Originally published on Progress and Poverty on August 10, 2026. Republished on Progress.org with permission.
Individual ballpoint pens famously retailed for the equivalent of $230.65 in today’s money when they were first introduced in 1945, but eighty-six years later, they sell for as little as eleven cents apiece, a more than 2,000x reduction in price. There is no “ballpoint pen” monopoly, and therefore no “pen windfall” for anyone to fight over, because endless competition has driven the price down to just above the bare cost of producing, marketing, and shipping them. The principal beneficiary is the consumer, who gets the full value of ballpoint pens for a very low price.
In contrast, consider petroleum. Although the global price of oil rises and falls, it consistently trades at a large premium well abovethe bare cost of the necessary labor and capital to pull it out of the ground and bring it to market. This is because, unlike pens, you can’t just make oil out of cheap and readily available materials, and you also can’t produce it anywhere you like: you need scarce natural resources, situated in specific physical locations. Those resources and locations are owned by someone, and you need to pay that someone for the right to participate in the oil business at all. The premium that someone is able to command for resource access is called a resource rent. This rent arises from fundamental scarcity, and it’s a key component of what Norwegian researchers Jonathon Moses and Anne Margrethe Brigham call “The Natural Dividend” in their book by the same name.
Progress and Poverty is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.
Artificial Intelligence is on everyone’s minds these days, and it seems clear that somebody, somewhere, is going to make a lot of money from it. But who makes that money–and how–is what will ultimately determine what the proper policy response (if any) should be. If AI turns out to be like petroleum, with large and durable monopoly rents accruing to gatekeepers, it makes sense to hit it with Norwegian-style regulations and taxes. However, if AI turns out to be like pens, policies like nationalizing the frontier labs would not only be nonsensical, they might even ironically amount to a bailout of AI investors and CEOs.
If AI is like petroleum, then someone should be able to capture a large and durable windfall by gatekeeping a durable chokepoint in the AI production chain. Let’s consider a few candidates:
Frontier labs
Hardware companies
Natural resource owners
Frontier Labs
Under what conditions can frontier labs like OpenAI and Anthropic capture an AI windfall? Imagine for a moment that we’re back in 2016, and we’re still debating whether AI will ever be able to beat the world’s best human in the board game Go, pass the Turing Test, etc. While we’re still arguing, a time portal opens and out plops Anthropic CEO Dario Amodei along with a few dozen 2026-vintage data centers running Claude. Dario shows off modern LLM capabilities to an absolutely stunned audience who are still a year away from reading Attention is All You Need (the seminal research paper that paved the way for LLM’s), let alone experiencing ChatGPT, Stable Diffusion, and the other harbingers of the modern AI revolution. Finally, Dario unveils Claude Code, and people lose their minds and their wallets.
Ten years ago, with today’s capabilities and no competition in sight, customers (and large companies especially), would be willing to pay very large sums for Claude subscriptions. Time travelling 2016 Anthropic could therefore capture the majority of the value chain because there would simply be nothing else like it on the market, and the competition would be a literal decade behind.
Fast forward to today, and the competitive landscape looks quite different. Today the question that drives willingness to pay is not: “how much total value does Anthropic produce for me?”, but “what has Anthropic done for me lately?” And it’s not just Anthropic. Today, there’s at least three companies widely recognized as keeping up the American AI frontier–Anthropic, OpenAI, and Google, and just as importantly, a storm of competitors nipping at their heels. Even worse for the leading labs, many of those competitors are giving their models away in the form of open weights, particularly the Chinese ones. This constrains the incumbents’ pricing power, since they can only charge for the marginal benefit they provide over the next-best choice.
Nevertheless, all of this is simply what’s happening today, and nobody knows for sure where things will stabilize. The AI bull thesis is that all the frontier labs are racing to get to the end of the rainbow, where a leprechaun named “AGI” will reward the first person across the finish line with a bottomless pot of gold.
This thesis depends on a big assumption: that a temporary lead will eventually be convertible into a permanent one. If that turns out to be false, “spend whatever it takes to reach the end of the rainbow first” becomes a dubious strategy. That’s because even in a world where almost all human labor is automated, if you can’t stop others from copying you, then everyone eventually gets the same advantage you have. You lose any special leverage you had over the economy and are left with a bunch of magic ballpoint pens, the same kind everyone else has, whose prices have fallen towards the bare cost of reproduction.
Hardware Companies
Now, what about hardware companies? AI is software, and depends on physical computers for both training and inference. Those computers are designed and manufactured by hardware companies. Will these companies be able to bottleneck the AI supply chain and secure windfall profits for themselves?
Maybe, and the best evidence for this is the fat stacks of cash money dollars already piling up in their bank accounts. While Anthropic brags about becoming “operationally profitable” in the near future, hardware companies are getting rich today.
The biggest players are companies like NVIDIA, who designs the chips, as well as SK Hynix and ASML, who make the machines that make the machines that make the chips. Downstream, there’s also all the companies that make less fancy components like good old fashioned RAM and SSD’s, which are now in such high demand that normal consumer electronics have all shot up in price. A good example is Valve’s new Steam Machine, which was originally supposed to sell for around $750, which now lists for $1050 to make up for surging memory prices.
There’s a credible case that in the short to medium term, we’ll see a real “AI windfall” accrue to these hardware companies while demand continues to outstrips supply. That’s because these hardware companies aren’t trivially copyable; you can’t just spin up a rival org, hire a bunch of people who seem smart, throw lots of money at them, and get the same results right away.
