Beyond the First AI-RAN Call: T-Mobile's Dual-Track Bet and What It Means for Open RAN Operators

October 28, 2026
9 mins
Technology
5g ai ran ericsson nokia nvidia open ran t mobile telecom

Beyond the First AI-RAN Call: T-Mobile’s Dual-Track Bet and What It Means for Open RAN Operators

Executive Summary

T-Mobile’s completion of the world’s first AI-RAN call on a live commercial network—using Nokia’s anyRAN software and NVIDIA’s Aerial RAN Computer (ARC) platform—has been widely covered as a milestone. But the more consequential story for Open RAN operators is the carrier’s parallel, pragmatic bet on two distinct AI-RAN paths: one GPU-accelerated (Nokia/NVIDIA) and one software-only on existing silicon (Ericsson). The dual-track approach, revealed in T-Mobile’s own disclosures and third-party reporting, offers a template for how operators can capture near-term efficiency gains while positioning for a GPU-enabled future. For Open RAN vendors and operators, the message is clear: AI-RAN is not a single product but a continuum—and the winners will be those who can deliver value at every point along it.

The Milestone: A Live Call, Not a Lab Demo

On October 26, 2026, T-Mobile, Nokia, and NVIDIA announced the completion of the first AI-RAN call on a commercial 5G network in Bellevue, Washington. The call ran over T-Mobile’s live production network using a commercial smartphone, a Nokia Massive MIMO radio, Nokia’s 5G anyRAN software, and servers built on NVIDIA’s Aerial RAN Computer (ARC) platform. Crucially, it connected to T-Mobile’s existing 5G Standalone core—no isolated testbed, no lab rig.

This is a genuine technical achievement. AI-RAN enables network infrastructure to handle both connectivity and AI workloads on the same computing platform, potentially improving spectrum efficiency and enabling new services closer to users. The fact that it worked on a live network with commercial hardware and a real handset is a significant step beyond the lab trials that have dominated AI-RAN headlines for the past two years.

But the milestone alone does not prove that AI makes calls better. As TechRadar noted, the most concrete performance numbers for the Nokia/NVIDIA platform ironically came from a lab: Deutsche Telekom reported typical uplink throughput gains of 10% to 15% from AI-based channel estimation in initial tests in Bonn and Oulu, with a field trial still to come. T-Mobile did not specify how many sites were involved in the Bellevue call. So while the call is a world-first, it is not yet a commercial-scale deployment.

The Real Story: T-Mobile’s Dual-Track Strategy

The same week, T-Mobile’s chief network officer Ankur Kapoor published a blog post detailing results from AI-managed beamforming trials with Ericsson in New York City. The AI model ran on Ericsson’s own silicon—not NVIDIA GPUs—and delivered a 16% improvement in downlink user throughput and 12% better spectral efficiency during peak hours. These gains came from software updates to equipment already in the field, a point emphasized by Omdia analyst Gabriel Brown, who called it a “wow moment” because the improvements were achieved without new hardware.

This is the crux: T-Mobile is pursuing two parallel AI-RAN strategies. One is the GPU-accelerated, cloud-native path with Nokia and NVIDIA, aimed at creating a distributed AI compute platform. The other is a software-only, silicon-embedded path with Ericsson, aimed at extracting immediate efficiency gains from existing infrastructure. Kapoor himself framed AI-RAN as “a continuum, not a single product,” and said the carrier is developing AI-RAN software with both vendors.

For Open RAN operators, this dual-track approach is instructive. It acknowledges that the journey to AI-native networks will be incremental, with different technologies maturing at different rates. It also highlights a tension: GPU-based AI-RAN promises long-term flexibility and new revenue streams, but software-based AI on existing silicon delivers near-term ROI with lower risk.

The Economics: Where the Money Is

The financial case for AI-RAN is still being built. SNS Telecom & IT forecasts that annual spending on AI-RAN automation and optimisation will reach USD $1.1 billion by 2030. That figure covers software and services that use AI and machine learning to automate and optimise mobile radio networks, including service management and orchestration (SMO) platforms, non-real-time RAN intelligent controllers (RIC), and AI-infused RAN software. While$ 1.1 billion is a meaningful market, it is a fraction of overall RAN capex—and it underscores that the near-term opportunity is in software and automation, not necessarily in ripping and replacing hardware.

The report also notes a market shift from older self-organising network (SON) technology to more open, application-based architectures built around the RIC and SMO. In this model, rApps running on the non-real-time RIC layer are emerging as a preferred route to network-wide automation across multi-vendor environments. Operators including AT&T, Verizon, Telus, Vodafone, Deutsche Telekom, Swisscom, and Telstra are increasing adoption, while Rakuten Mobile, China Mobile, STC, and TDC NET have reached limited levels of highly autonomous operation for specific use cases.

