Hello,

This week, three global tier-1 operators formalised AI-native RAN partnerships in 17 days this March. Europe, North America, Asia. Different vendors. Same direction.

In today's edition:

Let’s dive in.

On 1 March 2026, Nokia and Deutsche Telekom announced an expanded strategic collaboration to advance AI-native and Open RAN. This was not a product purchase. It was a joint co-creation programme covering AI-powered receivers, adaptive beamforming, predictive network optimisation, and autonomous fault resolution, validated jointly across Deutsche Telekom Labs and live commercial field environments.

The scope matters. Previous operator-vendor agreements involved operators buying platforms that vendors had already developed. This agreement has Deutsche Telekom engineers co-designing AI capabilities with Nokia's Bell Labs teams from specification stage, with commercial timelines tied to Deutsche Telekom's own 5G-Advanced and 6G planning cycles, not Nokia's product release schedule.

Nokia's CTO described the shift on 1 March: AI-native RAN represents "a fundamental shift in how networks will be built and operated." Operationally that means AI stops being a layer added on top of the RAN and becomes embedded in the radio architecture changing how beamforming decisions are made, how interference is managed, and how energy is allocated across the spectrum.

This announcement sits inside a 17-day window. On 15 March, T-Mobile and NVIDIA went live with physical AI applications running on AI-RAN-ready infrastructure the first US operator to run AI workloads and radio functions on shared hardware in a production environment. On 18 March, Ericsson and SK Telecom signed a multi-year MoU through 2031 covering autonomous channel learning, energy efficiency, and AI-driven resource optimisation across multi-vendor 5G.

Deutsche Telekom chose Nokia. SK Telecom chose Ericsson. T-Mobile chose Nokia radio software and NVIDIA compute on shared hardware. Three tier-1 operators. Three different vendor configurations. None of this was accidental.

Production benchmarks from April add weight. Agentic AI deployments in live European networks are reporting up to 25% opex reduction and 30% downtime reduction. Vodafone's 5G sites in London recorded up to 33% power reduction from AI-based optimisation measured in production, not pilots.

Strategic analysis: The vendor split is deliberate and reflects a shift in operator procurement logic. No operator committed exclusively to a single vendor. Each secured a primary co-development partner while preserving competitive alternatives the standard playbook before 6G standards lock in 2027–2028. Co-creation agreements give operators influence over 6G radio architecture before vendors bundle and price it. Deutsche Telekom and SK Telecom are writing requirements into joint development programmes. T-Mobile is doing it through production deployment data. The pressure on vendors to match these configurations is structural, not commercial.

What this means for you: If your vendor selection process still treats AI-native RAN as a feature tier in an RFP, you are behind. Operators who will have leverage in 6G vendor negotiations are those co-creating now, not purchasing later. Co-development agreements lock in architecture preferences before standards finalise.

The March cluster is best understood not as three separate announcements but as one coordinated market signal. Each operator locked in a primary AI-RAN development partner. Each chose a different vendor configuration. Each published their announcement publicly, ensuring competitors and vendors could see the full picture simultaneously.

In 2024 and 2025, vendors competed to convince operators their proprietary AI-RAN implementations were superior. Operators ran parallel pilots. Vendors charged premium pricing because differentiation was real , each platform had genuinely different capabilities. By March 2026, that differentiation window was closing. Standards from the O-RAN Alliance and 3GPP's Release 19 work are converging on common AI/ML interfaces. Operators who formalised co-creation agreements now hold the architectural influence. Vendors not inside these agreements are competing for the residual market.

T-Mobile's hybrid model, Nokia radio software with NVIDIA compute on shared hardware opens a third configuration that neither Ericsson nor Nokia fully controls. Ericsson and Nokia are not on the same side of these deals. SK Telecom chose Ericsson. Deutsche Telekom chose Nokia. That split means both vendors retain credible Tier-1 references, but neither can claim a dominant position going into 6G.

