AI-RAN Grows Up: How BT’s 2027 Trial Fits a Decade of RAN Evolution

When BT Group announced that it will begin live AI-RAN trials with Nvidia and Nokia on its EE mobile network in Birmingham in 2027, it did more than add another logo to the AI-RAN bandwagon. It marked the arrival of AI-native radio access at a mainstream European incumbent—a company that has spent the better part of a decade navigating the promises and pitfalls of network disaggregation. For an industry that has watched AI-RAN evolve from a research curiosity to a board-level imperative, BT’s move is a milestone worth understanding in historical context.

The story of how we got here is not a straight line. It is a tale of two parallel revolutions—Open RAN and AI—that have slowly, sometimes awkwardly, converged. BT’s trial, slated for 2027, is the latest chapter in that convergence, and it offers a window into what the next phase of network evolution will look like.

From Disaggregation to Intelligence: The Open RAN Prelude

To appreciate why BT’s AI-RAN trial matters, it helps to recall the industry’s original ambition. Open RAN was born out of frustration with the traditional, proprietary RAN model dominated by a handful of vendors. The promise was disaggregation: split the base station into open, interoperable components—radio units, distributed units, centralized units—and let operators mix and match vendors. The vision was compelling: lower costs, more innovation, and reduced vendor lock-in.

But the road was rocky. Early Open RAN deployments were plagued by integration challenges, performance gaps, and a steep learning curve. Operators invested heavily in trials and testbeds, but commercial rollouts were slow. By the mid-2020s, the narrative had shifted from “Open RAN will replace traditional RAN” to “Open RAN is a tool in the toolbox.” The technology matured, standards solidified, and interoperability improved, but the economics remained challenging. RAN spending, as Light Reading reported, fell from $45 billion in 2022 to$35 billion in 2024—a decline that forced vendors like Ericsson and Nokia to cut tens of thousands of jobs. The pie was shrinking, and Open RAN’s slice was still small.

Yet the disaggregation genie was out of the bottle. The architectural principles of Open RAN—virtualization, cloud-native design, open interfaces—became the foundation for something bigger. As operators virtualized their networks, they began to ask: if the RAN is now software running on commodity hardware, what else can we do with it? The answer, increasingly, was artificial intelligence.

The AI-RAN Spark: From Lab to Live

The idea of using AI to optimize radio networks is not new. For years, vendors have embedded machine learning algorithms into their RAN products to improve scheduling, beamforming, and energy efficiency. But the AI-RAN concept that gained traction in the mid-2020s was more ambitious: use AI not just to optimize existing functions, but to fundamentally transform how the RAN operates—and to create new services at the edge.

The turning point came when Nvidia, the GPU giant, turned its attention to telecom. Nvidia’s Aerial platform, combined with its Grace Hopper superchips, offered a way to run AI workloads directly in the RAN, enabling real-time inference and even training at the edge. Nokia, seeking to differentiate its RAN portfolio, partnered with Nvidia to build AI-RAN solutions. Together, they demonstrated impressive results: in laboratory testing, Nokia reported improvements exceeding 20% in spectral efficiency. T-Mobile, in the US, completed an AI-RAN call on its commercial network using Nvidia and Nokia technology—a proof point that AI-RAN was not just a lab experiment.

Other operators followed. In November 2025, IOH, Nokia, and Nvidia opened an AI-RAN research centre in Indonesia. In 2026, Orange, Nvidia, and Nokia announced a collaboration. SK Telecom launched a national AI-RAN test network. The AI-RAN Alliance, an industry group founded to promote the technology, grew to 132 members and 33 demos. The message was clear: AI-RAN was moving from concept to commercial reality.

But the journey was not without skepticism. Critics pointed out that AI-RAN’s benefits were often demonstrated in idealized conditions, and that the economics of deploying GPUs in every cell site were questionable. The RAN market was flat, and operators were cautious about new capital expenditure. The hype cycle, as it always does, collided with the reality of budgets and timelines.

BT’s Bet: A Pragmatic Path to AI-RAN

Against this backdrop, BT’s announcement stands out for its pragmatism. The company is not promising a nationwide AI-RAN rollout overnight. Instead, it plans live trials on its EE network in Birmingham in 2027, using Nvidia and Nokia technology. The trials will assess whether the spectral efficiency gains seen in the lab—over 20%—can be achieved on a live network. That is a critical question. If AI-RAN can deliver meaningful capacity improvements in real-world conditions, it could help BT manage growing data traffic without proportional increases in infrastructure investment.

BT’s interest in AI-RAN is also part of a broader technology strategy. The company is investing heavily in Openreach, its fiber infrastructure arm, and is navigating the regulatory complexities of its proposed TalkTalk acquisition. AI-RAN offers a way to improve network efficiency and potentially reduce the need for additional spectrum or cell sites—a compelling proposition for a mature market like the UK.

But BT is careful to temper expectations. The company notes that AI-RAN remains an emerging technology and is unlikely to make a material contribution to near-term earnings. That candor is refreshing in an industry often prone to overpromising. It also reflects the lessons of the Open RAN era: transformative technologies take time, and the path from trial to commercial deployment is rarely smooth.

