The Third Layer of the AI Story: Value Capture
A reflection on how the AI economy is splitting between model builders and infrastructure enablers, and why long-term value will shift toward applications, agents, and business outcomes.
In my previous reflections, I explored two dimensions of the current AI wave.
First, the responsibility that comes with systems that act. Then, the unprecedented scale of infrastructure investment and its financial implications.
There is now a third layer that completes the picture.
The structure of value itself.
We are starting to see a clear split in the AI economy.
On one side, pure-play model companies are scaling rapidly. Revenue is growing at extraordinary speed, but so are losses, driven by compute, training, and inference.
On the other side, hyperscalers and enablers are generating strong profits. But they are also committing to massive capital expenditure, with projections of $650–700 billion in AI infrastructure in 2026 alone.
This creates a structural imbalance.
The companies building intelligence are not yet capturing sustainable value. The companies capturing value are now exposed to the largest infrastructure investment cycle in modern tech history.
This is where the narrative shifts.
Because the question is no longer who builds the most powerful model.
The question is who captures the value.
And the answer is already starting to emerge.
Value is moving up the stack.
From models To applications To agents To business outcomes
Models are beginning to commoditize. Costs remain high. And differentiation is increasingly defined by how AI is applied, integrated, and monetized.
This is where margins will exist.
This is where defensibility will be built.
And this is where the winners will emerge.
For pure-play AI companies, the path forward is clear. They must evolve beyond model providers into product and solution companies with sustainable economics.
For hyperscalers, the challenge is equally critical. They must demonstrate that this infrastructure scale translates into durable and profitable growth.
Both sides are now tied to the same constraint.
Monetization must catch up with investment.
If it does, this becomes one of the most important value creation cycles in modern history.
If it does not, the outcome will not be a failure of technology.
It will be a correction of expectations.
And history suggests those corrections tend to arrive faster than anticipated.
Where do you think the real value in AI will be captured over the next 3 years?
Comments