- Ericsson laid out a road map for operators to build AI businesses on 5G before 6G arrives
- The path runs through 5G standalone, AI for the network and advanced technologies such as physical AI, said Ericsson CTO Erik Ekudden
- Federal agencies need to move faster to free up spectrum for 6G, said Ajit Pai, CTIA CEO and former FCC chairman
ERICSSON INDUSTRY ANALYST EVENT 2026, BOSTON — Ericsson wants operators making money from AI on their 5G networks well before 6G arrives around 2030, building business that carries into the next generation, the company's top technologist said.
That strategy can help operators avoid repeating 5G's slow start. "I think we were way too slow in creating models that work for enterprises," said Erik Ekudden, SVP and group CTO at Ericsson, in the opening presentation at the analyst event here.
What does Ericsson's path to 6G look like?
The first step is using 5G standalone and slicing to move from best-effort to differentiated connectivity and to sell network APIs, fixed wireless access and mission-critical services. Most leading operators are there now, Ekudden said, but just over 90 of the roughly 390 service providers with commercial 5G have launched 5G standalone, the version of 5G that enables network slicing and other premium services, according to the June 2026 Ericsson Mobility Report.
The next step is applying AI to the network itself, with autonomous operations in selected domains, AI-driven energy optimization and features such as uplink capacity for AI glasses. The most advanced operators have reached that phase, Ekudden said.
Then, operators should add businesses in physical AI, edge inference, satellite coverage and sensing, Ekudden said. No operator has reached that stage yet, he added.
AI-native 6G follows, with new spectrum, a radio designed for massive IoT, integrated sensing and fully autonomous networks beyond 2030, according to Ericsson's slides.
Operators don't need to wait for 6G to implement AI. "No need to wait for new hardware or new infrastructure to bring AI into the network," he said.
More than 90% of Ericsson's 5G time division duplex (TDD) massive MIMO sites outside mainland China are already AI-ready, according to the company's slides. Ericsson's newest massive MIMO radios, introduced in February, add neural network accelerators — chip cores built to run machine learning models inside the radio — that deliver 10 times the compute power for AI inference. Existing Ericsson RAN Compute hardware can run AI with close to zero added energy consumption, based on field measurements.
Ericsson is seeing 20% spectral efficiency improvements in the field, Ekudden said. In trials with T-Mobile, Ericsson's AI-native link adaptation software raised spectrum efficiency by close to 10% and downlink throughput by up to 15%, the companies said in May.
The AI-6G connection
Ajit Pai, president and CEO of CTIA and a former FCC chairman, joined Ekudden by video. He said that AI and 6G go together. "To me, those two technologies are inextricably intertwined," Pai said. Physical AI such as robots will make the network more important, with connectivity serving as "the tissue that holds everything together," he added.
Will operators make money on inference?
Joe Madden of Mobile Experts asked how operators could compete with hyperscalers on AI inference. "I think that the honest answer is that we don't know," Ekudden said.
GPU-as-a-service may commoditize quickly into a hosting business, Ekudden said. Token services, which sell access to AI models billed by the token, could let operators differentiate on energy efficiency and network-aware models. Selling AI outcomes is very early and most likely to emerge in industrial and physical AI, he said.
Operators' edge assets could help them deliver inference as a business. Some operators have spare power at network assets — perhaps a megawatt or 10 megawatts —and not every inference job needs a data center drawing hundreds of megawatts, Ekudden said.
Ericsson's OSS/BSS software can already charge down to the individual token for some large customers. Operators should sell differentiated connectivity, APIs and accelerator capacity backed by service-level guarantees, orchestrated together across edge and cloud and priced by outcome or token, Ericsson said.
What does Pai want from Washington on 6G spectrum?
The FCC voted in July to auction 160 megahertz of upper C-band spectrum by July 2027, and it aims to hold three more auctions in 2028 in bands around 7 GHz, 4 GHz and 2.7 GHz. FCC Chairman Brendan Carr has publicly committed only to the 2.7 GHz auction so far.
Delay is the biggest risk. "One of the most powerful forces in the public sector, at least in the United States, is inertia," Pai said. Speed is the first lesson of 5G, he said: "There's a time value of spectrum."
The second lesson is that the most valuable applications are hard to predict. "Arguably one of the killer apps thus far in 5G may be utter disruption of the fixed broadband market," Pai said. But New Street Research expects fixed wireless access additions to plateau in 2026.
Ericsson projects uplink traffic will roughly triple every five years through 2035, while total traffic roughly doubles, according to its slides. Uplink traffic already grew faster than downlink at about four of every five of the 55 operators Ericsson measured.
Networks have been built for downloads for 25 to 30 years, Ekudden said. "We need to have new ways to start to reallocate resources to the uplink because that's where the value is. That's where there is a willingness to pay," he said. AI glasses, multimodal AI apps and robots are driving that demand, Ekudden said.
Learn more about AI and the network
T-Mobile fetes AI RAN innovation with Ericsson
Nokia, Ericsson disagree on how to apply AI in the network
Mavenir sees dollar signs for operators to make money via AI tokens