A key limitation for competitors is all the “tacit knowledge” embedded in the procedures, processes, and workflows that exists only in the minds of the extremely skilled people who work at these companies. Even when a competitor manages to crack that vault, they still have to re-assemble all the supply chains and relationships which are just as much a part of the business as the factory specs and operations themselves. For these reasons, hardware companies have the potential for a strong and durable moat.
That said, “durable” isn’t the same thing as “perpetual.” Machines wear out and have to be replaced, and luxurious margins are a giant blaring signal screaming “compete with me.” Absent artificial restrictions, capital will flow into this sector to do just that. The only question is how long it will take. For this reason, hardware (in the long run) looks more like pens than petroleum.
Natural Resource Owners
All the hardware AI runs on is made of atoms, and the electricity it consumes is made of energy. Atoms and energy have to come from something, and that something is natural resources.
On the other hand, data centers absolutely do use a lot of energy, which is increasing over time and can have a direct impact on what regular people pay for electricity. This is why instead of handing new sites 100% property tax abatements, localities should negotiate with data centers and insist they pay market rates for everything they consume.
If growth continues to accelerate, we expect to see a growing “resource rent” accrue to those who control the resources AI uses which are fundamentally scarce in supply. Interestingly enough, even though this scenario is the one that’s most like petroleum and least like pens, even here the gains flow not to AI itself but what AI consumes and depends on; whoever controls the rights to those resources will capture the greatest windfall.
Speaking of control, we must also be careful that in (rightfully) regulating these resources, we don’t set ourselves up for regulatory capture. Moses and Brigham point out in their book that when the government regulates access to resources, it creates an additional “regulatory rent.” Since regulation limits competition, savvy incumbents adept at navigating those regulations can charge higher prices. This happens even when the regulations are objectively necessary and good, and is not an argument against regulation–just a harsh reality Moses and Brigham urge us to be aware of. The researchers argue that whenever a state creates regulatory rent, it should also take care to recapture that unearned monopoly windfall that would otherwise fall into private hands.
Should we see both resource rent and regulatory rent arise from AI’s increasing demand on our natural resources, Moses and Brigham give us a straightforward policy framework to follow (for a full treatment of this subject, see our detailed review of Moses & Brigham’s book):
Identify scarce natural resources as the rightful property of the people in common
Enact value capture policies such as severance taxes to capture the unearned windfall
Supplement this with proper incentives for discovery, exploration, and risk-taking
As a final note, we should consider that resource scarcity can be broken under the right circumstances. This is happening right now to diamonds in Botswana, which have long funded a Norwegian-style sovereign wealth fund in that country, only to now face competition from cheap lab-grown synthetics from China.
Nobody
If frontier labs descend into bloody price wars, and hardware companies get commoditized, and natural resources turn out to be sufficiently abundant, then we are left with a potential scenario where there simply is no “AI windfall.” Highly leveraged late-coming investors will undoubtedly go broke, but AI will continue to exist and there will still be (marginal) money to be made in continuing to develop it.
In this world, consumers themselves capture the windfall, in the form of “consumer surplus.” The cool new gadgets are just as capable as ever, it’s just that the whole chain of producers are unable to fend off competition. In many ways this is a best-case scenario, because everybody gets the benefits of AI development, without any single group holding durable monopoly power over the public’s head for decades on end.
Sounds great, right? There’s just one wrinkle: there’s one more group we haven’t considered, and even in a maximalist “consumer surplus” scenario, they might be the ones who wind up seizing the “AI windfall” by default.
Landowners
If neither frontier labs, nor hardware companies, nor natural resource barons are able to capture the bulk of the gains from AI advancements, then we are left with just another general purpose technology that increases human productivity for everyone. And general purpose technology that increases human productivity for everyone has a well-established tendency for its gains to get sucked into the price of land.
The best evidence for this is that it’s already happening. In San Francisco and Seoul, South Korea, the two epicenters of the AI boom, real estate values are already shooting up. Real estate values in both cities were already astronomical, but they’ve spiked again in response to recent stock market surges and IPOs in AI-related industries. All that cash burning holes in newly minted AI millionaires’ pockets is superheating the price of land in the urban centers where social pressure and urban clustering effects insists everyone “has to be” to get in on the AI craze.
In this scenario the AI premium flows not to pens, nor petroleum, but to property prices.
It’s not yet clear which of these worlds we’re going to land in, which is why pursuing an aggressive policy that confidently assumes an “AI windfall” will land with any particular sector is still premature. For instance, nationalizing the frontier labs, as both Bernie Sanders and Donald Trump seem to support, could backfire in one of two major ways. First, it could amount to a bailout of investors and billionaire CEOs whose businesses were already on the cusp of being commoditized anyways. Second, it may create the very sort of monopoly it seeks to avoid by anointing and entrenching well-resourced companies at the same time it constrains or even outright bans the competition.
Given this uncertainty, we should strongly consider policies that remain on solid footing regardless of which scenario we find ourselves in.
Tax Resources, Regulation, and Land
Any AI windfall will tend to fall on whatever remains scarce: natural resources, legal monopolies created by regulation itself, and land.
In the case of natural resources, we can straightforwardly follow the battle-tested Norwegian model that has been deployed for over a century, first in hydropower, and then in petroleum. As for regulatory rent, the most important thing is for lawmakers to realize that even when regulations are good, they create monopoly privileges, which should not be given away for free. The details for handling both of these cases are found in Moses & Brigham’s book.
As for land, it is something that everybody needs but which we cannot make any more of, and the AI powered economy will only accelerate the bidding war for scarce urban land in the most valuable cities, accelerating the ongoing housing crisis and worsening inequality. The solution, as always, is Land Value Return, our umbrella term for the many policies (including Land Value Tax) that return the value of land to the people.
Progress and Poverty is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.
