Red Hat Charts the Three-Stage Path from AI for RAN to Truly AI-Native Networks

In a quiet conference room during RCR Wireless News’ Telco AI Forum this July, Shujaur Mufti, Red Hat’s director of telco ecosystem solution architecture, leaned into the microphone and delivered a message that cut through the hype: operators aren’t waiting for a wholesale radio overhaul to start harvesting AI gains. “I think AI for RAN is starting first because you can see, visualize the savings, for example, opex,” he said. The room nodded—here was someone speaking the language of CFOs and CTOs alike.

Mufti’s three-stage roadmap for AI-RAN, grounded in Red Hat’s work with SoftBank, Fujitsu, and NVIDIA, offers telecom leaders a clear, incremental playbook rather than another moonshot vision. It begins with “AI for RAN”—targeted intelligence layered on today’s networks—progresses to “AI and RAN,” where compute is shared, and culminates in “AI on RAN,” where the radio access network itself becomes an AI platform. For Open RAN operators already embracing disaggregation and open interfaces, the implications are immediate and profound.

From SON to AI-Enhanced Intelligence Without Rip-and-Replace

The first phase, AI for RAN, is already delivering measurable returns on legacy infrastructure. Mufti highlighted classic use cases: energy optimization that trims opex, spectral efficiency boosts, faster fault detection, and evolved self-organizing networks (SON) now infused with machine learning. These capabilities sit comfortably inside existing SMO frameworks, including the xApps and rApps that Open RAN operators are deploying today.

Critically, no massive hardware refresh is required. “You don’t have to modernize your RAN network, and then you can get the benefits right there,” Mufti emphasized. For operators running multi-vendor Open RAN deployments, this means AI-driven rApps can optimize both open and traditional radios from the same orchestration layer. Early adopters are already seeing the payoff in lower power bills and improved downlink throughput without touching the radios themselves.

This pragmatic starting point explains why AI for RAN will dominate industry focus through at least 2027. It lowers the barrier to entry and builds internal confidence in AI tooling before bigger architectural bets are placed.

The Middle Phase: Shared Infrastructure and Proofs of Concept

Between 2027 and 2030, attention shifts to “AI and RAN”—the era of converged infrastructure. Here, AI inferencing and radio workloads begin sharing GPU-accelerated platforms at the edge. Red Hat sees proof-of-concept activity accelerating as 6G standards take shape.

Mufti cautioned against assuming every cell site will run on GPUs tomorrow. “We should not think GPU everywhere in the RAN,” he said. “Maybe some selected sites as a starting point.” Targeted edge deployments—stadiums, enterprise campuses, dense urban clusters—will serve as the proving ground. Operators like SoftBank and T-Mobile are already exploring these models, running Layer 1 and Layer 2 functions on GPU platforms without real-time kernel dependencies.

For Open RAN ecosystems, this convergence plays to the architecture’s strengths. Open interfaces and cloud-native design make it easier to introduce AI workloads alongside DU/CU functions. Red Hat’s AI Grid initiative is positioned as the connective fabric, delivering a RAN-ready AI infrastructure platform that spans data center, core, edge, and radio domains.

The Long Game: AI-Native Services and New Revenue

Beyond 2030, coinciding with commercial 6G rollouts, the industry enters “AI on RAN.” The radio network becomes a platform for AI-native applications—new services that monetize excess compute and deliver enterprise-grade intelligence at the edge. Mufti stressed that deployments will scale only where clear economic value is proven: better network quality plus fresh revenue streams.

One path forward is to start with pure AI inferencing at the edge, measure the business case, and then decide how much GPU capacity to allocate to radio functions. This measured approach resonates with operators wary of stranded assets.

Why Open RAN Operators Should Pay Attention Now

Open RAN’s disaggregated, standards-based foundation is uniquely suited to this phased journey. The same open APIs that enable multi-vendor interoperability also accelerate the insertion of AI xApps/rApps today and shared AI-RAN workloads tomorrow. Red Hat’s collaborations demonstrate that open-source tooling can underpin both the operational intelligence layer and the underlying cloud-native platform.

Operators who treat AI-RAN as part of a broader AI-native transformation—applying lessons from core, OSS/BSS, and autonomous networks—will build the consistent fabric needed for the next decade. Those who wait for perfect 6G specs risk falling behind on the operational and financial gains already available.

Mufti’s message is ultimately one of disciplined optimism: AI-RAN only makes sense when it delivers technology and economic benefits. The roadmap he outlined gives the industry a realistic timeline and a clear set of milestones to hit along the way.

Sources

• https://www.rcrwireless.com/20260717/carriers/red-hat-outlines-ai-ran
• https://ai-ran.org/
• https://www.o-ran.org/