Strategic analysis: Vendor strategy has shifted from feature leadership to ecosystem integration. The operators who coordinated these March deals have effectively set the terms for the next round of vendor competition. Differentiation will now move to integration execution and regional support , not feature exclusivity. Vendors who cannot demonstrate ecosystem fit within an operator's existing infrastructure will lose deals to those who can.

What this means for you: Watch which vendor configurations your regional peers are locking into. The co-development deals formalised in March will shape 6G radio architecture. Operators outside these agreements will be buying from platforms shaped by operators who were inside them.

On 10 February 2026, Aira Technologies and Axiata announced a partnership to accelerate AI-native autonomous network deployment across Axiata's operations in Malaysia, Indonesia, Bangladesh, Sri Lanka, Cambodia, and Nepal. The scope covers autonomous network operations — AI-driven fault detection, self-healing, predictive maintenance, and energy optimisation — across a footprint of more than 170 million subscribers.

Axiata does not have Nokia Bell Labs co-development agreements or NVIDIA compute co-investment. It has a different problem: a multi-country, multi-technology network spanning six regulatory environments, constrained capex, and high customer growth pressure. The Aira partnership addresses that constraint directly — deploying AI-native operations as a managed service layer on top of existing infrastructure, without requiring a full network rebuild.

The deployment logic mirrors patterns HYPHN has tracked across African operators. When capital constraints prevent overbuild-and-optimise cycles, AI moves earlier in the network operations process. Axiata cannot afford to deploy, measure, and then optimise retroactively. AI-driven autonomous operations reduce the cost of running diverse infrastructure at scale.

The contrast with European deployments is instructive. Deutsche Telekom is co-creating AI-native RAN at the architecture level. Axiata is deploying AI-native operations at the management layer. Different layers, different constraints, same direction: networks running with less human intervention, at lower opex, with faster fault response.

Strategic analysis: Southeast Asian operators are reaching production AI faster than European peers on the network management side, even while European operators lead on RAN architecture. The difference is implementation path. European operators have the capex to redesign the RAN. Southeast Asian operators are optimising operations within existing infrastructure. Both paths lead to autonomous networks. Southeast Asia's path is faster to deploy, cheaper, and generating measurable ROI now.

What this means for you: The autonomous network is not one deployment — it is a stack of AI layers across RAN, transport, core, and operations. Operators who cannot start at the RAN layer can start at the operations layer and deliver commercial results within months. Axiata's model is replicable across any multi-country, capital-constrained environment.

On 6 April 2026, Red Hat published its blueprint for an open, intelligent, and efficient AI-RAN era at the Future of RAN Summit hosted by TelecomTV. The framework addresses one of the most contested questions in the RAN market: can open, cloud-native infrastructure deliver the performance and reliability required for production AI-RAN, or does AI-native RAN require proprietary hardware integration?

Red Hat's position is that open platforms built on containerised, cloud-native architecture can support AI-RAN workloads at production scale, provided operators standardise on open interfaces from the O-RAN Alliance specification and use purpose-built AI compute frameworks rather than adapting general-purpose cloud platforms.

Publishing this five weeks after Deutsche Telekom, T-Mobile, and SK Telecom formalised bilateral co-development agreements is not coincidental. Red Hat is positioning open infrastructure as the neutral path for operators not inside those deals. For every Axiata or regional Tier-2 operator without access to the co-creation track, the open AI-RAN path through Red Hat's blueprint is the alternative.

Production results from the same week support the case. An 8 April Open RAN in Production session documented operators across Europe reporting double-digit energy savings and measurable opex reductions from AI-enabled RAN deployments on open infrastructure.

Strategic analysis: The open versus proprietary RAN debate has been running since 2019. What changed in Q1 2026 is that production data now exists on both sides. Proprietary AI-RAN co-development is generating measurable field results. Open AI-RAN deployments are generating comparable numbers on energy and opex. The vendor lock question is now a strategic choice with evidence, not a theoretical risk.

What this means for you: If your RAN roadmap does not yet have a documented position on open versus proprietary AI integration, the April 6 Red Hat blueprint and the April 8 O-RAN production data are the two documents to review before your next vendor negotiation.