This aligns with T-Mobile’s dual-track approach: the Ericsson software-based AI beamforming is a form of RAN automation that can be deployed today, while the Nokia/NVIDIA GPU platform is a bet on future AI workloads at the edge. For Open RAN operators, the implication is that investments in RIC/SMO and rApps may yield faster returns than waiting for GPU-based AI-RAN to mature.

The Vendor Landscape: Nokia’s Urgency and NVIDIA’s Ambition

Nokia’s position in AI-RAN is complex. The company has aggressively marketed its anyRAN software and AI-RAN platform, and it has announced collaborations with nearly 20 telecom operators worldwide. Yet, as Light Reading’s Iain Morris pointed out, Nokia badly needs big AI-RAN deals—and a deal with SpaceX’s Starlink could be a game-changer. Starlink’s reported talks with tower companies for up to 80,000 sites suggest that a GPU-powered RAN from a well-funded entrant could disrupt the traditional vendor landscape.

NVIDIA, meanwhile, is positioning its ARC platform as the compute foundation for AI-RAN. The company’s collaboration with Nokia and T-Mobile is a showcase, but NVIDIA’s broader strategy involves turning mobile networks into distributed AI computing platforms. The convergence of AI and RAN is central to NVIDIA’s long-term vision, and the T-Mobile call is a proof point.

Ericsson, by contrast, is taking a more conservative approach. Its AI-managed beamforming runs on its own silicon, avoiding the need for GPUs. This may appeal to operators seeking near-term gains without a hardware refresh. Ericsson’s recent AI-in-RAN subscription launch and its RAN Compute platform at NTT Docomo suggest the vendor is betting on software-driven AI as a practical path.

For Open RAN operators, the vendor landscape is fragmenting. The choice between GPU-based and silicon-based AI-RAN is not just technical; it is strategic. It affects vendor lock-in, cost structure, and the ability to run third-party AI workloads. Open RAN’s promise of multi-vendor interoperability may be tested as vendors pursue different AI-RAN architectures.

Implications for Open RAN Operators

1. AI-RAN is a continuum, not a binary choice

Operators should not view AI-RAN as a single decision between GPU and non-GPU. Instead, they should map their AI-RAN roadmap along a continuum: starting with software-based AI for RAN (e.g., AI beamforming, link adaptation) that delivers immediate efficiency gains, then evolving to AI for RAN workloads that require more compute, and eventually to RAN for AI, where the network itself becomes a platform for AI services. T-Mobile’s dual-track approach validates this staged model.

2. The RIC and SMO are the near-term battleground

The $1.1 billion AI-RAN automation market is largely about software—RIC, SMO, and rApps. Operators should prioritize investments in these areas to enable multi-vendor automation and create a foundation for more advanced AI-RAN use cases. The shift from SON to open, application-based architectures is an opportunity for Open RAN operators to avoid vendor lock-in and foster innovation.

3. GPU-based AI-RAN is a long-term play with uncertain ROI

The Nokia/NVIDIA platform is compelling, but the business case for GPU-accelerated AI-RAN at scale is not yet proven. The lab results from Deutsche Telekom (10-15% uplink throughput gains) are promising, but field trials are needed. Operators should treat GPU-based AI-RAN as a strategic bet, not a near-term revenue driver. They should also consider the total cost of ownership, including power and cooling, which are significant for GPU servers.

4. Vendor diversification is essential

T-Mobile’s dual-track approach with Ericsson and Nokia/NVIDIA demonstrates the value of not relying on a single vendor for AI-RAN. Open RAN operators should similarly cultivate relationships with multiple vendors to hedge against technology and market risks. The emergence of new entrants like Starlink (if it moves into RAN) could further disrupt the vendor landscape.

5. Live trials matter more than lab demos

The T-Mobile call was a live network trial, which is a step forward. But operators should push for commercial-scale trials with clear KPIs. The industry needs more data on performance, reliability, and cost in real-world conditions. Open RAN operators should demand transparency from vendors and share their own learnings to accelerate the ecosystem.

The Road Ahead

T-Mobile’s AI-RAN call is a milestone, but it is not the finish line. The carrier’s dual-track strategy reveals a pragmatic approach that balances near-term gains with long-term innovation. For Open RAN operators, the key takeaways are: embrace the continuum, invest in software and automation, diversify vendors, and demand real-world proof.

The AI-RAN market is still in its early stages. The $1.1 billion automation forecast by 2030 is modest, but it signals a shift toward software-defined, AI-native networks. As operators like T-Mobile, Deutsche Telekom, and others advance their trials, the industry will learn what works and what doesn’t. Open RAN’s principles of interoperability and openness can help operators navigate this transition—provided they remain vigilant about vendor lock-in and demand multi-vendor support.

Ultimately, the first AI-RAN call is a beginning, not an end. The real work of turning AI-RAN from a showcase into a scalable, profitable reality is just starting. Operators that take a staged, pragmatic approach—like T-Mobile—will be best positioned to capture the benefits.