Nokia bets its radio comeback on NVIDIA, claims 20% spectrum boost
Curated by the Inblix editorial team
Nokia dropped what it calls the industry’s first GPU-accelerated AI-RAN platform on July 15, and the numbers it’s putting up are the kind that make mobile operators lean forward. The vendor says it has already demonstrated more than 20% gains in spectral efficiency, with a roadmap targeting 50% by 2027 and a doubling of capacity by 2028. That last figure is the real headline: the promise of wringing far more performance out of spectrum carriers already own, without a hardware rip-and-replace. Operators would buy in through a software subscription, with deployment options ranging from a plug-in card for existing AirScale sites to a standalone cloud build.
The launch is, at its core, a repair job. CEO Justin Hotard has been blunt that Nokia’s mobile business wasn’t delivering acceptable returns. The answer is a deep tie-up with NVIDIA, which invested $1 billion for roughly a 3% stake last October. By building on NVIDIA’s silicon and CUDA software instead of custom chips, Nokia is betting it can redirect costly R&D dollars toward software and recurring subscription revenue, a shift Hotard has framed as a clean break from a legacy hardware model. Investors have bought the story, re-rating Nokia shares sharply through 2026.
But the “industry’s first” label needs a hard squint. Ericsson already has a commercial AI-in-RAN software subscription in the market, running on operators’ existing baseband silicon with no GPU required. It claims up to 20% higher downlink throughput across more than 15 live deployments. Nokia’s claim to a first rests on a narrower definition: a GPU-accelerated platform, a different architecture entirely. Both things can be true, which is exactly why the framing deserves scrutiny rather than a straight repeat.
The strategic fork runs deep. Nokia has tied its radio roadmap to NVIDIA’s stack, and CTO Pallavi Mahajan has acknowledged that at least some Layer 1 software depends on the underlying hardware. Ericsson has gone the opposite way by design, keeping its AI features silicon-agnostic. Nokia points to merchant silicon from Marvell and calls the platform Open RAN-compliant, but the performance case it’s selling today runs through NVIDIA, for which no alternative exists. Outsourcing the silicon race to the dominant AI-chip supplier is a defensible bet for a business Nokia struggled to fix alone. It’s also a bet that the platform’s commercial fate, still two years out from those headline efficiency targets, now rides on a partner whose own roadmap Nokia doesn’t control.
💡 Key Takeaways
- Nokia has demonstrated more than 20% spectral efficiency gains on its new platform but is targeting 100% by 2028, which could double existing spectrum capacity.
- The platform is not yet commercial; pilots start late this year and general availability is not expected until 2027, placing its most ambitious claims two years in the future.
- Ericsson already sells a commercial AI-in-RAN subscription that runs on existing baseband silicon without a GPU, undercutting Nokia's 'industry first' claim depending on how you define the architecture.
Keep reading: See related articles below for more coverage on this topic.
Get smarter about AI
The sharpest AI news, curated daily. Delivered free to your inbox.