AI shouldn’t be built in one postal code: Cloudflare CEO Matthew Prince

Matthew Prince, Co-founder and CEO of Cloudflare, at the India AI Impact Summit 2026, made a strong pitch for decentralised AI infrastructure anchored in a free and open internet, has said that Artificial intelligence should not be built by a handful of companies in the same postal code.

In a keynote that set the tone for the summit, Prince argued that while AI today appears concentrated and expensive, the factors driving that concentration are unlikely to remain permanent.

Prince broke down what he described as the two core reasons AI remains “very, very hard” and costly at present.

The first is hardware dependency. “AI requires lots and lots and lots of chips,” he said, noting that these are largely produced by one dominant manufacturer and consume significant power. 

Had chips been designed from scratch to build AI systems, he suggested, they would look very different today.

The second constraint is talent concentration. “There’s a very small set of people in the world who know how to build these models and run these systems,” Prince said. Five years ago, he pointed out, AI was still seen as a field full of unfulfilled promises from the 1970s, 80s and 90s. “The AI professor was kind of shunted off to the side,” he remarked.

That perception, however, has changed dramatically. Enrolment in computer science programmes globally has surged, and AI theory courses are “off the charts,” he said.

Prince argued that both talent and hardware bottlenecks are already beginning to ease. “Enrollment in specifically AI theory courses is off the charts. Every university that used to sort of shudder their course is now standing it up and building it like crazy,” he said, adding that over time, “we’re going to have more and more people who are able to do this,” making today’s enormous AI salaries unsustainable in the long run. On the hardware side, he noted that history suggests shortages don’t last forever. “Any time there has been a silicon shortage, it turns into a silicon glut over time,” he said, predicting that the price per unit of AI compute will inevitably decline.

Importantly for startups, Prince stressed that competition in AI infrastructure is widening rapidly. “From startups as well as incumbent players, from hyperscalers and other players that are getting involved, there are so many people who are making this silicon that no matter what, the price per unit of work done is going to come down.” 

In a forward-looking projection, Prince said that within five years, it could be possible to build a frontier-level AI model within a specialised domain for $10 million or less.

“My prediction would be that you’ll be able to build models that are on the frontier — more specialised, but on the frontier — for tens of millions of dollars in the not-so-distant future,” he  said, putting a five-year timeline on that shift.If that plays out, the current hundreds of billions of dollars flowing into leading AI companies may not represent a permanent cost structure.