Nokia will pilot Nvidia-based AI-RAN at the end of 2026 and sell it in 2027, promising more than double spectral efficiency by 2028. Operators still lack a like-for-like field baseline and full deployment economics, while rivals say existing silicon can run AI features.
Nokia has attached a two-year capacity promise to a product that has not yet entered its announced pilot phase. Operators now have to decide whether programmability and future software gains justify changing the compute architecture of the radio network—or whether narrower AI upgrades on existing silicon offer the better return.
Nokia says pilots of its new AI-RAN platform will begin at the end of 2026, followed by commercial availability in 2027. The platform combines Nokia’s anyRAN software with Nvidia’s Aerial AI-RAN technology and is intended to support 4G, 5G and a later transition to 6G, according to the launch announcement.
The company says AI-driven radio work has already produced more than a 20% spectral-efficiency gain. It targets 50% in 2027 and more than 100% in 2028, which in Nokia’s formulation would more than double the capacity extracted from existing spectrum, particularly in dense cells.
Those percentages should not yet be read as a continuous, independently verified field series. The retained launch material does not specify the network configuration, traffic conditions or comparison baseline behind the demonstrated 20%, nor does it show that the 2027 and 2028 targets have been reached. Analysts interviewed for an independent report called for results from a highly optimized network and questioned whether GPUs are necessary for the proposed gains.
The available comparison with Ericsson also uses a different scope:
| Supplier claim | Comparison described in the source | Evidence status |
|---|---|---|
| Nokia: more than 20% demonstrated; 50% targeted in 2027; more than 100% targeted in 2028 | The retained materials do not identify an optimized-network baseline common to all three figures | Company demonstration claim and roadmap |
| Ericsson: close to 10% for an AI-native link-adaptation feature | Existing optimized algorithms, with the feature described as running in commercial networks | Rival company claim about one feature |
Ericsson’s technical account therefore provides competitive evidence that AI can improve an optimized commercial RAN, but it does not validate or disprove Nokia’s broader platform forecast. The figures are not like-for-like.
Nokia is offering operators several ways to adopt the platform rather than requiring one network-wide replacement:
| Deployment path | Starting point | Constraint or advantage |
|---|---|---|
| AirScale capacity plug-in | Existing Nokia AirScale baseband | Preserves racks and radios and offers the simplest installed-base upgrade, but concentrates the advantage among existing Nokia customers |
| Standalone AI-RAN node | New sites, extensions or mixed deployments | Combines GPU and CPU compute without the plug-in’s power-headroom limit, but adds a new appliance |
| Cloud-native COTS servers | Virtualized or cloud RAN | Can expose spare compute to other AI workloads, but must still meet time-sensitive RAN performance requirements |
The easiest path is not equally open to every operator. An account of the product mix says most services available through the Nvidia systems are reserved for operators already using AirScale, even though Nokia says anyRAN can work with other vendors’ equipment and Open RAN-compliant radio units. Open radio interfaces do not by themselves make baseband compute or upgrade economics vendor-neutral.
For the AirScale replacement, Nokia executives said the Nvidia product must fit the same 190-watt power envelope as the current Marvell-based unit so customers do not have to replace cabling or racks. Nokia head of RAN Mark Atkinson also pledged that the GPU-based products would not cost more than their Marvell-based predecessors, according to an interview about the architecture. These are supplier commitments; the sources contain no independent power measurement, operator total-cost analysis or subscription price.
Nokia’s case is that programmable merchant silicon can escape a pattern in which new software has forced a silicon replacement every three to five years. Its new subscription model would deliver algorithms and performance upgrades between hardware cycles. But the commercial model is not itself new: the same product account notes that Ericsson introduced an AI-in-RAN subscription first.
Nokia is replacing one Layer 1 dependency with another. Its most demanding baseband code is moving from Marvell custom silicon to Nvidia’s CUDA-based platform, while Nokia says a hardware-abstraction layer will keep as much code portable as possible. Atkinson described the forthcoming ABIP unit as Nokia’s final custom-silicon baseband product. Marvell remains in current Habrok massive-MIMO radio units, and Nokia is still evaluating the processor for a next-generation Habrok expected around 2028.
