Why AI in Telecom Is Still Stuck at Pilot Projects (And How to Break Through)

Key Takeaways

• 95% of enterprises investing in generative AI face challenges in achieving meaningful returns.
• 88% of telecom firms expected better ROI from AI, but only 12% are building AI-engineered products.
• 28% of telecom companies remain in the AI pilot phase, down from 49% the previous year.
• 50% of AI ownership in telecom sits solely with the CIO, limiting cross-functional impact.
• Greater than 50% of CSPs cite employee skill gaps as a major barrier to AI deployment.
• AI infrastructure costs are rising due to increased compute, memory, and bandwidth requirements and growth in data center and cloud traffic.
• Oracle laid off ~30,000 employees to fund AI data center investments.
• AT&T committed $250 billion to infrastructure supporting AI and connectivity

Right now, you're probably running at least one AI pilot. Maybe two or three. Some of them are producing results that look good in a slide deck. But none of them are scaling. And if you're being honest with yourself, you already know why, because it's the same reason they didn't scale last year either.

A 2025 MIT report found that 95% of enterprises investing in gen AI had challenges getting any good returns. At the same time, 88% of telecom firms say they expected better returns from their AI initiatives, but only 12% of them are building AI engineered products. So the belief in AI is there. The investments are going in. The pilots are running. The only issue is that the results aren't leaving the lab.

Yet another research has something entirely different to share. NVIDIA's State of AI in Telecommunications report found that 28% of telcos are still in pilot phases, which is down from 49% the previous year. It means that there are certain telecommunication businesses that are actually able to use AI the right way. So, let’s figure out what those companies are doing or what you might be lacking in the process that’s keeping you from getting the ROI you want.

Why AI in Telecom Keeps Stalling

The major problem we’re dealing with here is fragmented data, no unified governance, undertrained teams, and legacy infrastructure that was never designed to support enterprise-wide AI. Let’s discuss this in detail.

1. Fragmented Data and Legacy Infrastructure

Telecom networks generate enormous amounts of data. Every call, every handover, every billing event, every fault log, it's all being captured. The problem is where it lives. Network data sits in one system, subscriber data in another, and marketing, operations, and customer service? They all run on different platforms with different definitions of the same metrics.

AI is only as good as your data, but if you feed it with fragmented data, it amplifies it. Then there's the legacy infrastructure problem on top of that. Most telecom networks are not cloud-native and were built on vendor-specific hardware and customer software stacks. These legacy systems are outdated and were never designed for AI integration. And integrating AI with your existing telecom infrastructure can be complex, time-consuming, and quite expensive.

2. Siloed Ownership and No Real Governance

As we discussed, your AI is only as good as the data you feed it. Among all the bad data challenges, bad organizational structure can be just another one to add. And in telecom, the two show up together.

According to Bounteous research, 50% of AI responsibility in telecom still sits with the CIO alone. Only a small portion of firms have cross-functional governance in place. When your AI ownership lives in one department, you disconnect your AI initiatives from the KPIs that matter for your business. For instance, if cost savings is reported in IT and no one’s actually measuring the customer impact or the revenue outcome AI has created, your business case starts to slowly fall apart.

When AI ownership lives in one department, AI initiatives naturally get disconnected from the KPIs that matter to the rest of the business. Cost savings get reported in IT. But nobody can measure the customer impact or the revenue outcome. So the business case slowly falls apart. Hence, you need to unify your AI initiatives under a single governing body or Center of Excellence (CoE) to avoid silos and capture benefits across teams.

3. The Skills Gap Nobody Is Talking About Loudly Enough

This one is underreported. Everyone talks about the technology gaps. Far fewer executives talk about the workforce gap, which is often the reason technology gaps never get resolved.

A Google Cloud and Analysys Mason study found that over 50% of Communication Service Providers (CSPs) cite employee skillsets as a big concern in their AI deployments. And it’s not just about hiring data scientists and AI engineers, but the broader workforce as well.

4. Costs Are Going Up Without Matching Revenue

Training and deploying large-scale AI models requires serious compute power, memory, and bandwidth. These workloads drive increased traffic across data centers and between cloud regions. Due to this, more and more companies are investing in AI data centers and infrastructure. In fact, recently, Oracle laid off over 30,000 jobs to fund AI data centers, at the same time, AT&T announced to invest $250 billion to boot infrastructure in AI.

The problem is that these costs are not being offset by new revenues. No matter what, investing in AI is the need of the hour, the only thing that differentiate winning leaders is that they ensure that their business case is tied to specific, measurable outcomes; otherwise, the costs will just keep piling up.

Where AI in Telecom Actually Delivers Results

Despite all of the above, we all know that AI is already making a measurable difference in telecom, in targeted, well-defined areas. And these include:

• Predictive maintenance, where AI monitors network equipment and predicts when a component is likely to fail to ensure proactive maintenance before outages happen.
• Cybersecurity, where AI-powered systems detect DDoS attacks and flag suspicious patterns.
• Fraud detection, that can save you millions of dollars and reputational damage.
• Network slicing in 5G by allocating resources to different network slices based on real-time traffic load and quality-of-service (QoS) requirements.
• Overall customer experience with AI-powered virtual assistants, intelligent chatbots, and personalization engines.

How to Actually Break Through: What the 12% Telecom Companies Are Doing Differently

The telecom operators that are genuinely scaling AI share specific practices. Here's what they're doing.

• Fix the data foundation first. This is non-negotiable. You need to break down data silos and establish clear governance before trying to scale AI.
• Prove value in contained use cases before expanding. Don't try to transform everything at once. Pick a use case with a clear ROI path, whether it’s predictive maintenance, fraud detection, or a specific customer service workflow, pick one. And then prove it, measure it, and use that proof to fund the next phase.
• Establish cross-functional AI governance. Pull AI out of IT and create ownership structures that connect AI programs directly to enterprise KPIs, for instance, customer experience, revenue, and operations.
• Make AI fluency a workforce priority, not an afterthought. Build AI into how work gets done, instead of keeping it as a separate skill that employees have to go learn elsewhere.
• Run AI and infrastructure modernization in parallel. You don’t have enough time to modernize your legacy systems entirely before you can deploy AI. Your targeted agents can deliver value today while you fix your legacy bottlenecks.

The Bottom Line for Enterprise Leaders

AI in telecommunications is failing because most organizations are trying to run it on broken foundations. Fragmented data, disconnected ownership, undertrained workforces, and unclear success metrics, you name it.

The majority of AI pilot failures is a reflection of how those pilots have been set up to fail. To lead the next phase of the telecom industry, you’ll need to be the one building the conditions under which AI can actually scale, provide the best ROI on AI investment, and deliver the right business outcomes.

FAQs: