Every one of Nokia's four cloud deals this quarter reads the same on paper. Nokia links its network data to Google Cloud in June, to Amazon Web Services and Databricks a week later, then to Microsoft Fabric in September, each time promising the same outcome: operators get usable network data in minutes instead of weeks. Read in sequence, the four deals point one layer above where Nokia's real competition sits: the chip running the radio.

Nokia Repeats the Same Pitch Four Times

The pattern started June 22, when Nokia and Google Cloud expanded an existing partnership to embed six Gemini-powered AI agents inside the Nokia Assurance Center, aimed at cutting the manual troubleshooting that eats into service reliability (Google Cloud, 2026). Two days later, Nokia said its Autonomous Networks Fabric would ship on Amazon Web Services, pairing intent-based networking with Amazon Bedrock and SageMaker, while a separate proof of concept ran the same data-unification pitch through Databricks (RCR Wireless, 2026). By September 17, Microsoft had its turn: Nokia Data Suite plugged into Microsoft Fabric, closing the same weeks-to-minutes gap operators keep hearing about, and Nokia's stock rose 6.3% in midday European trading on the news (Quartz, 2026).

The partnerships share a common value proposition, even though they differ in platform and implementation.

Telecom data has been fragmented across vendors and formats for decades, so any one of these integrations could cut the weeks it takes an operator to get usable data in front of an AI agent. Stacking four of them into ninety days points to a separate motive: Nokia does not yet know which cloud its customers will standardize on, so it is reducing integration friction across all of them at once, ahead of the answer. Switching costs stay low and the deals are quick to sign.

The Bet That Isn't Cheap

One layer down, Nokia made a choice it cannot easily reverse. In October 2025, Nvidia took a $1 billion equity stake in Nokia tied to a commitment: Nokia's future radio access network software, an approach the industry calls AI-RAN, would run on Nvidia graphics processing units, or GPUs, instead of purpose-built chips (TechTimes, 2026). The AI-RAN platform built on that bet launched July 15, 2026, targeting more than 100% spectral efficiency gains by 2028, which would roughly double the traffic operators can carry on spectrum they already license (TechTimes, 2026).

Ericsson rejected that architecture outright. Thirty-four days before Nokia's platform launched, Ericsson already had its own AI-RAN running in more than 15 live operator deployments, built on custom silicon it ships inside its Massive MIMO radios rather than new GPU hardware (TechTimes, 2026). At Mobile World Congress in March, Ericsson deepened a separate, decades-long relationship with Intel, betting future Ericsson silicon on Intel's process technology and Intel Xeon processors for its Cloud RAN, and explicitly rejecting the idea that GPUs are necessary for 6G at all (Ericsson, 2026).

Samsung has landed on Ericsson's side of that argument too.

Radio processing has to finish inside a half-millisecond window. Ericsson and Samsung argue that constraint caps useful AI models at tens of thousands of parameters, a size where purpose-built silicon beats a GPU on cost and power every time. Nokia is betting that constraint loosens as its software matures, and that the extra compute headroom a GPU provides is worth paying for now, ahead of the physics catching up.

“6G is not merely an iteration of mobile technology.”

Börje Ekholm, Ericsson CEO (Tech Monitor, 2026)

What the Open Layer Protects

If Nokia's GPU bet works, the cloud deals reduce integration friction, but they do not validate Nokia's AI-RAN performance or economics. Operators will buy AI-RAN because it performs, and whichever data platform sits underneath becomes background plumbing. If the bet stalls, and Ericsson's already-shipping alternative keeps winning trials on cost and power, the multi-cloud data layer does not rescue the RAN business on its own. It means Nokia fights that battle on someone else's infrastructure instead of its own.

That asymmetry is worth naming plainly for anyone evaluating Nokia as an infrastructure partner right now. Their strategic value is more commercial and operational than architectural. They lower switching costs for customers who have not yet picked a side, which is a real advantage in sales cycles, but it answers a different question than whether the underlying radio architecture performs.

What Happens Next

• Nokia's 2026 field trials with T-Mobile and eight other operators move into 2027 commercial pilots. The spectral-efficiency numbers from those trials, not the cloud deals, will be the first real evidence either side is right.
• Ericsson's upcoming earnings disclosures are the place to watch for whether its 15-plus live AI-RAN deployments are converting into contract wins large enough to offset Nokia's Nvidia-backed marketing advantage.
• Whether a fifth hyperscaler joins Nokia's data layer before the AI-RAN trials report results will say something about how much confidence the market still has in the 2028 timeline.