March 2026 - Page 2 of 2 - Dell'Oro Group

AI RAN is moving to the center court. While operators have not fundamentally changed how they think about their RAN roadmaps—openness, intelligence, automation, and virtualization remain the core pillars of next-generation RAN platforms—the visibility and adoption of these technologies vary significantly. In the early phase of 5G, Open RAN and vRAN dominated the conversation. Today, AI RAN is the shiny object.

Events such as MWC2026 Barcelona and Nvidia GTC reinforced the message we have communicated for some time, namely that AI RAN is already happening. At the same time, the GPU conversation is shifting. Looking ahead, AI RAN is expected to see broad adoption across the RAN in the latter half of the 5G cycle and from the outset of 6G.

All roads lead to increased adoption of AI RAN. Differences will emerge across deployment models, compute architectures, hardware choices, functional splits, and underlying technologies.

At present, the majority of the AI RAN market is driven by distributed AI-for-RAN solutions focused on improving performance and efficiency, often leveraging existing 5G infrastructure. Vendors such as Huawei and ZTE have collectively shipped more than 0.6 M AI-enabled boards/plug-ins, underscoring that AI RAN is already happening at scale.

One of the key takeaways from MWC Barcelona is that nearly all RAN roadmaps—across both large and smaller vendors—now incorporate AI RAN capabilities across the full RAN stack, with a focus on AI-for-RAN. And it is not just the baseband—suppliers are now bringing intelligence into every RAN layer, including the radios. Ericsson’s launch of ten AI-ready radios featuring in-house silicon with neural network accelerators is a case in point. The question is no longer if AI RAN and AI-RAN will happen, but rather how, what, where, and when.

Dell’Oro’s long-term view of next-generation RAN has remained broadly intact. Events like MWC 2026 and NVIDIA GTC have done little to alter the underlying trajectory. The likelihood that AI RAN, Cloud RAN, and multi-vendor RAN will play major roles in the second half of 5G and the early 6G era remains high, moderate, and low, respectively. According to our latest forecast update, AI RAN is expected to surpass $10 B and account for roughly one-third of the total RAN market by 2029 (this is not new revenue).

Within the AI RAN domain, the prospects for GPU-RAN (and AI-and-RAN) are improving—still small, but no longer negligible. This shift reflects both low starting expectations and a gradual change in sentiment. The conversation is moving from outright skepticism to cautious curiosity. Much of this momentum is being driven by NVIDIA’s continued push and its vision that the world’s ~10 million macro sites could evolve into more than just base stations. As Jensen Huang put it during his GTC keynote: “That base station…is going to become an AI infrastructure platform.”

Early operator progress—from T-Mobile, SoftBank, and Indosat—combined with Nokia’s recent reiteration of its AI-RAN roadmap, is reinforcing this shift. Samsung and 1Finity, meanwhile, are exploring whether GPUs could make sense to diversify their computing platforms.

Part of the renewed interest in AI RAN—and GPU RAN specifically—stems from a broader realization: technological change is accelerating at a much faster pace than during the 4G-to-5G transition. This shift is reshaping how the industry views the role of mobile networks, the distribution of AI inference, and the trade-offs between hardware-based and software-defined architectures.

At the same time, “physical AI” is becoming more tangible. Concepts that once felt like science fiction—such as robots assisting with cooking or walking children to school—are now increasingly plausible in the near term.

That said, operators remain cautious for now about GPU RAN and broad-base AI inference distribution, even as skepticism gradually eases as the ecosystem matures. The constraints are structural. RAN deployments operate under tight power budgets, strict cost controls, and massive scale requirements. These factors make it challenging to justify deploying power-intensive compute at every cell site.

So, concerns persist about the performance-per-watt gap between GPUs and custom silicon, as well as the practicality and need to support non-telco workloads at both D-RAN and C-RAN sites—particularly in D-RAN deployments. For example, the SoftBank/Ericsson robot assistance demo at MWC operated with latency requirements of around 100 ms, which allows for centralized AI inference, with compute resources located in a data center using the User Plane Function.

In other words, AI RAN is moving from hype toward reality. While trade-offs across AI inference distribution needs, flexibility, performance, energy efficiency, TCO, and TTM will shape adoption paths over the near-term and long-term, the overall direction is clear: AI will become an integral part of every layer of the RAN.

Base-case projections suggest that non-GPU RAN will dominate AI RAN over the forecast period, reflecting both the ability to upgrade existing infrastructure, the constraints at the cell site, and the need for multi-purpose tenancy. This suggests NVIDIA still faces a meaningful challenge if it aims to position itself not only as the “inference king,” but also as the “AI RAN king.”

