Nokia AI-RAN Platform: NVIDIA-Powered Radio Comeback

The mobile radio business has been the ugly duckling of Nokia's balance sheet for years, a segment CEO Justin Hotard openly told investors was not delivering acceptable returns. On July 15, 2026, the company put a bet on the table meant to change that. The Nokia AI-RAN platform combines Nokia's own anyRAN software with NVIDIA's Aerial system, pitched as the industry's first commercial AI-native radio access network. The claim is that carriers can squeeze far more capacity out of the spectrum they already paid billions for, without a full hardware refresh. For any operator watching data traffic outrun capex budgets, that is the pitch that matters. Below is what the platform actually does, what the numbers really say, and where the "industry's first" label holds up.

This article covers:

1. What the Nokia AI-RAN platform actually does

2. The efficiency numbers, and the 2027 catch

3. Three ways operators can deploy Nokia AI-RAN

4. Why the NVIDIA partnership sits at the centre

5. The "industry's first" claim vs Ericsson's head start

6. What Nokia AI-RAN means for Justin Hotard's turnaround

7. Frequently Asked Questions

8. The verdict on Nokia AI-RAN

What the Nokia AI-RAN platform actually does

Nokia AI-RAN is a software layer that runs AI workloads inside the radio access network itself, promising more throughput from the same spectrum an operator already owns. It combines Nokia's anyRAN software with NVIDIA's Aerial system, and Nokia is selling the capability as a subscription rather than a box. The premise is simple: teach the radio to schedule, beamform, and adapt smarter, and the same tower carries more traffic.

That is a meaningful shift because for most of the last two decades, more capacity has meant more spectrum, more sites, or more baseband hardware. All three are expensive. Nokia AI-RAN reframes the problem as a software one, sitting on GPU-accelerated computing rather than the custom silicon that has defined radio for a generation.

The company has framed the launch, according to AI News, as one of the most significant shifts in radio network architecture in decades. That is vendor language, and it deserves scrutiny. But the underlying idea, that AI-native RAN can push more bits through the same megahertz, is exactly what carriers have been asking for.

The efficiency numbers, and the 2027 catch

Here is where the marketing meets the fine print. Nokia says the platform has already shown more than 20% spectral efficiency gains in testing, is targeting 50% by 2027, and is aiming for more than 100% by 2028, the point at which operators could roughly double the capacity of existing spectrum.

Read that timeline carefully. The 20% figure is a result. The 50% and 100% figures are targets. Nokia's own schedule puts pilots at the end of 2026 and commercial availability in 2027. So the headline numbers driving investor excitement about Nokia AI-RAN are, on the company's own admission, still two years out from a paying customer switching one on.

That is not a scandal, it is just how radio roadmaps work. But anyone reading the launch as a live product with double-the-capacity results today is reading it wrong.

What "spectral efficiency" actually means to a carrier

Spectral efficiency is bits per second per hertz, and it is the single number that decides whether a carrier needs to buy more spectrum at the next auction. A 20% lift is a real gain. A doubling, if it landed, would rewrite the economics of building 5G and eventually 6G networks. That is the difference between another spectrum auction and skipping one.

Three ways operators can deploy Nokia AI-RAN

One of the smarter parts of the Nokia AI-RAN platform is that it does not force a rip-and-replace. Operators can subscribe to the software and pick from three deployment paths, sized to how much of their existing kit they want to keep. That flexibility is important because most tier-one carriers have billions of dollars of AirScale gear already in the ground.

The three options Nokia has outlined:

1. A GPU-powered plug-in card for existing AirScale sites. The lowest-friction path. Add the card, subscribe to the software, and keep the site.

2. A standalone AI-RAN node. A purpose-built box for new deployments or focused capacity upgrades.

3. A cloud-server build delivered through partners. For operators pushing toward a cloud-native RAN architecture, aligned with wider AI-native RAN thinking.

The subscription model matters as much as the hardware choices. Radio has historically been a lumpy, capex-heavy business tied to hardware refresh cycles. Turning it into recurring software revenue is exactly the kind of shift telecom equipment vendors have been trying to engineer for a decade. For a parallel on how platform-versus-build decisions get framed elsewhere in AI, see our.

Why the NVIDIA partnership sits at the centre

You cannot understand the Nokia AI-RAN NVIDIA relationship without understanding what happened in October 2025, nine months before the platform launch. NVIDIA committed a $1 billion investment for roughly a 3% stake in Nokia, announced in October 2025 alongside the wider AI-RAN partnership. That deal put the chipmaker structurally inside the Finnish vendor's radio roadmap before the July 2026 platform announcement. By building on NVIDIA silicon and CUDA software rather than its own custom chips, Nokia can shrink a slice of costly in-house R&D. That freed budget goes toward software instead.

That is the shift Hotard has been describing to investors, moving away from a legacy hardware model. It is also, quietly, an admission. Nokia is outsourcing the silicon race to the industry's dominant AI chip supplier because it could not fix the mobile business on its own. The corporate context for the tie-up sits on NVIDIA's own newsroom, and the deal is what makes the partnership commercially rational for both sides, even as it creates a dependency.

And there is a dependency. Nokia CTO Pallavi Mahajan has acknowledged that at least some of the Layer 1 software is bound to the underlying hardware, a point reported by AI News in its coverage of the launch. Nokia points to merchant silicon from Marvell in its wider ecosystem and describes the platform as Open RAN-compliant. But the performance case it is selling, those spectral efficiency gains, currently runs through NVIDIA's stack, for which no equivalent alternative exists today. Openness in the messaging and NVIDIA dependency in the engineering are both features of the same launch. A useful contrast on how open-weight ecosystems handle the same tension sits in our write-up on.

