AI creates a familiar economic question in an unfamiliar form: when a key input becomes cheaper, who captures the difference?
The immediate answer seems obvious. If cognitive output becomes cheaper, every organisation should benefit. In practice, the gains are contested. They can accrue to infrastructure providers, model vendors, software platforms, firms that redesign workflows, professionals who control scarce complements, or customers whose prices fall.
The technology does not decide the distribution on its own.
A market shaped like a funnel
AI economics moves in two directions at once.
At the frontier, creating capability requires enormous capital: data centres, chips, energy, research talent, data, and long development cycles. This concentrates power among a small group of companies and states.
At delivery, models and model-like capabilities become cheaper, more accessible, and increasingly commoditised. What was frontier performance yesterday becomes a low-cost service, an embedded feature, or a local model tomorrow.
This creates a funnel:
Concentrated creation
→ broad distribution
→ competition over the scarce complements
The strategic question for most organisations is not whether they will own a frontier model. It is what scarce complement they can control around abundant model capability.
The scarce complements
Several complements remain valuable as generic generation becomes cheaper.
Trusted proprietary data
Not merely data volume, but data that is current, governed, meaningful, and connected to a decision context.
Workflow position
The ability to place AI inside a process where it can influence action rather than merely produce content.
Distribution and customer access
A model without users, channels, or embedded relationships captures little value.
Verification and accountability
In high-stakes environments, the party that can validate, insure, approve, or defend a decision often holds the economic power.
Physical execution
AI can make plans cheap; moving goods, repairing equipment, caring for people, and operating regulated infrastructure remain constrained by the physical world.
Institutional trust
Reputation, licensing, legal responsibility, and governance can become more valuable when plausible output is abundant.
Why model access alone is not a strategy
When many competitors have access to similar models, model access becomes a baseline capability rather than a differentiator.
The differentiator is how well an organisation combines AI with the assets others cannot easily copy:
- Its data model
- Its domain workflows
- Its relationship with customers
- Its ability to deploy safely
- Its ability to prove value
- Its ability to turn outputs into decisions and decisions into action
This is why generic chat interfaces are rarely enough. They create local productivity, but they do not automatically create a durable operating advantage.
A capture test
Ask five questions of any AI initiative:
- What cost or constraint is falling?
- What complementary resource becomes more valuable as it falls?
- Who controls that complement today?
- Can competitors acquire it easily?
- Does the workflow convert the advantage into a measurable customer or business outcome?
If an organisation cannot answer these questions, it may be adopting AI without a theory of value capture.
The AIxtract position
AIxtract operates in the layer where value becomes operational: connecting fragmented enterprise information, extracting structured signals, validating them against evidence and rules, and routing them into accountable workflows.
As generation becomes abundant, the premium shifts to trusted intelligence in context. The advantage belongs to the organisation that can answer not only “What did the model say?” but also:
What does it mean here?
What evidence supports it?
Who should act?
What happened next?
That is where the AI surplus becomes durable business value.
Adapted from the book Cheap Thinking: What AI Makes Abundant, What It Makes Scarce, and Who Captures the Difference by Daoyuan Li, PhD.