That architecture could give Nokia access to Nvidia’s large software ecosystem and more room to introduce complex models. It also exposes operators to a supplier-specific programming platform. Ericsson executives have warned about CUDA lock-in, while their company promotes purpose-built ASICs and vectorized CPUs as sufficient for many edge AI workloads.
Ericsson’s alternative is not costless or effortless. Its own technical account says real-time RAN inference can run on microsecond deadlines and that commercial-grade models need much finer, better-curated telemetry than conventional network measurements provide. The competitive dispute is therefore not AI versus no AI. It is whether the performance headroom of GPUs outweighs their hardware, energy and software-dependency costs for each workload.
Other vendors have not converged on Nokia’s answer. EJL Wireless Research President Earl Lum told the analyst report that Huawei and ZTE are not putting GPUs in base stations; the report also described Ericsson and Samsung as keeping multiple compute options open. That does not prove Nokia’s approach is wrong, but it makes the claim that GPU deployment is the natural industry path something operators can test rather than assume.
Nvidia is more than an arm’s-length chip supplier. Nokia announced in October 2025 that Nvidia would invest $1 billion through a directed issuance of 166,389,351 new shares at $6.01 each. Nokia said the proceeds would support its broader connectivity strategy and general corporate purposes, while it would accelerate 5G and 6G RAN software for Nvidia architecture, according to the investment announcement. The proceeds were not described as ring-fenced for this platform.
The issuance subsequently closed, and the new shares represented about 2.9% of Nokia’s enlarged share count, the company said in its completion notice. The relationship combines Nvidia’s role as silicon supplier with equity ownership. That financial alignment is not independent evidence for the roadmap’s performance.
The market forecast also favors a selective upgrade cycle over a wholesale GPU replacement. Dell’Oro Group expects more than $35 billion in cumulative AI-RAN revenue from 2026 through 2030, but little if any incremental RAN revenue by the end of that period, according to an analysis citing the firm. Dell’Oro expects near-term demand to favor AI-for-RAN software and existing-hardware upgrades; it forecasts the GPU-based RAN segment will surpass $1 billion by the end of the period. The five largest suppliers already held about 96% of RAN revenue in 2025, so subscriptions may reinforce installed vendors more than expand the market.
Nokia presents another possible economic return: use spare edge compute for non-RAN AI jobs. SoftBank says it has integrated its AITRAS Orchestrator with a Nokia platform that allocates shared resources to external workloads when internal RAN and AI demand leave capacity underused. But the project description says large-scale validation is still planned. It does not establish paying demand or show that external jobs can reliably coexist with time-critical radio work.
Nokia’s announced end-of-2026 pilots can resolve the central question only if operators publish comparable results. The evidence needs to show:
T-Mobile US is Nokia’s lead trial partner, and Nokia cites work with SoftBank, Deutsche Telekom, Telia, Orange, BT and Indosat Ooredoo Hutchison. Yet not every deployment option has been publicly confirmed with every operator, and the trial report notes that Nokia is no longer a main wireless infrastructure supplier at AT&T or Verizon.
The next decision is not whether AI belongs in the RAN; both Nokia and Ericsson treat that as the direction of travel. It is where a GPU produces enough additional, measurable value to beat a software upgrade on existing compute. Until operator data answers that question on a common baseline and at a disclosed total cost, Nokia’s 2028 doubling claim remains a product roadmap rather than a network planning assumption.
Zhipu reportedly reached $1 billion in annual recurring revenue in July, roughly four times a March estimate, but the unconfirmed run rate is not annual sales and still sits far ahead of recognized cloud revenue while margins remain thin.
Anthropic’s 20-year lease gives TeraWulf a customer for 401 megawatts of future AI infrastructure, but rent starts only after delivery and the proposed utility deal passes power and grid costs to TeraWulf, leaving financing and project margin unresolved.
