Who Gets Rich When Intelligence Is Free?
A map of the post-AI economy
That is the only AI question worth real money.
Everything else the industry argues about is a question of timing: which model tops which benchmark, whether this is a bubble, when the buildout corrects. Timing changes the price you pay. It does not determine where the value ultimately goes.
In June 2024, Leopold Aschenbrenner published Situational Awareness: The Decade Ahead. It mapped the industrial mobilization required to build machine intelligence: chips, power, data centers, transformers and capital. He subsequently launched an investment fund and translated the argument into a concentrated AI investment strategy, including substantial exposure to infrastructure companies.
Two years later, its central insight is no longer contrarian. The AI buildout is visible in every hyperscaler budget, clogged interconnection queue and multi-year transformer quote. Consensus is priced. Investors understand that intelligence needs electrons and steel.
Which means the live question has moved one step further:
When intelligence itself becomes cheap and ubiquitous, which companies, assets and entirely new categories capture the value?
That is what this publication exists to answer. This founding essay draws the map. The essays that follow will test it, territory by territory.
Where the value goes when intelligence gets cheap
When intelligence collapses in price as a factor of production, the value doesn’t stay in the intelligence layer. It moves to the complements intelligence consumes, to the assets it cannot reproduce, and to the products and markets that only become viable once cognition is nearly free.
Two ideas carry the argument: complements and possibility.
In his classic account of the productivity paradox, economic historian Paul David showed that electrification delivered its largest gains only after factories were reorganized around distributed electric motors. Merely replacing a central steam engine with a central electric motor changed far less. The real gains required redesigning the factory floor, production flow and surrounding infrastructure around the new technology. His paper, “The Dynamo and the Computer”, was published in 1990, but its lesson is current.
A cheap general-purpose technology does not create most of its value inside the technology itself. It creates value through everything reorganized around it.
Cheap cognition should follow the same pattern. The prize is not only in producing intelligence. It is in what intelligence runs on and cannot do without, and in every business that was uneconomic while it needed an expensive human mind on each unit, and suddenly isn’t.
That creates two maps:
Where existing value re-pools.
What becomes newly possible.
Both matter. Most of the AI conversation is standing on neither.
Why intelligence keeps getting cheaper
But why would one assume that intelligence gets cheap at all?
It is not a prediction. It is an extrapolation of curves that have been measured in public for years. Epoch AI’s current trend data estimates that training compute for frontier language models has grown roughly four to five times per year. Its research on algorithmic progress in language models found that the compute required to reach a fixed level of performance was halving roughly every eight months.
The two effects compound: more physical compute used more efficiently. The result appears in prices. When OpenAI launched GPT-4 in March 2023, its API cost $30 per million input tokens and $60 per million output tokens. Across fixed levels of capability, Epoch now estimates that inference prices have been falling between ninefold and nine-hundredfold per year, depending on the task and performance threshold being measured.
Capability is improving at the same time. METR’s research on the length of tasks AI agents can complete found a historical doubling time of approximately seven months. The exact estimate will move as evaluations improve, but the direction matters: systems are not only getting cheaper per unit. They are becoming useful across longer and more complicated units of work.
You don’t need a date for “true machine intelligence” to underwrite any of this. You only need the observation that curves this consistent, with this much capital behind them, don’t stop on a dime. Take the direction as given. The honest uncertainty is real, and it lives in the failure modes at the end of this essay.
Everyone is looking in the wrong place
The first default move is to own the intelligence itself: invest in the model layer or build an “AI app for X.”
But the model is the exact part racing toward free. Open-weight Chinese models such as MiniMax’s M2.5 were available on OpenRouter at $0.15/$0.90 per million input/output tokens, more than an order of magnitude cheaper than frontier models, while scoring only a few points below on SWE-Bench. Nor is the frontier gap durable. In February 2026, Anthropic said DeepSeek, Moonshot, and MiniMax had generated more than 16 million Claude exchanges to improve their own models. Epoch AI now estimates that the best open-weight models trail the closed frontier by only about four months. A frontier release quickly becomes training data for its cheaper successors:
This year’s frontier is next year’s free tier.
