Ericsson and Nokia at odds on AI-RAN silicon

Ericsson CEO Per Narvinger used the Q3 2026 earnings call on 15 October to defend custom ASICs for AI-RAN, according to RCR Wireless on 16 October. Asked by Deutsche Bank's Robert Sanders whether GPUs in the baseband had been ruled out, Narvinger said Ericsson is not philosophically against anything and will pursue what makes business sense. Its software and hardware are disaggregated, so GPUs remain an option, although they would require new hardware at the site. The function Ericsson is selling today, AI-native link adaptation, runs on basebands already deployed. Ericsson reports a 10% spectral efficiency gain and says the function runs on more than 60,000 base stations at 3 operators, including Optus, with AT&T, Bell Canada and T-Mobile reportedly testing it. All figures here come from RCR's account of the call. 

Narvinger put a 10% spectrum efficiency gain at about $800 million, against the $8 billion SpaceX was reported to have paid for spectrum the week before. RCR does not say what base the $800 million is calculated on, so I treat it as an illustration, not a measurement. The comparison does clarify the nature of the benefit. Spectral efficiency is capacity gained on spectrum already licensed, which is cost avoidance, the category I separated from new revenue in AI monetization for operators: separating revenue from cost avoidance. On a single cell the value arrives when a capacity upgrade is deferred, and Vish Nandlall's RCR Analyst Angle of 13 October put that at about $18,000 in present value for a cell at 91% load with a $200,000 upgrade deferred by about 1 year. An average gain across 60,000 sites says less than how many of those sites sit near an upgrade trigger.

Light Reading reported on 12 October that Nokia is building its AI-RAN on Nvidia GPUs and has stopped new custom silicon, with operators testing this quarter and commercial availability in early 2027. Ericsson's answer is that value can be delivered now on silicon operators already own. Both positions have an economic logic, and neither has been tested against the other in a published cost comparison. This is the gap between field-validated and commercially validated that I described in Is AI-RAN the future of telco?. A footprint of 60,000 sites is a field result for a software function on existing ASICs. It does not show that a GPU baseband earns its premium. Where accelerators are needed, the position in Non-RT RIC, AI-RAN, and the AI Grid applies: deployment starts where power, cooling and fibre are already in place, which points to the central office before the cell site.

Component costs add to the comparison. RCR reports that Ericsson beat Q3 forecasts on 7% organic sales growth, helped by customers buying ahead of expected AI-related price rises, and warned that AI-driven component costs will pressure margins through 2027. 

As you know if you follow this blog, I am skeptical of the cost-benefit of putting GPUs in a base station. The costs are high and growing, both form a capex and an energy perspective, while the benefits are entirely speculative. Would Nokia have taken the same architectural decision without NVIDIA's 1B$ investment? Unlikely. Between data centers, AI-RAN, sovereign AI, and AI factories, NVIDIA plays on a lot of different fields in telecom. They can afford for some of these explorations to backfire. Can Nokia afford to fail the next generation of radio architecture by betting it all on GPU?

Ericsson Makes Its Case for Edge Inference

Ericsson published a blog on 7 October, written by Benedek Kovacs, Wolfgang John and Mischa Dohler, arguing that where AI inference runs is becoming a strategic decision for enterprises and that communications service providers are well placed to answer it. Ericsson draws on an F5 report for the shift toward enterprises running some inference themselves. The blog states that service providers operate about 100,000 distributed network data centres, which could support over 100 GW of new AI compute over time. It names 3 advantages for the telco edge: access to live network context such as radio conditions and device location, stable low latency for physical AI, and confidential computing that keeps data private from the operator and from co-located workloads. It adds that a Swedish regulatory sandbox assessed the confidential computing approach as compliant with current data protection rules, subject to user-controlled attestation and key management.

The 100 GW figure divided by 100,000 sites implies an average of about 1 MW per site, which is my arithmetic and not Ericsson's. The blog gives it as an eventual ceiling with no timeline and no split between sites that have the power, cooling and fibre today and those that would need a rebuild. Those 3 constraints are why I placed AI Grid deployment at central offices and mobile switching offices rather than at the cell site in Operators Lean In On AI Grid Location. A count of 100,000 sites says little about how many of them can host accelerated compute.

Ericsson's own performance chart points the same way. At 30 tokens per second it shows roughly 7.0 billion parameters running on a device and roughly 30.8 billion at an edge site, which is the range of small and mid-size models, not frontier models. The workload Ericsson builds the latency case on is physical AI: robotics models of 10 billion to 100 billion parameters, control loops of 10 Hz to 20 Hz and responses under 100 ms. Those are real requirements, and they define a narrower market than general enterprise inference.

The operator role Ericsson describes

The more notable statement is about 6G. Ericsson expects specialised AI providers to deliver most early edge AI services on top of telco infrastructure, with the service provider's value in orchestration, network API exposure and end-to-end assurance. That is a vendor describing the operator as the infrastructure and assurance layer, not the seller of the AI service, and it is a shift from the position I covered in Ericsson Says the Telco Edge Was Too Early. Operators that accept that role should price edge sites on power and connectivity economics at the central office and metro layer, and should treat network context as a paid input that third-party AI providers consume through APIs.