The usual question is what AI will do to jobs. The harder question is what it will do to ownership.
That is where this piece begins. Not with chatbots, not with prestige demos, and not with another argument about whether office work vanishes on a fixed timetable. The deeper shift is that intelligence is getting cheaper to access while the systems that make it usable at scale are getting more expensive to build, finance, and control.
Using AI is not the same as owning AI.
| Position | What it means | Where value tends to settle |
|---|---|---|
| Using AI | Applying tools to work faster or cheaper | Productivity gains that are often competed away |
| Owning AI assets | Holding claims on chips, power, data centers, or workflow surfaces | More durable margins and compounding exposure |
| Controlling AI bottlenecks | Owning the layer others depend on | Pricing power, switching costs, and strategic advantage |
This is why How AI Will Transform Capital matters more than a simple jobs debate. It is also why the older distinction in The Foundations of Wealth still holds up. Income and ownership are not the same category, and they behave very differently once the operating environment starts to move.
The Popular Story Gets the Emphasis Wrong
There is real labor exposure here. Anthropic’s early usage research shows that AI is already touching a meaningful share of tasks in knowledge work, especially in computer and mathematics roles, with uneven spillover into writing, support, and administrative work in current usage data.
But labor disruption is only half the frame, and maybe not the most expensive half. A lot of commentary still assumes the main question is whether wages fall. That is too narrow. Wealth can concentrate even when wages do not collapse in a clean straight line.
In plain language, a smaller labor share does not automatically mean every pay packet drops at once. It means a larger portion of output can end up somewhere else. In practice, that usually means it moves toward the owners of capital, infrastructure, intellectual property, and operating systems.
That is the tension most surface-level AI writing misses. The visible product is the model. The less visible question is who owns the rails under it.
If AI becomes ordinary but ownership stays narrow, who actually compounds?
Where AI Value Is Concentrating Already
The funding pattern is already giving the shape of it away. According to the OECD, AI firms captured 61% of global venture capital in 2025, with infrastructure and hosting drawing a large share of that capital as the cycle intensified.
That does not read like a story where value is mostly settling in casual software usage. It reads like a story where value is pooling around physical bottlenecks and the systems that sit close to them.
Epoch AI estimates that global AI computing capacity has been growing at roughly 3.3x per year since 2022, which is close to doubling every seven months as chip production and deployment scale up. The same research puts AI data-center power capacity at about 30 GW by late 2025 as the power footprint keeps climbing.
Those are not abstract numbers. They point to buildings, transformers, cooling, financing, procurement, networking, and long supply chains. They point to Nvidia at the chip layer, to Amazon and Google at the cloud and data-center layer, and to a wider set of hosting and utility systems that rarely get treated as the starring part of the AI story even when they are carrying much of the economic weight.
- Chips set the ceiling for training and inference.
- Power turns models into live capacity.
- Data-center space converts demand into something deployable.
- Workflow control decides where recurring margin can stick.
| Scarce layer | Why it matters | Why value can stick there |
|---|---|---|
| Chips and advanced compute | They set the ceiling for training and inference | Supply concentration and technical difficulty limit substitution |
| Power and data centers | They turn model access into live capacity | They require huge capital, time, and coordination |
| Data pipelines and evaluation | They make systems usable and trustworthy | They create hard-to-copy operating memory |
| Workflow control | They sit where real work is reviewed, routed, and recorded | They build switching costs and durable dependence |
This is why the argument in The $212 Billion Bet: AI’s CapEx Gold Rush matters. It is also why Why Data Pipelines Are the New Oil Rigs of AI and The AI Economy’s Hidden Engine: Specialized Tools feel closer to the truth than another generic celebration of smarter models.
Cheap intelligence. Expensive delivery.
That line is doing more work than it first appears to. Plenty of people will get access to useful AI. That does not mean plenty of people will own the bottlenecks through which the gains are routed.
Why Wages and Wealth May Diverge
This is where the story gets messier, and better. The simple version says automation hurts labor and rewards capital. The more careful version is that wage effects, job effects, and wealth effects do not have to move together on the same schedule.
An IMF working paper on AI adoption and inequality makes exactly that point. Depending on how adoption happens, wage inequality can compress in some settings even as wealth inequality worsens, because the ownership layer keeps pulling ahead.
Restated more simply, workers can still benefit in some places while owners benefit more, for longer, and with less friction.
That is a different kind of pressure. It does not always announce itself through immediate unemployment. Sometimes it shows up through slower repricing. Skills still matter. Salaries still arrive. Teams can still look intact. But the part of the system that compounds most reliably starts drifting away from labor and toward capital-heavy control points.
Reading recommendation, if you want one frame that helps with this without turning it into ideology: Vaclav Smil, Energy and Civilization. AI looks digital from the front. It looks physical from underneath.
The practical question is not whether all cognitive work becomes worthless. It is whether more of the upside ends up attached to assets that fewer people can own directly.
The Ownership Question Beneath the AI Boom
The strongest version of this argument is not anti-technology and not anti-business. It is an observation about value capture. When a technology depends on scarce systems, gains tend to collect around whoever owns, finances, or controls those systems. That has been true in ports, rail, grids, and cloud. There is no reason to assume AI behaves differently once it leaves the demo phase and settles into infrastructure.
BlackRock’s 2026 annual letter put the point in unusually plain terms: the central question is not only whether AI raises productivity, but whether ownership of the gains broadens or stays narrow as the system matures.
That is the part of the reset worth watching. Not because everyone needs to become a chip designer or a data-center operator. Not because one company wins everything. But because the terms of participation are shifting. In some sectors the model itself will be the visible product. In others, the steadier margins may sit lower in the stack, or closer to the workflow, or inside the trust architecture around recurring decisions.
This is where The Three Waves of AI Wealth Creation remains useful. The first wave is often about efficiency. The later waves are usually about control.
No one needs a melodramatic theory of collapse to see the pattern. AI is not only changing what work gets done. It is changing which assets become more central, which dependencies harden, and which claims on future cash flow begin to look more valuable than the work that first drew attention.
The real reset is not that intelligence is becoming cheaper. It is that ownership of the systems around it may matter more than intelligence itself.
Does AI mainly threaten jobs or ownership structures?
Both matter, but the ownership side is easier to miss. Job exposure is visible and immediate. Ownership concentration is slower and often more durable because it determines who holds the claims on the systems that keep compounding. That is the distinction underneath How AI Will Transform Capital.
Why do compute and power matter so much in AI economics?
Model access alone does not create live capacity. Chips, power, networking, buildings, and cooling turn intelligence into a service that can run every day. That is where cost and scarcity start to bite, which is why the infrastructure framing in The $212 Billion Bet: AI’s CapEx Gold Rush matters.
Can wages hold up while wealth concentration still worsens?
Yes. Workers can still benefit in some areas while the larger and steadier gains go to owners of infrastructure, data flows, and operating surfaces. Wage outcomes and wealth outcomes do not have to move together on the same schedule, which is close to the older distinction in The Foundations of Wealth.
Is using AI enough to benefit from the AI economy?
Not necessarily. Using AI may improve productivity, but productivity and ownership are different. A user can get faster while the economic rent settles elsewhere, especially when workflow control and bottleneck ownership sit outside the visible tool. That is why The AI Economy’s Hidden Engine: Specialized Tools fits naturally here.
