The Ice Age of AI

I've been thinking about ice.

Watching The House of Guinness, I realised that ice used to be centralised. In the late 19th century, it was harvested in big blocks, stored in warehouses, and delivered daily from a central depot. It was a service: heavy, expensive, and reliant on a lot of infrastructure.

Then came the refrigerator, and everything changed. What had been industrial became personal. Cold became local.

The same thing happened with computing.

My dad worked in tech through the 1970s and 80s, selling mainframes for ICL, Honeywell, IBM, and later Siemens. Back then, computing power was centralised too. Offices were full of terminals, screens with no processors, all connected to a single powerful machine humming away in a sealed room somewhere. That was the future, or so everyone thought: big iron in the middle, smart people at the edges.

But that future didn't last.

Moore's Law meant ever-increasing processing power in a smaller space, so the room-sized company computers of the 20th century gave way to the more powerful pocket computers of the 21st. What had been rare and expensive became abundant and ordinary.

And now, it seems, we're in another version of the same story; this time with intelligence itself.

Today's AI runs on enormous clusters of GPUs, housed in billion-dollar data centres, powered by staggering amounts of energy and water. It's centralised in every sense of the word. But history suggests that won't last forever.

As chips get faster and smaller, and as open-source models improve, intelligence will move closer to us: to our laptops, our phones, even into the tools we use every day. Personal AI, running privately, locally, efficiently; intelligence that doesn't need to live in the cloud to be useful.

What the refrigerator did for ice, and the PC did for computing, small models might do for AI.

We may be living in the ice age of artificial intelligence, where everything still depends on central infrastructure. But sooner or later, the thaw will come.