OpenAI has put conversations and Projects back in its redesigned ChatGPT desktop app and enabled cloud Work threads to move across devices, correcting the launch's biggest usability failures without merging local Work or Codex histories.
Twenty-nine countries signed an agreement creating WAICO as an independent intergovernmental organization, while China paired the launch with capacity-building offers that are not yet confirmed as WAICO programs.
Meta reportedly plans to put departing AWS compute executive Dave Brown to work on its data-center buildout, adding hyperscale operating experience while leaving any customer-facing cloud business conditional and undefined.
Moonshot AI has made Kimi K3 available through its apps and API, pairing a 2.8-trillion-parameter architecture with early frontier-level results, but the model's open-weight claim cannot be tested until its weights and technical report arrive.
CIA Director John Ratcliffe said US intelligence is consistent with an estimate that Russian recruits last 20 to 30 minutes on Ukraine’s battlefield, but the public trail leads to an unsourced claim about assault troops and does not establish a representative average.
China has paired a five-year AI training offer for developing countries with cooperation centers, a weather-warning rollout and a new 29-country organization. The package gives Beijing a platform for influence, but no budget, selection rules or delivery timetable has been published.
Nebius has arranged its first senior secured facility against an operating GPU deployment and one customer’s cash flows. The deal adds project-level debt to its expansion toolkit, but an unnamed customer and a larger rival financing limit what it proves about the rest of Nebius’s backlog.
Huawei publicly displayed a 16-cabinet Atlas 950 configuration rated at 1 EFLOPS in FP8, providing tangible evidence of its system-scale AI strategy while leaving price, power use and sustained workload performance undisclosed ahead of the full system's planned fourth-quarter release.
UK testing places leading open-weight models four to seven months behind selected closed-model cyber results, yet longer attack chains, U.S. benchmarks and mixed cost comparisons show why that interval is a warning signal rather than a universal capability clock.
Moonshot AI's Kimi K3 added to a global technology selloff with near-frontier performance, but unreleased weights, mixed cost comparisons and a recommended 64-accelerator deployment leave its effect on chip demand unresolved.
Two binding EU decisions require Google to give rival AI services comparable access to 11 Android features and offer eligible search competitors a restricted, anonymized dataset, but phased deadlines, certification, pricing and privacy safeguards leave the competitive effect unproven.
Databricks has signed a term sheet for a Coatue-led financing at a $188 billion valuation, while unidentified sources put the round at $3 billion. The proposed capital would deepen its push into AI governance, data agents and operational databases, but the transaction remains open and the company supplied no new operating figures.
OpenAI says a handful of GPT-5.6 Sol file-deletion reports most commonly involved Full Access without sandboxing or Auto-review. Its own evaluations show a more complicated risk picture, and the company has not yet published an incident rate or evidence that its promised safeguards stop the failure.
Netflix says generative AI workflows were used on roughly 300 titles in 2026, but its only quantified example tied to that disclosure covers 17 minutes and does not establish how much finished material, spending or labor changed across the slate.
A rogue browser extension can trigger Claude for Chrome’s fixed Gmail, Docs, Calendar and business workflows; the attack is constrained by default approvals but can run silently for users who enabled unattended action.
Gemini 3.5 Pro missed its expected June rollout. An anonymous-source account says a late-June data change fell short of Google's coding goals, but Google has confirmed only partner testing—not the reported cause, a new date, or public results and pricing.
Gold Eagle has begun collecting and prioritizing AI-discovered software vulnerabilities, but the voluntary federal clearinghouse has not disclosed results, operating rules or the resources that would turn findings into deployed fixes.
Google is bringing Gemini Omni editing and a reusable face-and-voice avatar to Vids, but the sharper distinction is account-level identity across Vids and Gemini rather than a new category of AI video; Vids already offered Veo generation and customizable avatars, while specialist rivals already sell digital presenters.