The Road Ahead Looks Clear
Where scarcity moves when intelligence becomes abundant
For a long time, we’ve been watching the same thesis develop:
Intelligence is becoming abundant.
But until recently, much of the evidence was still indirect.
We’d put the thesis down, wait, and look for the subtle reinforcing signals that the world was beginning to catch up.
That has changed.
We’re now watching the consequences unfold in real time.
Ox Alpha.
Open-weight models.
Token economics collapsing.
Local models approaching frontier capability.
Inference efficiency outrunning compute growth.
Companies reducing proprietary-model spend.
Nvidia moving toward open models.
The enormous quantities of inference being given away.
The economics of neoclouds coming under pressure.
These aren’t isolated developments.
They are different manifestations of the same underlying transition.
Intelligence is losing its scarcity.
And when a scarce resource becomes abundant, value doesn’t disappear. It moves.
This is the part of the AI conversation I think is still being massively underestimated.
The question is no longer:
Who has the smartest model?
It increasingly becomes:
Who can turn abundant intelligence into a trusted decision?
Because when models become increasingly capable, cheaper, smaller, open and interchangeable, the model itself becomes less defensible.
The scarce layer moves upward.
It moves into:
→ Context — understanding what someone actually needs
→ Routing — choosing which intelligence to deploy
→ Specialisation — knowing which model is best for which task
→ Evaluation — knowing whether the answer is good
→ Trust — knowing which outputs to rely on
→ Distribution — being present at the moment of decision
→ Defaults — becoming the thing people don’t need to think about choosing
This is the emergence of the resolution layer.
The models generate possibilities.
The resolution layer decides what actually gets used.
And that changes the economics of AI.
The enormous capital being deployed today was largely justified by an assumption of continuing scarcity:
scarce compute,
scarce intelligence,
scarce frontier capability,
scarce access to the best models.
But technology is now attacking those scarcities simultaneously.
That doesn’t mean compute becomes irrelevant.
It means the economic value of compute has to be reconsidered when efficiency is improving faster than demand can absorb the resulting intelligence.
That’s the strange thing we’re watching now.
The constraint isn’t necessarily disappearing.
It’s being outrun.
And this is why Phase 2 matters.
Phase 1 was about making intelligence.
Phase 2 is about resolving intelligence.
The winners may not be the companies that produce the most intelligence.
They may be the systems that can reliably determine:
Which intelligence? For what purpose? For whom? With what evidence? And what should happen next?
That is a very different economy from the one the current capital stack was priced around.
And it’s the reason the road ahead increasingly looks clear.
We’re not building another intelligence engine.
We’re building for the layer above it — where abundant intelligence becomes trusted resolution.