T-Mobile, Ericsson Demo Cloud RAN on NVIDIA GPUs

TL;DR

• Ericsson ran its Cloud RAN software on NVIDIA AI infrastructure using Aerial CUDA during an over-the-air trial at T-Mobile’s AI-RAN Innovation Center in Bellevue, proving hardware portability across multiple platforms.
• T-Mobile’s CNO says the demo shows the carrier evolving “from a connectivity pipe to an intelligent platform” — setting the stage for AI-native services and 6G.
• The showcase hits Mobile World Congress 2026 in Barcelona (March 2-5), positioning Ericsson against Huawei’s vertically integrated AI-Native framework and validating NVIDIA’s RAN acceleration ambitions.
• Hardware-agnostic Cloud RAN gives operators flexibility to switch suppliers and dodge vendor lock-in — a critical selling point as carriers weigh massive AI-RAN infrastructure investments.

Ericsson’s Cloud RAN Software Runs on NVIDIA Acceleration

Ericsson successfully demonstrated its Cloud RAN software stack running on NVIDIA AI infrastructure during an over-the-air trial at T-Mobile’s AI-RAN Innovation Center in Bellevue. The demo used NVIDIA’s Aerial CUDA platform — a specialized accelerated computing framework designed for radio access network workloads. Ericsson plans to showcase the solution at Mobile World Congress 2026 in Barcelona, scheduled for March 2-5.

The demonstration proves that Ericsson’s Cloud RAN software maintains portability across multiple hardware platforms while supporting NVIDIA’s accelerated computing infrastructure. Mårten Lerner, Head of Networks Strategy at Ericsson, framed the achievement as validation of architectural flexibility: “Cloud RAN software is portable by design. By running the same RAN software stack across multiple hardware platforms, we reinforce our commitment to providing mobile operators with true flexibility without compromising on high performance.”

T-Mobile’s Chief Network Officer Ankur Kapoor positioned the trial as a strategic shift in network architecture. He said, “At T-Mobile, every network decision starts with customer centricity, ensuring our investments deliver the best performance, reliability, and value for those who trust us with their connectivity. Our over-the-air trial in the AI-RAN Innovation Center in Bellevue demonstrates how we are evolving from a connectivity pipe to an intelligent platform, laying the groundwork for AI-native services and future innovation in advanced AI and 6G.”

Why Ericsson’s Hardware Portability Gambit Threatens Huawei’s Playbook

This demonstration isn’t just a technical milestone — it’s a strategic wedge against Huawei‘s vertically integrated AI-Native framework. Huawei’s approach bundles RAN software and custom silicon into a single optimized stack, promising maximum performance at the cost of reduced flexibility. Ericsson’s bet is that operators will choose portability over integration, especially as AI-RAN infrastructure investments balloon into the billions.

And I think that bet makes sense in a market where vendor lock-in has historically burned carriers. Operators remember what happened when they couldn’t easily swap hardware suppliers during previous network transitions. This time, they’re demanding architectural escape hatches.

The demonstration also validates NVIDIA’s ambition to become the accelerated computing standard for AI RAN deployment — a market where it competes with Qualcomm’s RAN accelerators and custom silicon from traditional telecom equipment manufacturers. By publicly endorsing NVIDIA’s Aerial CUDA platform, Ericsson signals that the market favors specialization: AI acceleration chips separate from RAN processing. That’s bad news for vendors who built their competitive moats around end-to-end integration.

Think of it like the shift from integrated graphics to discrete GPUs in gaming PCs. Sure, an all-in-one solution from a single vendor can be optimized — but when performance demands spike and workloads diversify, specialized components from best-of-breed suppliers win. Cloud RAN with NVIDIA acceleration follows the same logic.

But here’s the tension nobody’s saying out loud: hardware portability might dilute NVIDIA’s pricing power. If Ericsson’s software truly runs across multiple accelerated computing platforms without modification, operators can pit NVIDIA against Qualcomm and Intel in procurement negotiations. NVIDIA’s current dominance in AI compute doesn’t automatically translate to RAN market lock-in if the software layer remains genuinely portable.

The demonstration also positions Ericsson as a favorable partner for operators concerned about vendor lock-in — a direct shot at Huawei’s integrated approach. Nokia’s AI-driven RAN solutions face the same strategic question: integrate tightly and risk lock-in concerns, or embrace portability and compete on software performance alone? Ericsson just planted its flag firmly in the portability camp.

What does this mean for operators evaluating AI-RAN infrastructure investments? They now have a credible alternative to Huawei’s “our way or the highway” architecture. T-Mobile’s willingness to conduct an over-the-air trial — not just a lab demo — suggests the technology works in real network conditions. That’s the kind of proof point that moves procurement committees.

The Broader Shift Toward Disaggregated RAN Architectures

Ericsson has been investing in Cloud RAN technology as a core strategy to help operators reduce capital expenditure and operational complexity. The partnership with T-Mobile and NVIDIA reflects broader industry momentum toward disaggregated RAN architectures where software and acceleration hardware come from different vendors. This approach directly challenges Huawei’s traditional integrated model and represents a fundamental architectural divergence shaping the 6G transition.

Disaggregation isn’t new — it’s been the dominant trend in data center infrastructure for over a decade. But applying that philosophy to radio access networks introduces unique challenges. RAN workloads demand ultra-low latency and real-time processing that data center applications don’t require. The fact that Ericsson’s software runs on NVIDIA’s general-purpose AI acceleration platform suggests those latency requirements can be met without custom silicon.

The timing matters. As the industry begins sketching the outlines of 6G standards, architectural decisions made now will echo for a decade or more. If disaggregated Cloud RAN becomes the dominant model, vendors like Ericsson and Nokia gain leverage. If vertically integrated AI-Native frameworks win, Huawei’s approach looks prescient.

T-Mobile’s emphasis on “AI-native services” and “future innovation in advanced AI and 6G” hints at where this is heading. Operators don’t just want faster pipes — they want programmable infrastructure that can support new revenue streams. Cloud RAN’s software-defined architecture enables that flexibility in ways traditional integrated RAN can’t match.

What to Monitor as AI-RAN Architecture Wars Heat Up

Watch whether other Tier 1 operators follow T-Mobile’s lead and conduct over-the-air trials of disaggregated Cloud RAN on NVIDIA infrastructure. Lab demos are one thing; production deployments at scale are another. If Verizon, AT&T, or European carriers like Deutsche Telekom or Vodafone announce similar trials in the next six months, Ericsson’s architecture wins momentum. If adoption stalls, Huawei’s integrated approach retains its appeal.

Pay attention to NVIDIA’s pricing strategy for Aerial CUDA. Right now, NVIDIA dominates AI compute and can charge premium prices. But if Ericsson’s portability claim holds and operators can genuinely switch between NVIDIA, Qualcomm, and Intel acceleration platforms, NVIDIA faces pricing pressure it hasn’t encountered in the AI training market. How NVIDIA responds — whether it competes on price or doubles down on performance differentiation — will shape the economics of AI-RAN deployments.

Finally, monitor the reaction from Nokia and Samsung. Both compete with Ericsson in the RAN market, and both face the same strategic choice: embrace portability or defend integration. If they follow Ericsson’s lead and demonstrate hardware-agnostic Cloud RAN, the industry consensus shifts decisively. If they stick with tighter hardware coupling, the market fragments — and operators face a confusing array of incompatible architectures as they plan 6G transitions.