Owning the asset whose whole trajectory is getting cheaper is a strange plan for getting rich, and the app wrappers are worse: a thin skin over a capability their own supplier can swallow in a release note.
The second is to fight about the bubble. Is it one? When does it pop? That is the mainstream conversation, and it is the wrong fight. A bubble is a timing error. It changes what you pay to get in, not where the value goes. You can be dead right that the buildout is overheated and still miss the ten-year migration running underneath it.
The migration is already visible
Microsoft, Amazon, Alphabet and Meta currently plan a combined $695 billion to $725 billion of capital expenditure in 2026: Microsoft: ~$190 billion, Amazon: ~$200 billion, Alphabet: $180-$190 billion, Meta: $125-$145 billion. The physical suppliers are printing the demand. NVIDIA’s data-center revenue reached $75.2 billion in the quarter ending April 26, 2026, up 92% from a year earlier.
Cloud demand accelerated in the same reporting window: Azure grew 40%, Google Cloud grew 63%, AWS grew 28%. Microsoft said demand continued to exceed available capacity. Alphabet made the same constraint visible in its expanding cloud backlog.
Then the thesis turns physical. Large power transformers averaged 128-week lead times. More recent reporting found generator step-up transformer lead times exceeding 160 weeks. Sightline Climate is tracking approximately 16 gigawatts of data-center capacity intended to come online during 2026. Only about 5 gigawatts were under construction when it published its February assessment. The remaining projects faced the slower world of power procurement, permits, substations, equipment and construction.
Meanwhile, intelligence consumption continues to compound through the price decline. OpenRouter, which routes traffic across hundreds of models, reported that weekly volume rose from 5 trillion to 25 trillion tokens in six months. At that pace, the platform processes more than a quadrillion tokens a year.
Lay those facts beside one another:
The intelligence is becoming cheaper.
The amount consumed is growing.
The physical and institutional systems underneath it are becoming more valuable.
The market is already paying for electrons and steel while everyone argues about which model wins.
The layer-two effect: value moves to the complement's complement
Value doesn’t migrate once; it migrates in layers. Everyone now agrees that cheap intelligence needs power, so power attracts capital. But supplying power requires its own constrained inputs: transformers, switchgear, cooling systems, transmission equipment, turbines and land with a credible route to interconnection.
The same pattern repeats elsewhere. Robots need intelligence, but they also need precision gears, motors, sensors and reliable actuation. Autonomous transactions need agents, but they also need identity, payment rails, permissions and proof. Continuous AI services need models, but they also need trusted data and systems of record.
The obvious complement fills first. The deeper mispricing often sits in the complement’s complement. “Buy the picks and shovels” is now consensus. The edge is the pick that makes the shovels.
Map 1: six zones where existing value concentrates
Six zones where value pools as cognition gets cheap. A few lines on each here; the essays that follow go deep, zone by zone, wherever the evidence is freshest.
Energy and the physical substrate. Power generation, the grid, cooling, land that comes with power. Intelligence is manufactured from electrons, and every capability gain adds load. This is the binding constraint of the era; the queue numbers above are its receipts, and they are why the map leads here.
Actuation. Precision motion, sensors, machine vision, the robot supply chain. A mind without a body cannot move matter, and the gears don’t rot the way chips do. The market is busy pricing the robot’s brain, which depreciates like the silicon it is. The durable money is in the body.
Transaction rails. Payment networks, exchanges and clearing, licensed data: the tolls that volume must cross. Turn a billion agents loose and they multiply transactions through rails they cannot rebuild.
Trust and verification. Identity, security, provenance, proof that an autonomous action was real. Cheap intelligence makes fakes and attacks cheap too, so proving who did what stops being optional spend.
