AI’s next chapter

It’s getting louder. The voice of doubt in the age of AI exuberance.

From The Economist to The Financial Times, from venture investors to developers, the consensus is beginning to fracture. The trillion-dollar AI boom that promised to remake the economy is starting to show its cracks.

And that’s a good thing. Because healthy skepticism isn’t the enemy of innovation, it’s the foundation of value creation.

For the past two years, capital markets have treated AI as a suffix that adds a zero to every valuation. From single players like OpenAI valued near half a trillion dollars to flagship labs pulling in $50–100 b+ valuations, the industry is treating foundation models as gold mines, even when revenue and robustness are uncertain. But scaling costs, model fragility and uncertain economics are forcing a reset. The same investors who once fought to secure GPU allocations are now asking:

What are we really buying?

The truth is that AI has entered the classic phase of any technological gold rush when promise outruns proof. Even as model sizes soar into the trillions of parameters, the performance curve is flattening. Each new leap in scale brings smaller gains and exponentially higher costs.

This is not a crisis of potential. It’s a crisis of proportion. We have built remarkable tools but we have overestimated their readiness for production and underestimated the complexity of delivering reliable value at scale.

Rational valuation in AI means pricing companies on what they can deliver, not what they can imagine. On verified outcomes, not speculative scale. It asks a simple question:

Can the economics match the excitement?

The next phase of AI’s evolution won’t be won by those shouting the loudest or training the biggest. It will belong to those who can deliver rational AI. Systems that are right-sized, cost-efficient and context-aware.

The next phase of AI progress will be defined by models that perform reliably in the wild, systems tuned to the imperfect, high-stakes realities of the businesses they serve.

We’re already seeing hints of this shift. Enterprise-focused players like Cohere are finding traction by delivering domain-tuned models that integrate cleanly with client data and infrastructure. They prove that scale alone doesn’t create value. Applicability does.

The market doesn’t need more inflated valuations. It needs measurable outcomes.

• What’s the ROI of each inference?
• Can enterprises audit, govern and trust their AI stack as much as they do their financial statements?

Rational valuation in AI starts with rational delivery. That means grounding performance metrics in actual business impact. It means recognising that the next frontier in AI isn’t cognitive imitation but value optimisation.

AI’s trillion-dollar moment might correct, and it should. But what remains after the froth will be healthier, more sustainable, and far more valuable.