AI-RAN’s $1.1 Billion Question: Where Should the Intelligence Actually Live?

Every few months, a new report lands on my desk with a big number attached to AI and radio access networks. Most of them are easy to skim and forget. But a fresh forecast from SNS Telecom & IT caught my attention this week for a different reason: it puts a precise figure on a market that has been maddeningly vague, and it quietly frames the real fight that will shape the next five years of network architecture.

The headline number is this: annual spending on AI-RAN automation and optimization software and services is expected to reach $1.1 billion by the end of 2030, growing at a compound annual growth rate of about 9% over the next four years. That’s not a giant market in telecom terms. It’s smaller than many single vendor product lines. But the report’s real value isn’t the dollar figure — it’s the taxonomy it uses to break the market apart, and what that taxonomy reveals about where the industry is placing its bets.

If you’re an operator, a vendor, or just someone trying to understand where Open RAN is heading, this is worth understanding. Because the $1.1 billion isn’t spread evenly. It’s concentrated in a specific layer of the network, and the scramble to control that layer is already well underway.

What exactly is AI-RAN, anyway?

Before we get into the numbers, let’s clear up the terminology, because the industry has done a spectacular job of making this confusing.

The SNS report breaks the broader AI-RAN movement into three distinct pieces, and this is genuinely useful:

AI-for-RAN is the most commercially mature piece. This is about using AI and machine learning inside the radio access network to automate and optimize how it works — replacing static, rule-based algorithms with software that learns and adapts. Think of it as giving the network a brain for its own operations.

AI-and-RAN is the co-hosting story. This is about running RAN workloads and external AI workloads on the same shared, GPU-accelerated infrastructure. The idea is that a base station site could simultaneously handle your phone call and process AI inference for some other customer.

AI-on-RAN is about enabling edge AI services for mobile subscribers — using the RAN as a platform to deliver AI applications to end users.

Of these three, AI-for-RAN is where the money is today. And within AI-for-RAN, there’s a further split that matters enormously.

The two paths to smarter networks

Here’s where it gets interesting. AI-for-RAN spending is flowing into two very different architectural approaches, and they represent fundamentally different visions of how networks should be built.

Path one: Centralized automation. This is the SMO and RIC approach — Service Management and Orchestration, plus the RAN Intelligent Controller. In plain English, this means putting the intelligence at the network level, in centralized software platforms that coordinate across many cell sites. Operators deploy Non-Real-Time RICs that run rApps — small applications that handle things like network-level optimization, energy management, and policy enforcement. The advantage is coordination: one brain overseeing many cells can spot patterns and make decisions that individual sites can’t.

The report notes that mobile operators are scaling SMO and Non-RT RIC deployments for exactly this kind of network-level automation through AI/ML-enabled rApps.

Path two: Embedded intelligence. This is the AI-for-RAN software upgrade approach — putting AI directly into the RAN nodes themselves, replacing static rule-based algorithms with embedded real-time intelligence. The advantage here is speed: decisions happen right at the radio, without a round trip to a central controller.

The report describes this as “AI-for-RAN software upgrades for embedded real-time intelligence and the replacement of static rule-based algorithms in RAN nodes.” That’s a mouthful, but the point is simple: instead of centralizing intelligence, you distribute it.

Both paths are growing. But they lead to very different network architectures — and very different competitive dynamics.

The xApp problem and the E3 escape hatch

Here’s where the report gets really candid, and where the Open RAN community should pay close attention.

In the pure Open RAN vision, the Near-Real-Time RIC (Near-RT RIC) runs xApps — applications that control the RAN with latency measured in the 10-millisecond to 1-second range. These xApps are supposed to be the heart of Open RAN’s intelligence layer, enabling multi-vendor innovation and competition.

But according to SNS Telecom & IT, adoption has been held back by “reluctance among DU/CU vendors.” That’s a polite way of saying that the companies building distributed units and central units — the baseband processing hardware and software — haven’t been eager to hand over control to third-party xApps running on someone else’s RIC. It’s a classic platform battle: whoever controls the interface controls the ecosystem.

This is not a new story. Anyone who has followed Open RAN for the past few years knows that the RIC has been the great promise and the great disappointment. The interfaces exist. The standards are written. But commercial deployments of multi-vendor xApps have been slow, because the vendors with the most to lose from openness have the least incentive to accelerate it.

Now here’s the twist: the report highlights growing interest in something called the E3 interface. This is a newly proposed standard that would let dApps — a different class of application — directly interface with Open RAN baseband nodes for closed-loop control. The key difference is latency: E3 would operate below the 10-millisecond limit that constrains xApps, enabling use cases like lower-layer performance optimization, geolocation, integrated sensing, spectrum sharing, and interference cancellation.

In other words, E3 is an attempt to route around the bottleneck. If vendors won’t open up the Near-RT RIC, maybe they’ll open up a different interface that does similar things but faster. Whether that works depends entirely on whether the same vendors who resisted xApps will embrace dApps. Color me skeptical, but it’s a development worth watching.

The proprietary alternative

The report also notes that some vendors are pursuing an entirely different strategy: offering proprietary AI-enabled, real-time optimization capabilities as software upgrades to existing RAN platforms, sold through subscription-based licensing models.

This is the path of least resistance for incumbent vendors. It doesn’t require opening interfaces or sharing control. It doesn’t require multi-vendor coordination. It just requires operators to pay a subscription for smarter software on hardware they already own.

And here’s the uncomfortable truth: for many operators, this might be the most practical option. The dream of a vibrant, multi-vendor app economy on the RIC has been slow to materialize. Meanwhile, vendors are shipping real AI features that deliver measurable improvements in energy efficiency, spectral efficiency, and operational autonomy. If you’re running a network and your board wants results, the proprietary upgrade path is right there.

