Mobile network operators have been flirting with the cloud for some time…without much success. MEC was a failure, and various initiatives (remember Edge Gravity?) were attempts to match mobile networks with the cloud. Nvidia recently tried to force the issue by telling operators to put RAN and cloud on the same chip. None of that worked.
Mobile Experts has released predictions through 2032 that show the radio access network (RAN) and cloud markets going in wildly different directions. This is a sign that mobile networks are making the transition from “growth market” to “commodity market," while telcos are investing in computing as a new growth area. These recent trends are a prelude to RAN and cloud converging successfully.
Mobile RAN software
Mobile network software is quickly becoming a stand-alone business that is not tied to hardware. We’ve all heard this train coming into the station for several years, as operators have fully tested vRAN and are ready to use vRAN in the 6G cycle.
Last week, Ericsson released its “AI-in-RAN” portfolio, which is essentially a software upgrade that operators will buy through an annual subscription. That’s an important development that shows the transformation of the RAN market, from a hardware business to a software business. Operators are now willing to accept a subscription-based pricing model.
As a result, we think AI-RAN software revenue will grow by 100% CAGR through 2032.
We’re predicting that, in 2031, RAN software that’s licensed independent of hardware will have grown to 20% of the RAN market, from only 7% in 2025. That includes vRAN, AI-for-RAN and SMO software. This kind of dramatic transformation doesn’t happen often.
As the networks become cloud-native, they will be flexible and automated enough to support dynamic network slices and the demands of non-human customers. The RAN will be ready to transition from “best effort” to a solid platform that handles next-gen workloads.
RAN hardware
In a commodity market, the hardware moves toward one of two endgame positions: either a custom ASIC where one company dominates a large percentage of the market, or a hardened standard IC that captures the needs of multiple vendors. We see RAN and chip vendors fighting over these positions now.
Some of the big RAN vendors have moved away from custom silicon and are looking at off-the-shelf options. Other vendors continue development of custom silicon and are building tensor pools and DSPs into custom chips to handle their AI models. The result is a fragmented market for RAN baseband chips, ranging from 18A ASICs to x86 to ARM and even GPUs.
The GPU question is starting to be resolved, as we see Nvidia's fans in the mobile operator world rejecting the idea of ARC Compact Pro for the radio site, and asking for something scaled down even more. The details are in our recent report. I am now a believer in GPUs for RAN, as I see Nvidia finally adopting the advice I gave four years ago.
In the end, we expect the RAN market to migrate into an “Application Specific Standard Product” model. Qualcomm made an early try for this, with QRU and X100 products, but some customers felt hogtied by the Qualcomm stack.
Marvell and NVIDIA are now partnering on something that might have broad appeal. And x86 vRAN is ready for market, with many operators planning a scale-up. These competing products will be resolved over the next five years into a small number of simple choices. That’s the inevitable gravity of a commodity market at work.
Together, the changes in hardware and software will make the mobile network flexible and automated. The network will be reliable enough to become the spinal cord for future physical AI applications. In other words, the industry is on a path toward higher value for connectivity.
Telco compute
Many operators are shifting their capex money from networks to computing. Asian telcos are investing in gigawatt AI factories, with AI models trained and adapted to local languages and culture. In Europe, telcos are forming a consortium to invest together in edge computing data centers across the continent. None of this derives much value from the mobile network. Instead, its value comes from requirements for sovereignty and local control. This value is tremendous, and we’re anticipating $100B of new telco revenue by 2031 (excluding China).
One interesting note is that people in Singapore and Germany talk about “Sovereign AI” and “Sovereign Computing," but American operators don’t mention these words in the same way. Because the biggest hyperscalers and the big AI models are American, the American telcos don’t have much opportunity to compete with their own data centers or their own stack. Normally, the USA is a good indicator of future trends-but in this case the telcos are different than the rest of the globe.
In all regions, it’s clear that cloud and AI inferencing services are differentiated along the lines of sovereignty and control, not based on latency, packet loss or other factors related to the network.
Does this mean that there’s no hope for Mobile and cloud to fall in love?
Based on recent discussions with robotics companies and chip vendors, I believe that physical AI will bring these two spheres of influence together. Today, Gen-AI is all the rage. Gen-AI can tolerate high latency and doesn’t care much about networks. But as cloud computing and AI models shift to support physical AI, the network will matter.
Ten years from now, we expect that the big SKT gigawatt AI factory will be well adapted to physical robots running on an SKT network. That’s where the smart money is being invested…not just for one wave (Gen-AI) but for the rising tide (intelligent things).
Why did MEC fail?
In 2015-2023, the idea of MEC failed because operators had no edge in the market. Edge computing didn’t have an “edge”. Low latency wasn’t strong enough to create a market for telcos. This is a different angle: the operators will create a market based on sovereignty and then back into edge applications because they will own the stack and have a pool of developers.
This is how Adam Smith’s “invisible hand” works in the market: each player makes the moves that serve his own interests at the time….and somehow, thousands of self-serving decisions result in a good outcome for everyone.
In our case, the RAN guys will optimize the heck out of the network for operational efficiency and automation, making it a good platform for robots to rely on.
At the same time, telcos will be investing in sovereign compute platforms that will begin as centralized models, and they will be in position to migrate the computing platform to the edge when the robots are ready to roam the streets.
Mobile and cloud's meet-cute moment
Doesn't this all sound a lot like a romantic comedy meet-cute? A man and a woman meet by chance. Crazy barriers get in their way. Each one solves his/her own issues through a series of mishaps. In the end, our hero, the mobile network, has cleaned up its bachelor pad and decided to be more flexible. The lovely cloud has moved to Telco Town and her side hustle has grown to become a lucrative business. They give birth to a robot baby. And they live happily ever after. We hope.
Joe Madden is principal analyst at Mobile Experts, a network of market and technology experts that analyzes wireless markets. Disclaimer: Nokia is a client of Mobile Experts.
Opinions from industry experts, analysts or our editorial staff do not represent the opinions of Fierce Network.