AI Traffic Is Reshaping Networks—and Open RAN Operators Need to Pay Attention

If you run a mobile network, you’re used to traffic growing. But the next wave isn’t just more video streaming or social media. It’s AI agents talking to other AI agents, and it could change the rules of the game. Recent forecasts suggest AI-related network traffic could triple within three years—and that’s just the beginning. For operators, especially those investing in Open RAN, this isn’t a distant problem. It’s happening now, and it’s forcing tough decisions about capacity, latency, and how flexible their networks really are.

The Rise of the Machines (Talking to Each Other)

We’ve all heard about AI writing emails or generating images. But the next big thing is “agentic AI”—software that doesn’t just respond to a prompt but takes actions on its own. Think of a virtual assistant that not only books your flight but also checks you in, orders a car to the airport, and monitors for delays, all without you lifting a finger. These agents don’t just talk to humans; they talk to each other, and they do it constantly.

That’s a whole new type of network traffic. Instead of a person watching a video for an hour, you have bots exchanging small bursts of data millions of times a day. Cisco forecasts that this kind of AI-related traffic could triple in the next three years, and some think it could grow even faster. At a recent industry event, AT&T’s SVP for public policy, Giulia McHenry, noted that the operator has seen a 15% increase in overall data traffic since 2023, and while not all of it is AI, they’re making sure they’re ready for it.

Even more striking is what’s happening on networks like Lumen. According to their SVP, up to 50% of internet traffic on their network is now driven by autonomous agents. That’s a massive shift. If you’re a mobile operator, you need to ask: are we ready for a world where half our traffic is machine-to-machine?

The Triple Threat: Capacity, Permits, and Supply Chains

It’s not just about having enough bandwidth. A recent Light Reading article highlights three big challenges: capacity, permits, and supply chains. Let’s break those down.

Capacity is the obvious one. More traffic means you need more spectrum, more cell sites, more backhaul. But spectrum is finite, and building new sites is expensive and slow. Operators are already feeling the squeeze. They’re looking at technologies like network slicing to prioritize different types of traffic, but that only helps if you have the underlying capacity.

Permits are a huge bottleneck. In many countries, getting approval to install a new antenna can take months or even years. With AI traffic growing so fast, the traditional timeline for network expansion just doesn’t cut it. Operators need to find ways to accelerate deployment, or they’ll be left behind.

Supply chains are another headache. The global chip shortage taught us how fragile the supply chain can be. For Open RAN, which relies on a mix of hardware and software from multiple vendors, supply chain issues can be even more complex. If you can’t get the servers or radio units you need, your rollout stalls.

These three threats are interconnected. You can’t solve capacity without addressing permits and supply chains. And for Open RAN operators, who are often trying to build more flexible, software-defined networks, these challenges are especially acute.

Latency: No Longer a Nice-to-Have

In the past, low latency was a bonus—great for gamers, but not critical for most users. That’s changed. AI applications, especially autonomous agents, require real-time responses. If an agent is controlling a robot or making a split-second trading decision, even a few milliseconds of delay can be disastrous.

As Lumen’s SVP put it, “Latency is no longer a preference. There’s a floor on latency for many of these AI use cases and applications.” That means operators have to guarantee low latency, not just hope for it. This has huge implications for network design. You can’t just rely on centralized data centers; you need edge computing to bring processing closer to the user. And you need a network that can dynamically route traffic to minimize delay.

Open RAN has a potential advantage here. Because it uses software-defined networking and open interfaces, it’s easier to implement intelligent routing and edge computing. But it also requires a level of automation that many operators haven’t yet deployed.

Programmability: The New Table Stakes

As AI traffic grows, operators need more control over their networks. They need to be able to programmatically allocate resources, adjust policies, and respond to changing conditions in real time. This is where the concept of a “programmability layer” comes in.

In the past, networks were largely static. You configured them, and they ran the same way until you changed something. But with AI, traffic patterns can change in an instant. One minute you’re handling a normal load, the next you’re dealing with a swarm of agents all trying to access the same service. The network has to adapt automatically.

This is a key area where Open RAN can shine. The whole point of Open RAN is to disaggregate hardware and software, making it easier to innovate and customize. With a programmable network, operators can deploy new AI-driven features quickly, without waiting for a vendor to release a new software update.