The "industry's first" claim vs Ericsson's head start

This is where the Nokia AI-RAN pitch needs the most care. Ericsson began selling a commercial AI-in-RAN software subscription in June 2025. The Swedish rival says its platform delivers up to 20% higher downlink throughput and up to 10% better spectral efficiency across more than 15 live deployments, and, crucially, it runs on operators' existing baseband silicon with no GPU required.

By availability, Ericsson is already in the market. Nokia is not. Nokia's claim to a first rests on a narrower definition: a GPU-accelerated AI-RAN platform, a different architecture from AI features layered onto existing hardware. Both statements can hold at once, which is exactly why the framing deserves scrutiny rather than a straight repeat.

Two roads to the same destination

The architectural fork here is straightforward. Nokia has bet on GPU-native AI-RAN, tying its performance ceiling to NVIDIA's silicon and stack. Ericsson has bet on silicon-agnostic AI features that run on whatever baseband a carrier already owns. Same destination, more capacity per hertz, but two very different roads, and each carries a different set of risks for the operators buying in.

The divergence runs deeper than timing. Ericsson has kept its AI features silicon-independent by design to avoid exactly the kind of dependency Nokia has embraced. Nokia has tied its radio roadmap to NVIDIA. The bet is that the performance ceiling on GPU-accelerated AI-RAN will eventually pull ahead of what any silicon-agnostic approach can deliver. Qualcomm AI-RAN work and other silicon-side efforts sit in the same broader ecosystem, but no one else has yet paired an AI-native RAN platform with an NVIDIA-scale compute partner. For a deeper look at how vendor lock-in plays out in a different AI hardware category, see.

One route is faster to market and lower risk. The other, if the 2027 and 2028 targets land, has a potentially higher ceiling. Neither has been proven yet at commercial scale on a live tier-one network.

What Nokia AI-RAN means for Justin Hotard's turnaround

To read the launch only as a product story is to miss why it matters to Nokia as a business. Radio has been Hotard's hardest problem since he took over as chief executive in 2025. At the November capital markets day, he told investors the mobile business had not delivered acceptable returns, and he folded it into a new mobile infrastructure segment alongside further cost cuts.

The NVIDIA partnership sits at the centre of that repair job. According to AI News, analyst house Omdia has put the cumulative AI-RAN opportunity above $200 billion by 2030, and Nokia shares have re-rated on the strength of the company's AI and cloud momentum. The AI-RAN launch landed days before the second-quarter results, which is not an accident. 

Investors have rewarded the story, and the direction of travel is real. The open question is how much of the $200 billion Nokia AI-RAN can actually claim as a lead, given that a major rival is already selling into the same demand. The subscription model gives radio the recurring revenue its hardware cycles never did. The GPU dependency gives it a performance story that hardware-agnostic rivals cannot yet match on paper. Both of those are real assets. Whether they are enough to move Nokia's mobile business from cost-cutting to growth is the question 2027 will answer.

The verdict on Nokia AI-RAN

None of this makes the strategy wrong. Outsourcing the silicon race to the industry's dominant AI chip supplier is a defensible answer to a business Nokia had struggled to fix on its own, and turning radio into a software subscription gives the segment the recurring revenue its hardware cycles never delivered.

But the platform is not yet commercial, its headline efficiency numbers are still two years out, and at least one major rival reached the market first by a different road. For Nokia, this is a comeback in motion, not one already won, and its trajectory now runs, for better or worse, through NVIDIA. Watch the pilots at year-end 2026. That is the moment the Nokia AI-RAN story stops being an investor deck and starts being a network.

FAQs

What is Nokia AI-RAN in plain terms?

Nokia AI-RAN is a software platform that adds AI-driven optimisation to the radio access network, letting operators carry more traffic on the same spectrum. It runs on NVIDIA's Aerial computing stack and Nokia's anyRAN software and is sold to operators as a subscription. Deployment options range from a plug-in card for existing sites to a full standalone node.

When will the Nokia AI-RAN platform actually be commercially available?

Nokia's own roadmap puts operator pilots at the end of 2026 and full commercial availability in 2027. The 50% spectral efficiency target is set for 2027 and the 100%-plus target for 2028, so the biggest headline numbers are still two years out from paying deployments.

How is Nokia AI-RAN different from Ericsson's AI-in-RAN?

Ericsson's AI-in-RAN, launched in June 2025, is a software subscription that runs on operators' existing baseband silicon with no GPU required, and is already live in more than 15 deployments. Nokia AI-RAN is GPU-accelerated, built specifically on NVIDIA's stack, and is not yet commercially available. Ericsson leads on availability, Nokia is betting on a higher performance ceiling.

Is Nokia AI-RAN really Open RAN-compliant?

Nokia describes the Nokia AI-RAN Architecture as Open RAN-compliant and points to merchant silicon from Marvell in its wider ecosystem. However, CTO Pallavi Mahajan has acknowledged that some Layer 1 software is bound to the underlying hardware, and the specific spectral efficiency gains being marketed currently run through NVIDIA's stack, with no equivalent alternative today.

Why did NVIDIA invest $1 billion in Nokia?

NVIDIA committed roughly $1 billion in October 2025 for about a 3% stake, tying the chipmaker structurally into Nokia's radio roadmap. For Nokia, it funds a shift from custom silicon toward software. For NVIDIA, it extends its AI compute footprint into the telecom radio access network, a market Omdia values at over $200 billion cumulatively by 2030.

What does AI-native RAN actually mean?

AI-native RAN means the radio access network is designed around AI workloads from the start, rather than having AI features bolted onto existing baseband hardware. In practice that usually means GPU-accelerated compute at the base station and machine learning models driving scheduling, beamforming, and interference management. Nokia AI-RAN is one implementation of that idea, Ericsson's approach is another.