The human premium. Status, attention, live experience, health and longevity, prime physical space. When the machine does the thinking, the freed surplus flows to what a machine cannot be, and cannot fake.
Hard money. The least direct zone: not a complement to cognition, but an expression of the monetary regime financing the buildout. If infrastructure expansion increasingly depends on fiscal support and balance-sheet growth, scarce monetary assets become more valuable. If it does not, this zone weakens.
The anti-map: do not own the donors
The six zones sharpen when placed beside their inverse. A whole class of business exists whose profit is the white-collar cost that cheap cognition is eating:
Per-seat software without proprietary data or workflow control
Outsourced back-office operations
Agencies and consultancies billing primarily by the hour
Businesses whose scale depends on adding people in proportion to revenue
These companies may initially appear cheap. Their historical earnings may look stable. Their customers may be slow to change. But when their economic function is the cost being removed, a low multiple does not make them protected. Cheap because it is melting is not a bargain.
Map 2: five categories cheap cognition makes possible
The first map tracks where existing value re-pools. The second tracks what gets created: everything that becomes viable because cognition stopped being the expensive part. Five mechanisms, each with a one-question test.
Cost-collapse services. Anything that needed a costly expert per unit goes continuous and cheap: the personal CFO, legal triage, health navigation. Was this a €200-an-hour human?
Long-tail markets. Segments too small, too custom or too slow to serve profitably become serviceable: micro-vertical software, the one-person enterprise. Was the TAM “too small”?
Continuous systems. Work done annually because labor was expensive goes always-on: audit becomes continuous assurance, market research becomes standing intelligence. Was the frequency set by labor cost?
Coordination-unlocked businesses. Value blocked by coordination overhead rather than cognition: multi-party logistics, complex procurement. Did this die in email threads?
Abundance-native primitives. No pre-AI analog at all: agent-to-agent markets, machine-speed negotiation layers. Highest uncertainty, highest ceiling.
One corollary stitches the two maps into a single thesis: every new autonomous category drags a verification demand behind it. The more agents transact, the less optional it becomes to prove who did what. The second map feeds the fourth zone.
I can’t hand you next decade’s winners by name here. Most are private, too early, or don’t exist yet, which is exactly why this is the frontier and not the trade. But a map that only shows what to defend, never what’s being built, is half a map. The next companies at today’s giants’ scale will come at least as much from this side.
How value moves: buildout, digestion, diffusion
None of this lands at once, and the order matters more than any date. Three phases, spanning roughly 2025 to 2035. Markers govern, not the calendar; each phase announces itself in observable facts.
Buildout, roughly 2025 to 2027, and where we are as I write. The physical substrate is the bottleneck, and the money is with whoever supplies the scarce infrastructure. The marker: lead times quoted in years and interconnection queues that don’t clear. It is a real mobilization, visible on balance sheets: Amazon’s trailing free cash flow fell from $25.9 billion to $1.2 billion as AI investment drove a $59.3 billion increase in equipment spending; Alphabet raised $84.75 billion specifically to expand AI infrastructure and compute.
Digestion, roughly 2027 to 2030. The bill comes due, and the market demands the return show up in audited earnings rather than keynote slides. Monetization is accelerating, but the relationship between AI revenue and infrastructure spending remains unresolved: OpenAI and Anthropic together were reportedly running at roughly $55 billion in annualized revenue by April 2026, while the four largest hyperscalers were planning several hundred billion dollars of annual capital expenditure. The figures are not directly comparable, and that is the point. Digestion is where the market begins demanding a credible bridge between them. The marker: capex growth rolling over while adopter margins are asked for in audited prints. It will not be comfortable. It is a phase to pass through, not an end state.
Diffusion, roughly 2030 to 2035. Rails, trust and the complements compound; the migration completes; the second map becomes ownable at scale. The marker: agents transacting at scale in reported financials, and AI gains visible in ordinary companies’ margins.