At the same time, the conversation is clearly evolving. Operators are no longer asking why GPUs might be relevant, but rather where and when they make sense. If NVIDIA succeeds in expanding the role of the RAN—from a single-purpose connectivity layer into a distributed AI platform—the long-term opportunity could be significantly larger than what is currently reflected in our base-case assumptions. As Amara’s Law suggests, the risk may not be overestimating the short-term impact of AI RAN, but underestimating the demand for more distributed intelligence over the long-term.

Open RAN has made significant progress since the O-RAN Alliance was formed in 2018 to “re-shape the RAN industry and ecosystem towards more intelligent, open, virtualized, and interoperable networks.” However, the results to date have been mixed. Open fronthaul (Open FH) is increasingly being specified as a baseline capability for next-generation RAN platforms. At the same time, supplier diversity has not improved. In fact, RAN market concentration is higher today than it was before the alliance was established. Also, uneven adoption across greenfield, early-adopting, and early-majority operators contributed to a sharp capex deceleration following the Open RAN peak in 2022. That slowdown fueled concerns about the movement’s momentum, even with single-vendor Open RAN.

Market conditions improved in 2025. Following the roughly 40 percent decline between 2022 and 2024, preliminary findings suggest worldwide Open RAN revenue grew at a double-digit rate in 2025. Virtualized RAN (vRAN) revenue also stabilized, although at a more modest pace. Several factors help explain this reversal, including easier year-over-year comparisons, more favorable RAN spending trends in regions with strong Open RAN exposure, and, to a lesser extent, increased activity among early-majority adopters.

Vendor rankings did not change significantly, but the broader RAN landscape evolved in ways that also affected the Open RAN and Cloud RAN ecosystems. Both Mavenir and NEC revised their RAN strategies. Mavenir is now focusing more on small cells and non-terrestrial networks (NTN), while NEC is prioritizing vRAN and Massive MIMO. Meanwhile, 1Finity moved up one spot in the Open RAN ranking.

The incumbent Western suppliers are fully hedged. Ericsson and Nokia continue to support Open RAN while maintaining integrated portfolios. According to Ericsson’s latest update, 160 radio models will be Open-RAN-proven by the end of 2026. Likewise, Nokia’s recently introduced AI-RAN-ready Doksuri radios include compatibility with Open fronthaul standards.

Looking ahead, the positive momentum is expected to continue into 2026, with both Open RAN and vRAN projected to grow this year. The longer-term outlook for Open RAN and Cloud RAN also remains favorable. We have not changed the long-term assumptions communicated in the most recent forecast update. To recap, near-term Open RAN revenue projections were revised downward, while long-term growth expectations strengthened.

Virtualization remains a key pillar of next-generation RAN platforms. At the same time, Cloud RAN projections were lowered in the most recent five-year forecast. Still, Cloud RAN is expected to account for roughly 15 to 20 percent of the total RAN market by 2030.

Although the narrative around Open RAN improving supplier diversity has clearly cooled, the emerging GPU-RAN and software RAN wave is reopening the conversation about non-traditional suppliers playing a larger role in the RAN ecosystem. That said, the base case outlook for mixing and matching vendors remains limited. Multi-vendor RAN is still expected to account for less than 5 percent of total RAN deployments by 2030.

For more information about our RAN and Open RAN coverage, please see https://www.delloro.com/advanced-research-report/openran/

Following the 14% revenue decline between 2022 and 2024, telecom equipment investment conditions improved in 2025. Preliminary findings indicate that aggregate worldwide telecom equipment revenues across the six programs tracked by Dell’Oro Group—Broadband Access, Microwave & Optical Transport, Mobile Core Network (MCN), Radio Access Network (RAN), and Service Provider Router & Switch—increased 4% year over year (Y/Y) in 2025, supported by an exceptionally strong fourth quarter (accounting for 29% of full-year revenue).

Improved market conditions were supported by easier year-over-year comparisons, inventory stabilization, favorable currency movements, healthy demand for both wireless and wireline equipment, and robust investment from cloud providers, which contributed meaningfully to the overall growth of the telecom equipment market.

From a regional perspective, double-digit growth in North America and EMEA (Europe plus the Middle East and Africa) more than offset the more challenging conditions in the Asia Pacific. North America and China together accounted for slightly more than half of the overall market in 2025.

While growth was supported by both wireless and wireline segments, Optical Transport and SP Router & Switch stood out, partly reflecting their exposure to data center infrastructure investments.

Relative to our expectations heading into 2025, market performance was slightly stronger than the flat outlook initially outlined, supported by better-than-expected growth in MCN, Optical Transport, and SP Routers. Per the MCN report, the 5G MCN market reached an inflection point in 2025.

Global supplier rankings remained largely unchanged, although revenue shares shifted modestly. Nokia gained share, while Huawei and Ericsson remained broadly stable. Nokia’s share gains were partly driven by its acquisition of Infinera.

