Nokia’s dApp Gamble: E3 Interface, Nvidia GPUs, and the Fight for Open RAN’s Next App Economy

Nokia is making a calculated, multi-front bet that the future of radio access networks runs on merchant silicon, disaggregated software, and a new class of real-time applications. According to a report covered by ETTelecom on September 30, 2026, the Finnish vendor is positioning Nvidia GPUs as the long-term general-purpose substitute for custom ASICs in the RAN, while simultaneously championing a new “dApp” application model built on the O-RAN E3 interface. It is a strategy that touches every layer of the stack—silicon, RIC architecture, and the application ecosystem—and it deserves a close engineering read.

The Silicon Bet: From Marvell Co-Design to Nvidia GPUs

Nokia’s current custom silicon story is well established. The company co-developed its latest baseband ASIC with Marvell Technology, a partnership that continues to underpin its shipping 5G portfolio. But the ETTelecom report makes clear that Nokia’s long-term vision is different: it anticipates Nvidia GPUs becoming the primary alternative to custom silicon in the RAN.

This is not a trivial architectural claim. Custom ASICs deliver deterministic, power-efficient L1 processing at massive scale. General-purpose GPUs, by contrast, offer programmability, a mature software ecosystem (CUDA), and the ability to run AI inference workloads concurrently with radio processing. The trade-off has always been power and cost per bit. Nokia’s bet is that those curves cross—or at least converge enough—that the flexibility wins.

The strategic logic is visible in Nokia’s AI-RAN platform, built on its anyRAN software and Nvidia’s Aerial AI-RAN platform. Nokia has publicly targeted more than 100 percent spectral efficiency gains by 2028 on much of the existing infrastructure, a figure repeated in EME Outlook Magazine’s September 28 profile of the company. That is an aggressive number, and it is worth being precise about what it means: the claim is not that GPUs magically create spectrum, but that AI-driven scheduling, beamforming, and inference can extract more usable capacity from the same radio assets without a traditional hardware refresh cycle.

CEO Justin Hotard has claimed that 19 operators have signed up for “initial deployments” of Nokia’s forthcoming AI-RAN offers. That is a headline figure, but the LightReading analysis from September 28 adds important texture: the three companies Hotard identified in his LinkedIn post are less well known, and none is a telco. That gap between the headline number and the named participants is exactly the kind of thing engineers should track. “Initial deployments” can mean anything from a lab trial to a limited commercial cluster. Until operators are named and the deployment topology is described, the 19-operator claim should be treated as directional rather than proven.

What is more concrete is the operator validation Nokia has accumulated around GPU-accelerated AI-RAN. The company says it has demonstrated the technology with BT, Elisa, Vodafone Group, and NTT DOCOMO, validating that AI inference workloads and radio processing can run concurrently on shared infrastructure. That concurrency point is the technical crux. If AI inference and L1 processing can genuinely share the same GPU without violating latency budgets, the economic case for merchant silicon in the RAN strengthens considerably.

The dApp Model: Why E3 Changes the Application Game

The more interesting architectural story is Nokia’s embrace of dApps. The ETTelecom report quotes a Nokia executive explaining that the company has “taken the whole concept of APIs and decoupling layers in a very serious fashion,” enabling an O-RAN-compliant E3 interface that supports sensing, geo-precise location monitoring, and improved spectral efficiency.

To understand why this matters, it helps to recall the trajectory of RAN applications. The original xApp model was built around the near-real-time RIC and the E2 interface. xApps operate on timescales of roughly 10 milliseconds to one second—useful for policy, mobility optimization, and some interference management, but too slow for the most latency-sensitive control loops. The industry’s enthusiasm for xApps has been muted; deployments exist, but the ecosystem has not exploded.

dApps are designed for tasks demanding even faster execution, running on a real-time RIC and leveraging the newer E3 interface. The E3 interface is the key enabler here. Where E2 connects the RIC to the RAN nodes, E3 is intended to expose lower-level, more granular control and data paths—enabling applications to act on timescales that approach the radio itself. That opens the door to use cases that were previously the exclusive domain of vendor-internal firmware: advanced sensing, precise positioning, and dynamic spectral efficiency optimization.

Nokia’s support for E3 is a meaningful signal. It means the vendor is willing to expose interfaces that historically sat inside its proprietary domain. For operators, the promise is a more competitive application market. For Nokia, the calculation is that a richer app ecosystem makes its RAN more valuable, even if it cedes some control.

The Startup Ecosystem: Cohere, DeepSig, and ZaiNar

The dApp model only works if there are applications to run. The ETTelecom report notes that startups including Cohere Technologies (referred to as Pulsone in the report), DeepSig, and ZaiNar are already designing RAN software for Nokia-built networks.

Cohere is the most commercially advanced of these. Its Universal Spectrum Multiplier (USM) technology has been publicly endorsed by Vodafone, Australia’s Telstra, and Bell Canada after field trials—and all three have now made investments in the startup. Cohere CEO Ray Dolan, speaking on a Telecoms.com podcast, identified Nokia as the vendor partner during trials in the Bell Canada network. Those trials used a common, off-the-shelf server deployed next to a basestation at a mast site to host Cohere’s technology. But as Dolan noted, a dApp would be closer to a direct integration and would likely come with service benefits. It would achieve what xApps did not.

That last point is worth unpacking. The xApp ecosystem struggled partly because the near-real-time RIC added latency and complexity without delivering compelling enough gains for many operators. If dApps can deliver Cohere’s spectral efficiency improvements as a native, real-time integrated function rather than a bolt-on server, the value proposition shifts from “interesting experiment” to “measurable capacity gain.”

DeepSig brings AI-powered spectrum sensing to the table, and ZaiNar contributes precise location and timing capabilities. Together, these three startups sketch out a dApp ecosystem focused on the physical layer and spectrum domain—exactly where real-time execution matters most.