Notice what the map is not claiming: a date when machines get smart. Take the direction as given; that is what the trendlines above are for. Argue instead about what is actually contested:
The speed of the transition
The bottlenecks along the route
The durability of commercial moats
Who captures the surplus while it moves
And a clock runs through both maps. The complements are ownable today, mostly in public markets. The new categories mostly are not yet; they come within reach as the phases turn. When each category flips from merely buildable to actually ownable deserves its own essay, and it will get one. Without the sequence, the map is a menu. With it, it’s a route.
What would make this wrong
This thesis only works if a chain of assumptions holds. The load-bearing assumption comes first:
Defensibility has to survive somewhere above cheap cognition.
The whole map assumes value has somewhere to pool. If intelligence becomes cheap and dissolves every moat it touches, then value has no destination. It leaks to customers as consumer surplus. Margins flatten. The map comes back blank.
Call that world the Great Leak. It could look like utopia from the consumer side and economic rupture from the capital side. I do not think it is the most likely world. Physics, land, licenses, networks and trust do not commoditize on the same curve as software. But the possibility cannot be dismissed, because the entire structure depends on it being wrong.
The other failure points follow.
The capability can’t plateau. If agents stay impressive demos that can’t run unsupervised past narrow, checkable tasks, the demand paying for the buildout breaks, and the crash takes the near-term names with it. The most bearish branch, and a real one.
The intelligence providers have to stay in their lane. The map assumes cognition stays cheap and comes from many suppliers. If a handful of labs integrate downstream and capture the complements themselves (the rails, the power, the trust layer), the map redraws around them.
Cheap has to mean cheap, not rationed. Prices must keep falling and power and chips must scale with demand. If cognition ends up abundant in theory but expensive and rationed in practice, “abundance” quietly becomes “premium.”
Adoption can’t stall, and someone has to be able to pay. Diffusion assumes firms actually reorganize and buyers still have money. Paul David’s factories are the warning: the gains waited on reorganization, not on invention. A long adoption lag, or a demand shock, leaves the surplus real on paper with no one left to spend it.
Purchasing power breaks. The second map assumes there are customers able to buy the new abundance. A technology can create enormous productive surplus while simultaneously disrupting the incomes required to purchase its output. If gains concentrate too narrowly, demand may become the constraint.
And beneath all of that sit two conditions that aren’t investment risks at all; they are the ground the map stands on:
The intelligence has to stay controlled and aligned. Everything above assumes AI remains a tool that does what it is pointed at. Lose that, and the question stops being where capital migrates, because there won’t be a stable enough world to ask it in.
The political order has to hold. This transition displaces a lot of white-collar work, fast. It reaches diffusion only if societies absorb the shock without the rupture that smashes data centers, freezes capital markets, or tears up the property rights the map depends on.
Note the uncomfortable symmetry: a safe, aligned, socially stable world with abundant intelligence can still falsify this thesis, if the surplus leaks, if adoption drags, if the providers eat the complements. The catastrophe-avoidance conditions are necessary, not sufficient. The map is worth drawing only in the worlds where we get to keep drawing.
How the map gets tested
This is version one of the map. It will change as the evidence changes.
The essays that follow will develop it in three directions:
deep dives into the zones where value concentrates,
essays on the mechanisms that create new categories,
scenario pieces that examine how the thesis could fail.
The sequence will follow the strongest evidence, not a fixed curriculum. Once a quarter, a watchlist will record which scenarios have become more or less plausible, and why. When the map is wrong, the watchlist is where it will say so first.
One disclosure matters: I have skin in the game and invest behind this framework. That is not proof the map is right. It is why the framework exists—and why I want it challenged.
So return to the opening question: where will the trillion-dollar companies of the next decade come from? They will not win simply by owning the smartest model. They will own the scarce systems the models depend on, control the assets the models cannot reproduce, or build the products and markets that become viable only when cognition is cheap.
Intelligence will become abundant. The largest fortunes will be built around what it makes scarce — and what it makes possible.
Subscribe to follow the map as the evidence changes.