Instead of simply carrying packets from one endpoint to another, a Neighborhood AI Factory is also running inference and training between humans and AI
You’ve probably driven past one without realizing it. A low, windowless brick building next to a cell tower or up on a ridge by the antenna site. In the telco world, we call them central offices or hub sites. They’re locked down, with multiple doors and access points, often a security guard, and always racks of equipment humming in the dark. For decades, these sites quietly powered our connectivity, handling switching, routing, and regulated communications.
That world is changing. Those same buildings are now being refitted as Neighborhood AI Factories — secure, near-edge compute hubs packed with NVIDIA GPUs, designed to run AI workloads right next to the people and businesses that use them.
From phone switches to AI powerhouses
These central offices were built for one thing: trust and uptime. Under the FCC’s CPNI rules, they carried conversations that could never be tapped, sold, or mined. That’s one reason a plain old phone call often sounds better and has lower latency than a FaceTime audio call; it’s running on a regulated telco network rather than the open internet. It remains the most secure form of communication, second only to face-to-face conversation.
For most of their history, these sites handled specialized workloads: small, predictable, low-latency tasks like switching voice calls. Massive cloud capacity wasn’t needed. Meanwhile, Amazon looked at its underutilized servers and turned that surplus into AWS, a $100 billion business in 2024. Telcos, focused on their core workloads, missed that wave.
Now, the opportunity is back, and this time, the stakes are higher. Will AI workloads run on the network or Over-the-Top (OTT)?
The full-circle journey of compute
If you follow the path of compute power over the past two decades, you can see why it’s finding its way back into telco infrastructure.
In the early 2000s, NVIDIA wasn’t selling “GPUs” the way we know them today. They were high-performance CPUs with programmability, marketed for gaming. That programmability made them ideal for special-purpose computing — multi-threading, matrix math — tasks CPUs weren’t built for.
A decade later, GPUs were powering blockchain mining rigs, settling transactions at speed. Then came the AI explosion: Large Language Models and, increasingly, Small Language Models, which deliver high-precision results while demanding far less compute. GPUs handle these workloads exceptionally well.
The result is a full-circle moment: these same chips are now going into telco central offices, co-located with the radio access network (RAN) and tied directly to high-capacity backhaul. In other words, they’re physically close to the user, just like the old voice-switching gear used to be, only now they’re handling AI instead of analog calls.
What exactly Is a Neighborhood AI Factory?
Think of it as the same secure telco hub, but with a new purpose. Instead of simply carrying packets from one endpoint to another, it’s also running inference — or even training — between humans and AI.
This is where the distinction from cloud or even traditional edge computing matters. In cloud, a request might travel hundreds or even thousands of miles. With a Neighborhood AI Factory, it’s processed within your metro area, sometimes just a single hop from your handset. That proximity slashes latency and brings greater security to data handling.
Why telcos have the advantage
The reason this shift works in telco territory comes down to three things:
- Distribution: Thousands of central offices and hub sites are already embedded in communities. No hyperscaler can match that footprint overnight.
- Spectrum Ownership: Every packet of wireless data rides on licensed frequencies (regulated, high-capacity “pipes” that only telcos can operate).
- Trust: Privacy in telecom isn’t just a promise; it’s backed by law. The FCC’s Communications Act and its updates give consumers protections OTT providers aren’t required to match.
The opportunity is massive. According to the 2025 Telco AI Market Pulse Report, operators are projected to invest $9.2 billion in AI infrastructure for internal efficiency use cases and $21.6 billion for external monetization use cases by 2027. With that much capital in play, the question isn’t whether AI will move to the network edge, it’s who will own the value it creates.
Inside the factory
Walk inside one of these upgraded sites and you’ll see GPU servers in neat rows, connected by high-throughput switches to the telco’s fiber backbone. The cooling systems once designed for legacy switchgear now keep AI hardware at optimal temperatures. In many cases, existing infrastructure can be leveraged, avoiding costly greenfield builds.
This setup is ideal for running Small Language Models. They’re compact enough to operate efficiently on local gear, yet powerful enough to deliver personalized, low-latency AI responses to the neighborhoods they serve.
What happens when AI is this close?
Once AI runs this close to the user, entirely new categories of service become viable:
- Manufacturing: A factory or warehouse leveraging supply chain and robotics automation can run workloads with ultra-low-latency, driving real-time physical actions without needing the cloud.
- Small Business: A mechanic, plumber, or deli can run an AI Receptionist near-edge, fielding customer queries and taking actions, directly through the network.
- Smart Cities: Traffic signals, public transit sensors, emergency systems, and IoT devices can run workloads in real time without the delays of cloud round-trips.
In every case, the workload stays on the telco’s network, which means faster responses and stronger privacy guarantees. That’s fundamentally different from OTT services which run over Wi-Fi and aren’t bound by FCC privacy rules.
It’s a way to deliver AI value without letting every interaction become someone else’s training data.
A new role for telcos
For telcos, this presents a business model transformation opportunity. Instead of just carrying traffic, they can deliver trusted, hyperlocal intelligence services.
The strategic question is simple: Do you want to give the next $100 billion opportunity to Google or Meta again? History offers a warning. Before Netflix, some telcos launched mobile video streaming. The bandwidth and device compute weren’t there yet, so those efforts stalled — and streaming went to the tech giants. With AI, the hardware, models, and demand are here now.
The next three years
Looking ahead, it would be a missed opportunity if, within three years, the 30 million small businesses in the US weren’t using AI services from their local telco hub. The infrastructure exists, the regulatory trust exists, and the technology is proven.
Hyperlocal AI is the natural evolution of telecom. The central offices are ready to host it, the spectrum is already licensed, and the expertise to run it sits within the industry. All that remains is the will to lead.