The Counterpoint: Ericsson Demurs, and the Market Is Flat

No engineering strategy exists in a vacuum, and two data points from the same news cycle provide essential context.

First, Ericsson is not matching Nokia’s GPU enthusiasm. The LightReading headline from September 28 is explicit: “Nokia puts new open RAN tech in 5G revival bid as Ericsson demurs.” Ericsson’s reticence is not irrational. The Swedish vendor has a massive installed base of custom silicon and a conservative operator customer base that values proven performance and predictable power consumption. Betting early on GPUs in the RAN is a high-risk, high-reward play. Nokia, with less to lose in RAN market share and more to gain from differentiation, is making the aggressive move.

Second, the RAN market itself is not growing. Dell’Oro Group’s Stefan Pongratz, speaking at the Intelligent RAN Forum, said the post-5G inventory correction is over, but the market will stay essentially flat through 2030. Growth is concentrated in specific segments—AI RAN, massive MIMO, and cloud RAN—rather than spread across the board. Pongratz also noted a structural constraint: RAN is about a fifth of total wireless capex, and the two have held a 98% correlation. “If you want to change the capex trajectory, you have to change the revenue trajectory,” he said.

This is the economic backdrop against which Nokia’s dApp and GPU strategy must be judged. In a flat market, vendors cannot rely on volume growth. They must either take share or expand the value pool. Nokia’s bet is that AI-RAN and dApps expand the value pool by enabling new services and efficiency gains that justify incremental spend—even if the total RAN envelope stays flat.

The Latin American Proof Point: Quantifying AI-RAN Gains

While Nokia’s strategy is the headline, a separate proof-of-concept from GlobalLogic, reported by Intelligent CIO LATAM on September 30, offers some of the most concrete performance numbers in this news cycle.

GlobalLogic, owned by the Hitachi Group, tested AI and GPU acceleration in a 5G Open RAN environment. The results: processing times improved by as much as 15 times under certain conditions, and power consumption fell by around 40%. A second experiment using deep-learning models to distinguish genuine network-access attempts from noise or interference reported a 99.98% reduction in false detections.

These are lab results in a controlled environment, not commercial network data. But they are directionally important. A 40% power reduction in processing is a meaningful operational expense improvement for operators. A 15x processing time improvement suggests that AI-accelerated workloads can handle tasks that would be impractical on traditional silicon. And the 99.98% false-detection reduction speaks directly to the reliability concerns that have slowed AI adoption in RAN operations.

The GlobalLogic work also reinforces a point that Open RAN advocates have made for years: disaggregation creates the environment for alternative computing architectures to play a larger role. By separating network functions and encouraging interoperability, Open RAN makes it possible to insert GPUs, FPGAs, or other accelerators where they make engineering sense, rather than being locked into a single vendor’s silicon roadmap.

What Engineers Should Watch

The next twelve months will be decisive for Nokia’s strategy. Several technical questions remain open:

Power and thermal density at the cell site. Cohere’s trials used an off-the-shelf server next to a basestation. If Nokia’s AI-RAN platform requires similar edge compute at scale, operators will need to solve power, cooling, and real estate constraints at thousands of sites. The 40% power reduction GlobalLogic measured is encouraging, but it is a processing-level metric, not a site-level one.

E3 interface maturity. Nokia says it supports an O-RAN-compliant E3 interface. The O-RAN Alliance’s E3 specification is still evolving, and multi-vendor interoperability is unproven at scale. Operators should ask hard questions about which E3 functions are actually exposed and how they are secured.

The 19-operator claim. Watch for named operators and deployment details. The gap between “signed up for initial deployments” and “running commercial traffic” is where most AI-RAN promises have historically stalled.

Ericsson’s counter-move. If Nokia’s GPU bet shows measurable gains in live networks, Ericsson will face pressure to respond. But Ericsson’s installed base and conservative customer relationships give it room to wait and see. The competitive dynamic here is a classic innovator’s dilemma, and it will play out in operator procurement decisions over the next two years.

The application ecosystem. dApps are only as valuable as the applications available. Cohere, DeepSig, and ZaiNar are promising, but three startups do not make an ecosystem. Nokia needs broader developer adoption, and that depends on how open and well-documented the E3 interface really is.

The Bottom Line

Nokia is making a coherent, if risky, architectural bet: merchant silicon (Nvidia GPUs) plus open interfaces (E3) plus a new application model (dApps) equals a differentiated RAN platform that can grow value in a flat market. The strategy is technically credible, and the early operator validations—BT, Elisa, Vodafone, NTT DOCOMO—are real. The Cohere investment from Vodafone, Telstra, and Bell Canada is a strong signal that operators see value in the spectral efficiency story.

But the market is flat, Ericsson is skeptical, and the gap between lab results and commercial deployments remains wide. The 15x processing gains and 40% power reductions are impressive in a controlled environment; the question is whether they survive contact with real-world traffic, multi-vendor integration, and the operational realities of thousands of cell sites.

For Open RAN operators, the Nokia strategy is worth watching closely—not because it is certain to succeed, but because it is the most concrete attempt yet to answer the question that has haunted the movement since its inception: what is the application economy that justifies disaggregation? If dApps and E3 deliver, that question may finally have an answer.

Sources

• Nokia puts new open RAN tech in 5G revival bid as Ericsson demurs
• LATAM mobile networks could become smarter with AI-RAN and GPU acceleration – Intelligent CIO LATAM
• Nokia bets on dApps, Nvidia GPUs for 5G open RAN competitiveness: Report, ETTelecom
• RAN market flat through 2030 as Dell’Oro sees the mix shift
• Nokia : Intelligent Networks by Design - EME Outlook Magazine