But programmability also brings challenges. It requires a skilled workforce, robust automation tools, and a security framework that can handle the complexity. And it’s not something you can just switch on overnight.

What Does This Mean for Open RAN?

The rise of AI traffic is both a challenge and an opportunity for Open RAN. On one hand, Open RAN is often seen as more flexible and cost-effective than traditional, proprietary networks. That flexibility could be a major advantage in an AI-driven world. On the other hand, Open RAN is still maturing, and some operators worry about performance and reliability.

One thing is clear: the network of the future needs to be AI-native. That means not just supporting AI applications, but using AI itself to manage the network. Ericsson, for example, has been testing AI features in the RAN with SoftBank in Japan, claiming a 25% improvement in spectral efficiency and a 50% boost in downlink throughput. That’s a big deal. If AI can make networks more efficient, it could help solve the capacity problem.

But AI-native networks also require a different approach to design. As Gabriel Brown of Heavy Reading points out, future 6G networks will likely have an “AI plane” that runs alongside the traditional data and control planes. This AI plane would handle tasks like traffic prediction, resource optimization, and automated troubleshooting. Open RAN, with its open interfaces and software-centric approach, is well-suited to accommodate this.

The Role of Network Slicing

Network slicing is another tool that will become increasingly important. With slicing, you can create virtual networks tailored to specific use cases. For example, you could have a slice for autonomous vehicles with ultra-low latency, another for IoT devices with low bandwidth, and another for streaming video with high throughput. This allows operators to guarantee quality of service for different AI applications.

EE in the UK recently demonstrated network slicing in a 5G context, showing how it can provide dedicated capacity for specific services. For Open RAN operators, slicing is particularly relevant because it relies on software-defined networking, which is a core principle of Open RAN.

The AI ‘Death Zone’ and Its Impact on Telecom

While we’re talking about AI, it’s worth noting the broader context. A recent Fortune article describes an “AI death zone” where many corporate AI strategies are failing. The key insight is that the gap between frontier AI models (the best in the world) and commodity models (good enough and cheap) is widening. For telecom operators, this matters because they need to decide where to invest in AI.

Should they build their own AI models for network management, or use off-the-shelf solutions? The article suggests that being in the middle—neither the best nor the cheapest—is dangerous. Operators need to pick a lane: either invest heavily in cutting-edge AI or adopt cost-effective open-source models. For Open RAN, this might mean using open-source AI frameworks to keep costs down.

What Operators Should Do Now

So, what should an Open RAN operator do to prepare for the AI traffic wave? Here are a few practical steps:

1. Assess Your Network’s AI Readiness: Look at your current capacity, latency, and programmability. Where are the bottlenecks? Can you dynamically allocate resources?

2. Invest in Automation: AI traffic requires automated responses. If you’re still doing manual configuration, you’ll be overwhelmed. Start building automation capabilities now.

3. Embrace Open Interfaces: Open RAN gives you flexibility. Make sure you’re leveraging open APIs and standards to enable programmability.

4. Consider Edge Computing: To meet latency requirements, you’ll need to process data closer to the user. Explore edge computing solutions that integrate with your RAN.

5. Monitor AI Traffic Patterns: Don’t wait until you’re overwhelmed. Use network analytics to understand how AI traffic is growing and where it’s coming from.

6. Collaborate with AI Experts: You don’t have to do it alone. Partner with AI companies or join industry groups like the AI-RAN Alliance to share knowledge and best practices.

The Bottom Line

AI is not just another application on your network; it’s a fundamental shift in how networks are used. The traffic from autonomous agents will force operators to rethink capacity, latency, and programmability. Open RAN, with its flexibility and openness, is well-positioned to help operators meet these challenges. But it won’t happen automatically. Operators need to be proactive, investing in automation, edge computing, and AI-native management.

The future is coming fast. Are you ready?

Sources

• EE puts 5G in Fast Lane with Network Slicing - Light Reading
• AI networks face a triple threat: Capacity, permits and supply chains - Light Reading
• Why the first AI-native 6G cores will run on 5G foundations - Light Reading
• The AI ‘death zone’ is here and most corporate AI strategies are standing in it - Fortune