The $1.1 billion figure, then, isn’t just a market forecast. It’s a scoreboard in a contest between open and closed approaches to network intelligence. And right now, the closed approaches have momentum.

Why 6G changes the calculus

The report makes an important point about timing: the first 6G networks are expected to launch around 2030, and they’re being standardized as AI-native systems from the outset. That’s a significant shift from previous generations, where AI was bolted on after the fact.

If 6G is truly AI-native, then the question of where intelligence lives becomes foundational rather than incremental. You’re not adding AI to an existing architecture; you’re designing an architecture around AI from day one. That’s a much bigger opportunity for hardware-agnostic, software-led innovation — and a much bigger threat to vendors whose value proposition depends on proprietary hardware.

The report also highlights the role of RAN digital twins, which allow operators to validate AI-enabled air interface features and train optimization algorithms in high-fidelity virtual environments before introducing them into live networks. This is a practical necessity when AI systems are making real-time decisions about spectrum, power, and handovers. You can’t just flip a switch and hope for the best.

Digital twins are also a bridge between the centralized and embedded approaches. You can simulate both, compare results, and decide which architecture makes sense for your network and your business model. That’s valuable, and it’s one reason why test and measurement vendors are having a moment right now.

The bigger picture: automation is eating the RAN

Zoom out for a moment, and the $1.1 billion figure looks less like a market forecast and more like a leading indicator of something bigger.

The report notes that total spending on the wider RAN automation market — which includes traditional SON and proprietary network-level automation, AI-for-RAN upgrades, Open RAN SMO/RIC and d/x/rApps, RF planning tools, and automated test and measurement solutions — is expected to grow at a CAGR of approximately 4% over the same period.

That’s slower growth than the AI-RAN slice, which tells you something important: AI is taking share within the broader automation market. Operators aren’t necessarily spending more on automation overall; they’re spending differently. The static, rules-based automation of the past is being replaced by AI-driven systems that learn and adapt.

This is consistent with what we’re seeing across the industry. At IMC 2026 in New Delhi this month, the theme was “Scale Without Boundaries,” and the conversation was dominated by AI, 6G, intelligent RAN, cloud and edge computing. Vendors like VIAVI are showcasing AI-RAN test and validation platforms that support RAN simulation, training, testing, and validation across 4G, 5G, emerging 6G, Open RAN, and non-terrestrial networks. Samsung is demonstrating AI-RAN, X-MIMO, D-MIMO, and ISAC technologies in Silicon Valley. SK Telecom won a Network X innovation award for AI-RAN-based autonomous vehicle transport technology and is planning demonstrations at a South Korean finished-vehicle plant next year.

The momentum is real. But so is the fragmentation.

What operators should actually take away from this

If you’re an operator trying to figure out where to place your bets, here are the questions this report raises:

First, do you believe in the app economy? The Open RAN vision of multi-vendor xApps and dApps running on standardized interfaces is elegant, but it has been slow to deliver. If you’re waiting for that ecosystem to mature before investing in AI-RAN, you may be waiting a long time. The proprietary upgrade path is available now.

Second, how much latency do you actually need? The distinction between xApps (10ms+) and dApps (sub-10ms) matters for specific use cases like interference cancellation and integrated sensing. If your priority is network-level optimization and energy management, the centralized RIC approach may be sufficient. If you need real-time control at the radio layer, you’ll need something faster — and that something may not be standardized yet.

Third, who do you trust to control your network’s intelligence? This is the fundamental question. Centralized RIC platforms give you more control and more vendor options, but they require vendors to open up their baseband interfaces. Embedded AI gives you faster decisions but locks you into a vendor’s proprietary roadmap. There’s no free lunch.

Fourth, what’s your 6G timeline? If you’re planning to deploy 6G around 2030, the architecture decisions you make now about AI-RAN will shape what’s possible later. AI-native 6G is coming whether you’re ready or not. The question is whether you’ll be building on open foundations or proprietary ones.

The quiet significance of $1.1 billion

One billion dollars is a rounding error in the global telecom equipment market. But that’s not the point. The point is that AI-RAN automation is becoming a real, measurable market with real, measurable spending — and the way that spending is distributed will determine the architecture of the next generation of networks.

SNS Telecom & IT’s report doesn’t answer the question of where intelligence should live. But it frames it clearly: centralized or embedded, open or proprietary, standardized or vendor-specific. These are the choices that operators face, and they’re not just technical decisions. They’re business decisions about control, competition, and cost.

The $1.1 billion is a forecast. But it’s also a warning. The window for shaping AI-RAN architecture is open now, and it won’t stay open forever. Once vendors lock in their approaches and operators commit to their platforms, the path dependencies will be hard to break.

If you care about Open RAN — really care about it, not just as a slogan but as a set of principles about openness and competition — this is the fight to watch. Not the headline numbers, but the interfaces. Not the market size, but the architecture. That’s where the future of the network is being decided.

Sources

• AI-RAN Automation & Optimization a $1.1 Billion Opportunity
• AI-RAN Automation Spending to Reach $1.1 Billion by 2030, says SNS Telecom & IT
• Architecting Next-Gen Telecommunications: VIAVI on 6G, AI-RAN, and Digital Twins at IMC 2026
• VIAVI Showcases AI-RAN, 6G and Next-Gen Network Testing at IMC 2026
• SK Telecom Wins ‘Network X’ Innovation Award for AI-RAN Driverless Vehicle Transport Technology
• Samsung showcases AI-driven 6G in Silicon Valley, expands Korea–US carrier ties