Global carriers and industry leaders including Nvidia and Nokia have joined forces to invest $45 million in AI-RAN specialist ODC.
This Series A round—backed by a diverse mix of silicon designers, networking incumbents, legacy radio vendors, and major telecom operators such as AT&T, MTN, and Telecom Italia—represents an uncommon alignment of interests. Seeing an accelerator like Nvidia, a traditional network supplier like Cisco, and long-established radio equipment providers participate alongside global operators in an early-stage financing highlights the strategic importance of ODC’s work.
Their collective support reflects a growing industry consensus: the radio access network (RAN)—the most capital- and energy-intensive part of any cellular system—is approaching its physical and economic limits under current architectures.
ODC is developing technology that embeds AI directly into the radio baseband, replacing fixed, human-coded signal-processing rules with machine learning models that adapt in real time to changing network conditions.
For companies that depend on sophisticated connectivity—industrial automation, large-scale IoT rollouts, or private enterprise networks—this funding signals the direction corporate connectivity is taking: a shift toward software-defined, intelligent platforms that learn and optimise autonomously rather than relying on constant human tuning.
Ronnie Vasishta, SVP of Telecom at NVIDIA, explained that the industry is moving toward software-defined, AI-native networks that will be central to the emerging physical AI era. He said ODC’s AI-RAN stack helps convert current 5G deployments into a distributed AI compute fabric on the wireless edge and, by leveraging the NVIDIA Aerial platform, raises the bar for AI-RAN innovation while paving a path toward 6G.
Economic imperative for intelligent radio
To grasp why a $45 million investment is attracting so much attention, consider the core economics of modern telecoms.
The RAN consumes the bulk of an operator’s capital expenditure as well as most ongoing energy usage. Deploying and maintaining cell towers, antennas, and baseband units that process radio signals is expensive and operationally demanding.
Meanwhile, as global data traffic keeps growing, average revenue per user remains largely flat. Operators must reduce the cost per bit transmitted and squeeze more capacity from existing spectrum, driving interest in technologies that improve efficiency without requiring large new capital outlays.
Open RAN and virtualised RAN concepts aim to commoditise hardware and reduce vendor lock-in, but replacing custom silicon with general-purpose processors often introduces performance and power-efficiency penalties.
AI-RAN proponents like ODC argue that embedding AI into the physical layer solves this economic dilemma. Machine learning models operating in the baseband can improve spectral efficiency, predict hyper-local demand patterns, and dynamically manage power at the antenna—optimising capacity and cutting energy use.
Conventional radio networks transmit at relatively constant power levels based on pre-set parameters. An AI-native RAN can learn the unique topographical and behavioural characteristics of each cell sector and adjust resources accordingly.
For example, if an enterprise campus clears out in the evening, an AI model can autonomously power down specific frequency bands to save energy, then reactivate them milliseconds before a scheduled automated process begins. This kind of real-time, fine-grained resource management offers operators a credible path to improved profitability in a 5G world.
However, moving to AI-driven radio architecture requires a fundamental redesign of edge infrastructure. Legacy baseband units cannot simply receive a software update and start running complex inference workloads; they need to be replaced or augmented with platforms capable of high-performance computing at the cell site.
Merging IT-grade compute—GPUs and AI accelerators—with telecom-grade hardware in harsh, space-constrained environments presents logistical and engineering challenges. Edge sites lack the climate control, power redundancy, and physical space of hyperscale data centers, while AI accelerators generate heat and demand consistent power. Managing thermal and power dynamics across thousands of distributed AI nodes is a major operational hurdle.
Beyond hardware, the data demands are substantial. Training and refining RAN models requires ingesting, normalising, and analysing petabytes of radio telemetry each day. Many operators lack the data maturity and backhaul capacity to move that volume to centralized locations, making local edge processing essential to meet latency and bandwidth constraints.
Deploying sophisticated distributed software to process and act on telemetry at the edge is therefore necessary, but it also represents a steep learning curve for operators accustomed to more centralized, deterministic network stacks.
Governing the probabilistic network
A major conceptual challenge is shifting from deterministic engineering practices to probabilistic operations driven by AI.
Telecom infrastructure has long been designed for absolute predictability: specific inputs yield specific, standards-driven outputs. Machine learning models instead infer and predict, which introduces new operational risks. If an AI model misinterprets traffic patterns and disables a macro cell during an emergency, the consequences could be severe, drawing regulatory scrutiny and eroding enterprise trust.
To mitigate this, governance frameworks and guardrails must be established before AI systems control live traffic. Network engineers will need to enforce deterministic boundaries around AI recommendations so that physical equipment cannot exceed safety and operational limits regardless of model output.
Creating these safeguards requires combined expertise in radio frequency physics and machine learning validation—skills that rarely coexist in the same teams today.
Pallavi Mahajan, Chief Technology and AI Officer at Nokia, noted that AI introduces a new workload that reshapes network architecture, driving the need for software-driven platforms, intelligence at the edge, and continuous innovation. She said ODC’s AI-RAN approach aligns with the industry’s direction toward AI-ready RAN platforms across 5G and future 6G networks.
The human side of this transition is equally important. Network operations teams have historically specialised in hardware deployment and deterministic protocol management. Integrating AI adds requirements for data science, MLOps, and modern software engineering skills.
Operators will need to upskill their workforce and break down silos between IT and network engineering. Radio engineers and data scientists must develop a shared language to jointly troubleshoot models that misallocate bandwidth or mismanage resources during peak periods.
Machine learning also demands continuous retraining. The radio environment evolves with seasonal foliage changes, new construction, and shifting user behaviours. Models require ongoing updates, a robust deployment pipeline, and operational processes many operators are still building.
Why the heavyweights are backing AI RAN specialist ODC
The mix of investors in ODC’s round demonstrates a strategic competition over the future of the telecom edge. Nvidia and other silicon designers see cell towers as a major opportunity for enterprise AI compute, while network vendors and operators seek ways to future-proof infrastructure and extract more value from edge assets.
By enabling AI workloads on the same silicon that manages radio functions, operators could monetise idle compute cycles—offering local processing to enterprises during periods of low network demand, for example. For organisations planning private 5G deployments or campus upgrades, it is now critical to evaluate vendors’ AI-RAN roadmaps and ensure purchased hardware will support future machine learning workloads to avoid stranded assets.
Enterprises should also treat network telemetry as a strategic data asset. Intelligent infrastructure will depend on high-quality, well-governed data from connected devices. Companies that have invested in structuring, cleaning, and securing network data will be best positioned to leverage AI-enabled networks to drive operational efficiency.
The era of static, predictable networks is giving way to adaptive, learning systems that operate in real time—promising greater efficiency but requiring significant changes in architecture, operations, and culture.
Telecoms is powered by TechForge Media. Explore related enterprise technology events and webinars through organisers’ official channels.