AI Needs Its ISP Moment

In the 1990s, a handful of companies built the internet backbone. Fiber, routers, peering agreements, the whole plumbing. It mattered enormously, and almost nobody got online because of it.

People got online because a local ISP in their town sold them a modem and a dial-up account. That ISP answered the phone at nine at night when the connection dropped. It hosted the first web page for the hardware store on Main Street. It hired the kid down the block who turned out to be good with computers, and that kid became the town's first webmaster.

The backbone made the internet possible. The local layer made it real. I ran that layer. I was building commercial bulletin board systems before graphical interfaces existed, and in 1995 I founded PRIMELINE Internet Services, the first ISP in Western North Carolina, and ran it until we sold it in 1999. After that I built ISPs for underserved rural towns from Indiana to South Carolina, the ones the big providers wouldn't touch. I've been a developer for 30 years, and I watched adoption happen one small business at a time, one phone call at a time, in towns nobody in Silicon Valley could find on a map.

AI is in the same position the internet was in around 1994. The backbone problem is solved. The frontier models are extraordinary, capable of work that would have sounded like fiction a few years ago. And most businesses in most towns still have no idea what to do with them.

Almost every dollar in this industry is going into hyperscale data centers. That investment is necessary and I'm not arguing against it. But almost nothing is going into the layer that made the internet actually stick: local operators who onboard businesses and residents, train local talent, and figure out how a new technology fits into how a community already runs its errands, its schools, and its town government.

Here's the proposal. OpenAI should build a local operator program, modeled on how ISPs worked.

The frontier models stay exactly where they are. A model like that cannot run in a strip-mall office, and nobody should pretend otherwise. The local layer was never about owning the backbone, not for ISPs and not for AI. It's about access, integration, onboarding, data, and people. An ISP never ran its own long-haul fiber either. It plugged into it and did the work of connecting a town.

A licensed local operator in each market would onboard businesses, schools, local government offices, and residents. They'd build and maintain the integrations that make the models useful in a specific place, doing specific work. They'd host what can reasonably be hosted locally. And they'd hire and train in-house talent, the way that first webmaster got hired, so the expertise lives in the town and not three time zones away. OpenAI's role would be to invest in that community infrastructure and in operator capacity, alongside the data centers, not instead of them.

Here is what the ISPs actually did, and it's the part everyone forgets. They didn't just sell connections. They saturated their towns. Within a few years the internet was in the vocabulary of the whole community. The hardware store put its URL on the truck. The school taught it. The chamber of commerce held classes in it. The newspaper ran a column about it. Businesses rebuilt their processes around it because somebody local showed them how, and the local operator got paid for every one of those engagements. The market didn't grow because the backbone got bigger. It grew because every town had someone whose whole livelihood depended on getting their neighbors online.

That is the market OpenAI is leaving on the table. A cloud service, however good, sits in a browser tab waiting to be discovered. A local operator goes and finds the market, business by business, and pulls the models into the daily processes of the whole community until they're simply how things get done there. That's saturation, and there's no other way to get it.

It also answers the question people are afraid to ask out loud. AI arriving as a remote service means the work leaves town and nothing comes back. AI arriving through a local operator means the opposite: new jobs that didn't exist, local talent curated and trained where it lives, the town's own people running the orchestration layer instead of watching it happen to them. The community stays in charge of how AI gets used, taught, and trusted, because the people doing it are their neighbors. That's what makes adoption durable instead of a novelty and trusted instead of imposed. And it's the part nobody at scale is building right now.

I'm not proposing this from the sidelines. In 2000 my wife and I started WNC Times, a community news site for Western North Carolina. She ran the paper, twenty-six years of editorial judgment and more than 20,000 articles; I ran the technology. In 2024 I moved it from Joomla to Next.js and AI entered my daily workflow, and it now runs as an AI-assisted newsroom. I productized that work into NewsroomAIOS, a platform that lets a community launch its own AI-powered local newspaper.

From mid-April to mid-June of this year, I ran a 17-paper beta across 16 US markets, including Asheville, Atlanta, Chicago, Colorado Springs, Jonesboro, Kingston, Miami, Minneapolis, Oceanside, Omaha, Phoenix, Portland, St. Petersburg, San Diego, Seattle, and Xenia. It was a load test on the platform and the AI content pipeline, nothing more, and I'm stating it plainly for that reason. Two of those papers run live today: WNC Times and Oceanside News.

That's the local-operator model already working, in one vertical, local news, built by one developer. Now imagine that as a program, with a company the size of OpenAI behind it, across every vertical a town actually has: retail, healthcare, schools, city hall.

I've run the local layer twice now, once in the dial-up era and once in this one, and I'd put what I've learned on the table with anyone at OpenAI working on how these models actually reach the rest of the country.

Carl Farrington, Founder & Principal Engineer, Farrington Development LLC, Asheville NC