets are rising, but they are being aimed with precision. In Nvidia's 2026 industry survey, 89 percent of telecom respondents expected their AI budgets to grow over the next twelve months, up from 65 percent a year earlier, and network automation has overtaken customer experience as the leading use case for investment, deployment and return on investment. That reordering is the real signal. The near-term return sits in operational efficiency, in zero-touch operations, anomaly detection and predictive maintenance, rather than in a wholesale rebuild of the radio.

There are credible proof points at that operational tier. Deutsche Telekom and Google Cloud have built AI agents that reconfigure radio parameters in seconds during major events or emergencies, work that previously took hours. The value there is concrete, measurable and bankable today. The value of a fully AI-native radio chip remains largely promissory, dependent on a 6G cycle that has not yet arrived. Commercial 6G deployment is generally targeted for the early-to-mid 2030s, with standardization expected around 2028 to 2030. The capital being committed now is buying optionality on that future, not revenue in this fiscal year.

THE OPERATOR'S READ

Having sat on the capital-planning side of large network deployments, the dynamic is recognizable. Vendors optimize for the technology roadmap. Operators optimize for cost per bit and payback period. When those two curves diverge, the roadmap waits for the economics rather than the other way around. AI-RAN will not be decided by what is demonstrated in a Barcelona exhibition hall. It will be decided by whether efficiency gains and new AI service revenue arrive faster than the silicon, energy and integration bills.

The synoptic point connects telecom to the wider AI economy. The same tension playing out in the radio layer is playing out in every enterprise AI program. The demonstrations are spectacular, the unit economics are unforgiving, and the organizations that win are the ones that treat capital discipline as a design principle rather than a brake. Deloitte has framed 2026 as the year the roar around AI gets quieter and smarter, with progress coming from the unglamorous work of making AI usable at scale rather than from headline-grabbing models. That description fits the radio as well as it fits the data center.

THE BOTTOM LINE

AI-RAN is not overhyped. It is mistimed in the public conversation. The architecture is sound, the alliances are durable, and the long-term direction is correct.

The deployment curve, however, is governed by a profit-and-loss statement, not a keynote. The operators worth watching are the ones pairing their AI ambitions with ruthless cost control and a clear-eyed view of the energy equation. They will set the real pace, and the vendors will follow them rather than lead them.

The industry has spent a year asking whether AI can run the network. The more useful question for 2026 is whether the network can afford to let it, and which operators will answer first.