Regional dynamics vary significantly. Excluding China, the revenue distribution among the top three suppliers is more balanced. In contrast, excluding North America, Huawei’s overall revenue share reached a new high of 41% in 2025.

We attribute Huawei’s strong performance in markets where it is permitted to compete to three key factors:

First, a comprehensive telco strategy, with Huawei ranking as the #1 supplier by revenue across all six telco programs.
Second, technology leadership, supported by R&D investments that continue to exceed those of its competitors.
Third, footprint expansion, as Huawei has adapted to geopolitical constraints limiting its total TAM by focusing on share gains in markets where it can operate.

Looking ahead, the analyst team expects the positive momentum to extend into 2026. Global telecom equipment revenue across the six programs is projected to grow 2% to 4% in 2026, though the outlook for wireless infrastructure remains more muted.

A few months after Upscale AI introduced SkyHammer—its clean-slate, open-standards scale-up platform designed to make XPUs “behave like a single coherent machine”—the firm is now extending its vision for open AI networking infrastructure into the scale-out domain, where clusters expand horizontally across multiple racks and, increasingly, across multiple data centers. To that end, Upscale AI is announcing a strategic partnership with NVIDIA aimed at accelerating the deployment of open, scale-out AI networking infrastructure for next-generation data centers.

The collaboration brings together NVIDIA’s Spectrum-X Ethernet switch silicon and Upscale AI’s AI-optimized, SONiC-based networking software to deliver interoperable, high-performance Ethernet fabrics designed for large-scale AI workloads.

As enterprises and neocloud providers expand AI clusters, networking has emerged as a critical bottleneck. The partnership focuses on enabling these customers to deploy scalable, low-latency networking systems that support heterogeneous environments spanning compute, accelerators, memory, and storage.

Open Infrastructure for Heterogeneous AI Environments

As part of the initiative, Upscale AI has joined the NVIDIA Partner Network. The partnership is intended to give customers greater flexibility in how they design and procure AI infrastructure, including deploying Ethernet switching powered by NVIDIA Spectrum silicon in heterogeneous, multi-vendor environments. This collaboration reflects a step toward more interoperable Ethernet infrastructure for AI deployments, while maintaining operational consistency at scale.

Focus on AI-Optimized SONiC

A core element of Upscale AI’s approach is its AI-optimized implementation of SONiC, the open-source network operating system widely used in hyperscale environments.

At Dell’Oro Group, we expect SONiC adoption in AI back-end networks to accelerate much faster than what we have historically observed in front-end networks. This faster uptake will be driven by several tailwinds on both the demand as well as supply sides.

On the demand side, a growing number of fast-growing AI model builders and neocloud providers are evaluating SONiC to diversify vendors, reduce platform lock-in, and gain greater control over their network infrastructure. Vendor diversification also helps mitigate risk especially as supply availability tightens.

On the supply side, an expanding ecosystem of established vendors and new entrants is supporting the SONiC ecosystem. We expect SONiC-based switch sales in AI scale-out networks to grow at more than 50 % CAGR (2025-2030), exceeding $10 B by 2030.

Addressing a Critical Gap with Fully Integrated AI Infrastructure for Enterprise and Neocloud Customers

Historically, SONiC adoption has been spearheaded by hyperscalers. However, deploying and operating an open-source network operating system like SONiC demands substantial in-house engineering expertise and integration effort—capabilities many smaller cloud providers and enterprises lack. In addition, SONiC broader ecosystem support—such as turnkey distributions, enterprise-grade tooling, and vendor-backed support—has lagged proprietary network operating systems offerings, limiting SONiC adoption beyond hyperscale environments.

Upscale AI plans to bridge this gap by delivering fully integrated solutions that combine hardware, software, and lifecycle services targeted at organizations building medium and large-scale AI environments.

While the first wave of AI has been driven primarily by large AI model builders—namely hyperscalers—the second wave is expected to be led by other cloud providers, including neocloud providers, as well as large enterprises. Together, these customer segments are projected to account for the majority of the Ethernet data center switch sales in scale-out networks by 2030.

Stitching Together an Open Fabric for AI

SkyHammer was step one. Scale-out is step two. Upscale AI is stitching together an open networking story—from the scale-up interconnect that makes XPUs act like one system, to the Ethernet fabric that lets AI environments grow horizontally while preserving multi-vendor flexibility. The NVIDIA partnership helps validate that direction and accelerates the scale-out side of the roadmap, reinforcing Upscale AI’s broader goal: open, interoperable AI networking infrastructure from pod to cluster.

