The Intelligence Flywheel Repricing Everything
The consequences of AI can be summed up as follows:
“Every race after AI will be run with AI.”
That sounds like a geopolitical argument.
It’s also an economic map.
If artificial intelligence becomes an input into every future technology race, then AI does not create one new industry beside the old ones. It changes the rate at which every industry can change itself.
Energy companies gain artificial engineers. Drug companies gain artificial scientists. Manufacturers gain artificial process designers. Software companies gain artificial labor. Investors gain systems that can read markets, test assumptions, and monitor portfolios continuously. Entrepreneurs gain execution capacity that once required an organization.
The economy does not simply get a new tool.
It gets a second learning rate.
The first learning rate is biological. People study, experiment, make mistakes, pass knowledge to one another, and improve the system over time.
The second learning rate is machine-based. Models can copy what they learn, run in parallel, work without sleep, search huge possibility spaces, and feed each result into the next attempt. When these systems connect to software, instruments, robots, factories, and markets, the gap between the two learning rates begins to widen.
That gap will reprice nearly everything.
It will reprice labor because cognition is becoming cheaper.
It will reprice infrastructure because energy and compute become more valuable.
It will reprice companies because small teams can command much larger output.
It will reprice capital because the organizations that learn faster can compound faster.
And it will reprice ownership because the returns will flow toward whoever owns the machines, the bottlenecks, the data, and the systems that convert intelligence into cash flow.
That is the economic meaning of the last race.
If you haven’t read the original piece, pop over to our sister publication:
Intelligence Becomes a General Input
The old technology map divided the economy into sectors.
Software sat over here.
Energy sat over here.
Healthcare, manufacturing, finance, logistics, defense, and media each had their own economics, talent pools, investment cycles, and competitive structures.
AI crosses those boundaries because intelligence is already inside every business.
Every company must forecast demand. Every factory must schedule work. Every laboratory must choose experiments. Every bank must price risk. Every retailer must allocate inventory. Every founder must decide what to build, whom to hire, where to spend, and when to change direction.
Those are intelligence problems and AI lowers the cost of solving them.
The first-order effect is efficiency. A company completes the same work with fewer human hours. The second-order effect is expansion. Once analysis, code, design, research, and coordination become cheaper, the company does more of them.
It studies every customer instead of a sample. It tests ten products instead of one. It personalizes the service. It enters smaller markets. It maintains software no one could justify maintaining. It investigates operational failures that were previously tolerated because no employee had time to own them.
Cost reduction gets the first earnings call.
Demand creation gets the next decade.
This distinction matters because investors often treat AI as a margin story. It will be one. But the larger opportunity comes from economic activity that did not exist because cognition was too expensive to apply.
The long tail becomes a market.
Cognitive Deflation Meets Physical Scarcity
The cost of machine intelligence is falling while the demand for it is rising.
That creates an apparent contradiction: if models and inference keep getting cheaper, why are companies preparing to spend so much on chips, data centers, power, and networking?
Because cheaper inputs create more consumption.
When bandwidth became cheaper, the world did not transfer the same files for less money. It created streaming video, cloud software, social media, video calls, and entire industries built around continuous data. Efficiency expanded the market.
AI follows the same pattern.
Cheaper tokens make longer workflows economical. Agents can read more context, run more tools, test more paths, supervise other agents, and work in the background. A single answer may get cheaper while the total amount of cognition applied to the problem rises by a hundredfold.
The digital layer enters deflation.
The physical layer feels the pressure.
More intelligence requires more compute. More compute requires chips, memory, networking, cooling, buildings, transformers, transmission, and power. When AI generates thousands of promising drug candidates, laboratory capacity becomes scarce. When it creates new hardware designs every hour, fabrication and testing become scarce. When it improves robots, the supply of motors, batteries, sensors, and actuators becomes more important.
Scarcity does not disappear.
It migrates.
This is the central economic move. Value shifts away from tasks that intelligence can reproduce cheaply and toward the assets, rights, relationships, and physical systems required to turn that intelligence into reality.
The model can design a factory.
It cannot wish the factory into existence.
The Return of the Physical Economy
The internet era trained investors to love businesses that scaled without touching the physical world.
Software had low marginal costs. Distribution was global. Capital expenditure was someone else’s problem. The best companies could add customers faster than they added assets or employees.
AI keeps those economics in parts of the application layer. At the same time, it creates one of the largest physical buildouts in history.
The intelligence economy needs power plants, semiconductor fabs, data centers, transmission lines, batteries, cooling systems, robotic factories, and automated laboratories. It needs copper, steel, concrete, uranium, gas, specialized chemicals, optical equipment, and skilled trades.
The cloud is becoming an industrial economy with a software interface.
This does not mean every industrial asset suddenly becomes a good investment. Capital-heavy industries can destroy enormous value through overruns, commodity pricing, leverage, weak contracts, and political risk. Demand can be real while equity returns remain poor.
But the direction of capital is clear.
The AI boom pulls money upstream. It raises the strategic value of reliable energy, secure compute, domestic manufacturing, scientific instrumentation, and the bridge between digital intelligence and physical execution.
The biggest winners may not look like AI companies.
They may look like the companies that let everyone else run AI.
The Firm Becomes a Machine for Allocating Intelligence
For most of the modern economy, output scaled with headcount.
More customers required more support workers. More software required more engineers. More research required more analysts. More operational complexity required more managers to coordinate the people managing the complexity.
AI weakens that relationship.
A company can now add cognitive capacity without adding an employee for every new unit of work. Agents can research, draft, code, test, reconcile, monitor, and coordinate. They do not eliminate the need for people. They change what the people are there to do.
The human moves toward direction, judgment, trust, taste, relationships, and accountability.
The machine absorbs more execution.
This creates a new form of operating leverage. Revenue can grow faster than payroll because the company is scaling through tokens, software, and systems. Small teams can attack markets that once required large organizations. Existing companies can produce more with the same people.
But the gains will not arrive automatically.
Buying model access is not the same as building a productive system. The model needs context, memory, tools, permissions, workflows, evaluation, clean data, and escalation paths. Without that harness, cheap intelligence creates cheap output and expensive repair work.
The company of the future is not the company with the most AI subscriptions. It’s the company that converts machine cognition into accepted outcomes the most effectively.
Margins Will Separate
AI will not raise every company’s margins.
It may destroy many of them.
If every competitor can produce similar software, content, analysis, or service at a much lower cost, the savings do not necessarily stay with the producer. Competition can pass them directly to the customer through lower prices.
This is the trap inside the easy AI thesis.
Cheaper production is not a moat.
If the output becomes abundant and interchangeable, margins compress. The value moves somewhere harder to copy: distribution, proprietary context, brand, trust, regulatory rights, physical access, embedded workflow, unique data, customer relationships, or ownership of a real bottleneck.
AI expands the difference between a product and a position.
A product can be copied.
A strong position controls the path between demand and a scarce outcome.
The companies that win will use AI to strengthen that position. They will learn from more interactions, improve the workflow faster, deepen integration, reduce cycle time, and build a private record of what works.
Every completed job makes the system better at completing the next one.
That is the company-level flywheel.