The Vendor Chessboard: Nokia, Nvidia, and the Rest

BT’s choice of Nvidia and Nokia as partners is no accident. Nokia has positioned itself as a leader in AI-RAN, with a commercial AI-RAN platform and a growing list of operator trials. The company’s partnership with Nvidia gives it access to cutting-edge GPU technology, while its own RAN expertise ensures that the solution is integrated and optimized. Nokia’s mobile networks business has shown improvement: in Q3 2026, its Mobile Infrastructure segment posted €2.902 billion in sales, up 5%, with an operating profit of €420 million and a 14.5% margin. That is a clear sign that Nokia’s bet on AI-RAN and cloud-native RAN is starting to pay off.

Nvidia, meanwhile, is playing the long game. Its Aerial platform and GPU hardware are becoming the de facto standard for AI-RAN, much as its GPUs became the standard for AI training in the data center. By partnering with Nokia and operators like BT, Nvidia is embedding itself in the telecom infrastructure—a market that, while smaller than the data center, offers significant growth potential as networks become more intelligent.

Other vendors are not standing still. Ericsson, Nokia’s main rival, has taken a different approach, emphasizing custom silicon and AI features embedded in its radios. On its Q3 2026 call, Ericsson said its AI-native link adaptation delivers a 10% spectrum efficiency gain and runs commercially on more than 60,000 base stations. That is a concrete deployment number, and it supports the argument that Ericsson can monetize AI inside the network rather than only around it. Ericsson also claims 206 live 5G networks across 85 countries, and it has been pushing its own vision of AI-native RAN.

The competition between Ericsson and Nokia—and between custom silicon and GPUs—will shape the next phase of AI-RAN. BT’s trial will provide valuable data on the GPU-based approach, but it is unlikely to settle the debate. The industry is still in the early innings, and different operators will make different choices based on their network architectures, spectrum holdings, and business models.

The Economics of AI-RAN: Promise and Peril

The enthusiasm for AI-RAN is tempered by a hard truth: the RAN market is not growing. As Light Reading reported, RAN spending fell from $45 billion in 2022 to$35 billion in 2024, and it has remained flat. Vendors have cut tens of thousands of jobs—Ericsson cut more than 15,600 full-time jobs between early 2023 and September 2025, and planned 1,600 more in Sweden; Nokia announced 5,000 additional cuts in November 2025. Combined headcount at the two companies is down more than 36,000 since 2017, while Huawei added about 28,000 employees from 2017 to 2024.

In this context, AI-RAN must justify its cost. GPUs are expensive, and deploying them in every cell site may not make economic sense for all operators. The spectral efficiency gains—20% in the lab, perhaps 10-15% in the field—are meaningful but not transformative on their own. The real value of AI-RAN may come from new services: edge AI inference, network slicing, and autonomous network operations. But those services are still nascent, and the business models are unproven.

BT’s cautious approach reflects this reality. The company is not betting the farm on AI-RAN; it is running a live trial to gather data and inform its long-term strategy. That is a sensible approach, and it is likely to be replicated by other operators. The AI-RAN journey will be incremental, not revolutionary.

The Road Ahead: 2027 and Beyond

As BT prepares for its 2027 trial, the broader AI-RAN ecosystem continues to evolve. The AI-RAN Alliance is working on standards and interoperability, and the O-RAN Alliance is incorporating AI into its specifications. The 3GPP is studying AI-native air interfaces for 6G. The pieces are coming together, but the puzzle is far from complete.

BT’s trial will be watched closely by the industry. If it demonstrates real-world gains, it could accelerate AI-RAN adoption among European operators. If it falls short, it could reinforce skepticism and delay investment. Either way, the trial is a necessary step in the maturation of AI-RAN—a technology that promises to make networks more efficient, more autonomous, and more capable of supporting the applications of the future.

For BT, the trial is also a test of its own ability to innovate. The company has a long history of network leadership, from the early days of telegraph to the rollout of fiber and 5G. AI-RAN represents the next frontier, and BT’s willingness to experiment—cautiously, but seriously—is a sign that it intends to remain at the forefront.

The history of telecom is punctuated by moments when disparate technologies converge to create something new. The convergence of Open RAN and AI is one such moment. BT’s 2027 trial is a milestone on that journey, and it will help determine whether AI-RAN becomes a foundational technology or a footnote. The industry will be watching.

Sources

• BT Group (LON: BT.A) shares price forecast for 2026-2028: Analysts see 224p target
• Ericsson vs Nokia: AI Sales +123% vs 7% Growth [2026]
• ETSI deploys new standard to combat deepfakes
• Telecoms.com | news, & analysis for the telecommunications
• Batelco and Ericsson expand 5G network coverage and introduce AI-powered capabilities in the Kingdom of Bahrain | ZAWYA
• Comprehensive Analysis of the Open RAN Security Solutions Market Size: Projected Growth with a 9.3% CAGR from 2026 to 2033
• Top 15 Companies in Global Network Digital Twin Market: Statistics Report Till 2035