The Broadband Battleground Is Moving Beyond Speed

In our previous blog titled “The 2026 Broadband Pivot: Why ‘Better’ Beats ‘Bigger,” we argued that the broadband industry’s competitive battleground is shifting from headline speeds to experiential quality — that the operators who win the next decade won’t be the ones who lit up the most fiber wavelengths or pushed the highest downstream speeds, but the ones who invested in intelligence, automation, and the operational architecture to actually deliver on their performance promises.

Obviously, the vendors who supply the equipment and components for their operator customers to make this shift must make adjustments themselves. We are beginning to see a consistent stream of product releases and partnership announcements highlighting this shift in focus away from faster speeds toward reliability and automation.

DOCSIS 4.0 Is Triggering a New Outside-Plant Investment Cycle

One such announcement hit the wires last week. ATX Networks and Harmonic announced an integration between ATX’s GigaXtend GMC Series 1.8GHz Amplifiers and Harmonic’s cOS Virtualized Broadband Platform. The timing is not surprising. As operators prepare their networks for DOCSIS 4.0, outside plant upgrades are entering a multi-year investment cycle. Dell’Oro Group previously projected that amplifier, node, and passive equipment upgrades will drive roughly $10 billion in cable outside plant spending through 2030.

On the surface, the release highlighted the companies’ partnership. But diving deeper, the focus is on the challenges operators are facing when rolling out DOCSIS 4.0 and how the two companies are working to solve those issues.

Specifically, as operators begin to upgrade their headend and outside plant systems with platforms that offer a much higher degree of intelligence, control, and automation, the management architecture required to take advantage of these network components becomes substantially more demanding than what operators have historically required.

Extended Spectrum DOCSIS and Full Duplex both require tighter, more precise control of the RF environment. Noise ingress that was tolerable —or at least manageable— in a DOCSIS 3.1 upstream becomes a hard impairment when you’re pushing FDX traffic into overlapping spectrum. Operators need faster detection, faster diagnosis, and faster resolution—and they need these capabilities at a scale and frequency that makes traditional manual troubleshooting workflows economically unsustainable.

The Limits of Legacy HFC Management Architectures

The inconvenient reality for most MSOs today is that their HFC management architecture was not designed for that operational model. Amplifier telemetry lives in one system. Node performance lives in another. The CCAP platform sits somewhere else. NOC technicians are left triangulating across multiple tools to assemble a picture of what’s happening in the plant —and by the time they have assembled it, a truck is already rolling to the amplifier or node believed to be the source of the ingress noise.

But that might not be the only source of trouble in the outside plant. Often, when is truck is rolled to address the specific ticket that has been generated, any other issues along the cascade can get missed. That is just one of the problems ATX and Harmonic are attempting to solve through their partnership.

Integrating Amplifier Telemetry Into the Virtualized Broadband Platform

ATX’s GigaXtend amplifiers now communicate natively with Harmonic’s cOS platform through embedded transponders. That means amplifier performance data —spectrum capture, ingress analysis, real-time diagnostics—flows directly into the same platform managing the vCMTS, RPDs, and cable modems, rather than feeding into a siloed element management system that operators have to query separately. Technicians can access amplifier settings and troubleshoot impairments through Harmonic’s Sonar cloud tool without context-switching between platforms.

The potential benefits include fewer truck rolls, faster root-cause identification, reduced mean time to repair, and an overall improvement in operational efficiency. In essence, the operator gets a better, more reliable network, but also a network that costs less to run.

Further, there is also an architectural benefit that is gained. When amplifier telemetry becomes a native data stream inside the vCMTS and its management plane, an operator is now one step closer to the network automation that many vendors and operators are talking about. Ingress noise can be detected, correlated to a node segment, and then isolated to a specific amplifier. From there, a resolution workflow can be created and applied without forcing technicians to connect the dots manually. More importantly, technicians can also ensure that a visit to repair the node segment or amplifier includes adjacent amplifiers along the cascade, so that one truck roll can take care of all possible sources of noise.

Though amplifiers with transponders and controller platforms designed to aggregate performance data from an entire system of amps—have been available and deployed for many years, the difference here is the use of the vCMTS container of the Virtualized Broadband Platform and its expanded telemetry capabilities to directly correlate amplifier and node performance with cable modem traffic in a single pane. This insight allows operators to detect, diagnose, and resolve ingress noise issues far faster than before. Also, the integration and correlation of these data streams will allow for more efficient and automated plant operations.

Intelligence and Automation Become the Competitive Advantage

We have argued before that the operators who recognize the shift from raw speed to experiential quality early —and invest in intelligence and automation accordingly— will build competitive advantages that are much harder to replicate than simply deploying more infrastructure. This partnership, as well as those expected to follow, is a concrete example of how operators and their vendor partners are working to improve the perceived quality and reliability of their broadband networks and services.

Contact Us