THE NEW AI DIVIDE

Could smaller nations (still) use AI to unlock a new wave of economic growth?

Which, given the growing disquiet about AI, feels like a tone deaf question to ask.

But bear with.

Until recently, we have expected everyone to have relatively cheap access to the very best models OpenAI, Anthropic and Google can offer. It's estimated that when you hand over your €20 for access, you receive up to €400 in usage.

But as models have become larger, increasingly capable and more expensive to run, this subsidy - paid for by the AI labs - has swollen their costs of doing business.

And we are reminded by the histories of Amazon, Uber and Netflix that subsidies to build a customer base are unsustainable. At some point, economic gravity kicks in. Leaving an almost-monopolistic market.

Let’s not be dewy-eyed. This is the game the frontier model labs are playing. Yet we are already seeing the start of its end. Because, as models become more capable, enterprise demand curves have gone exponential. And we’re reaching a point where supply cannot keep up. The chips and data centres needed cannot be delivered fast enough.

There are other trends colliding too - the labs are tightening access to the increasingly sophisticated models for security reasons - and to prevent them from being copied. Plus, the US government is starting to wonder if frontier models aren’t a national strategic asset, to be more closely guarded.

All of which poses a problem for consumers and smaller businesses in the US - and almost every other nation on Earth.

As is usual with market-based models, the gains from AI will be unevenly distributed.

So what can middle nations do? Well, when they can source the chips, what the labs need next is electricity and compute. So nations that can deliver those at preferential rates are likely to cut deals for cheaper frontier access. Which is one way smaller nations could (still) use AI to unlock economic growth.

But only if political conditions allow. Which explains why, until the recent US-Iran war, Middle Eastern nations were investing in data centres heavily.

Now the good news. There is also another way to secure continued growth.

The truth is that not every workload needs a frontier model. Most enterprise value can be extracted from the tier below - where pricing pressure is far gentler and access is structurally unconstrained. In fact, many of these models can be run on local machines - outside of vast data centres.

So what’s today’s in-the-end-at-the-end?

UK telecoms monopoly BT re-popularised the phrase: ‘Work smarter, not harder’. And smaller nations will be able to enjoy the full benefits of AI - without the strain on energy grids or water supplies - by applying that principle.

Thus meaning that politics, economics and business practice could find a shared happy place again.