What Kills Cartels?

Plus! Diff Jobs; Strategic Investments; Media Form Factors; Fees; Critical Minerals; Brand Names

Byrne Hobart Byrne Hobart

In this issue:

audio-thumbnail
The Diff February 2nd 2026
0:00
/944.53551

Chat with this post on Read.Haus.

What Kills Cartels?

One dog-that-didn't-bark question worth asking from time to time is: how has encrypted chat failed to lead to every major economy being dominated by cartels? The economic logic of a cartel is pretty straightforward: get together a critical mass of industry participants, ideally all of them; set output where a monopolist would; divide up the profits. Not only is there plenty of profit to go around in this scenario, but the people involved get a nice free-time dividend from spending a lot less of their time trying to predict and short-circuit whatever their competitors are up to.

Cartels are illegal, of course, so that's one constraint. But they haven't gotten more illegal since it got easier to run them. And many technologies besides Signal and the like have also made it easier for companies to monitor each others' behavior: ERP systems mean that companies can do more accurate accounting, vertical software means real-time algorithmic pricing engines and benchmarking tools, there are third-party data sources that track the flow of goods within supply chains, and there are plenty of general-purpose alternative data sources that can make it very clear what every company in a given industry is up to. Not to mention that many companies are publicly-traded, so they're consistently sharing information on revenue and margins, and telling investors about their future plans. In fact, some of these information technology tools are so good that you can join an alleged cartel by accident, simply by using a yield management service that tells you to raise prices.

Cartels need a few other ingredients besides coordination mechanisms. For one thing, they rely on the relative traits of industry fragmentation and high barriers to entry. It would be very hard to form a tech/finance newsletter cartel, for example: if everyone raises their prices, that drags some talent out of industry and onto Substack. And since some newsletters are pseudonymous, there isn't much of a barrier to someone retiring under one name and reemerging under another.[1] Housing and hotels aren't a very concentrated market, but they're also a market with a wide skill gap between the best and worst operators, and it takes a while for new supply to show up.[2]

So cartels work well in cases where there's a high fixed cost, especially if there are returns to scale. Some of those returns will come from operations: you generally get a lower cost for a given amount of steel or refined oil products if they're produced at fewer, bigger sites. Whereas a business that achieves maximum benefits from scale when it has half a dozen employees is always going to face the threat of new entrants. It also helps if they have economies of scale with respect to their relationship with the government: before the Second World War, both Germany and Japan had fairly cartelized economies, at least for heavy industry. In Germany's case, this was an old model of state-business cooperation, where both sides found it helpful to negotiate with the other, and where businesses that benefited from government patronage could be expected to do favors to the state from time to time. In Japan, this process was more state-driven; in 1931, their government passed a law that allowed the government to compel non-cartel members to adhere to the cartel's rules. From the perspective of businesses, these state-approved cartels are just another kind of operating leverage: once you've structured your business around margins that can only be sustained with the government's help, it's basically in charge.

But cartels don't just show up in developmental states that treat them as one more part of the economic growth toolkit. OPEC is a fun case study here, because it's an early example of international nationalism—the oil-producing assets that underlie it were mostly created by Western oil companies, and later nationalized (those specific assets have mostly been depreciated to zero, but US and European companies stuck around for a while as contractors).

Individual nationalist movements found that they could seize control of oil assets, but that they'd often face sanctions and other limitations. Any given nationalization was mostly articulated based on nationalist interests, but there was also a group interest in putting together a cartel of oil-extracting countries that could coordinate their behavior. (They had a pretty solid moral case for this given that a generation earlier, the biggest oil companies had agreed to a similar cartel of oil buyers.)

OPEC still matters on the margin, but it's been a long time since they were among the most significant unilateral economic actors. When they made headlines, it often revolves around things like bribery ($, FT) instead of their former ability to induce a global inflationary recession. Which illustrates why cartels are so hard to sustain: they're basically a bright flashing light informing anyone and everyone that there's some easy money sloshing around. This can be accessed by:

A monopoly seems like it should be something completely different, because the intra-organizational communication problem is solved. But, as was once pointed out by someone who illustrates the difference between politicians and statesmen, corporations are people. Specifically, they're made of people who have asymmetric information and differing incentives. Within the management of a monopoly, you can imagine two basic views of that monopoly's durability: one side says that they're always one mistake away from being in a competitive business once again, and that they need to constantly reinvest the proceeds of their monopoly into expanding their competitive moat, often by providing consumers with either a better product or with some subsidized complement to the core business. The other cohort says: nah, that's not gonna happen, and then goes on vacation. In the short term, the second category produces much higher returns on investment, and also more employee satisfaction and very happy shareholders. But it tends to get blindsided. It's easy to think of examples of this in tech, especially because young industries are prone to monopolies (the first web directory, commercial browser, search engine, social network, video-sharing platform, e-commerce site, etc. had a monopoly by definition, and there's a correlation between seeing that some business model might be viable and foreseeing what could threaten it in the future.

In practice, you can think of any existing monopoly as a particular opaque cartel. Instead of a group of CEOs getting together in some conference room to carve up their market share, it's a group of EVPs in a conference room or Zoom room trying to strike a balance between a) maximizing immediate profits, b) maximizing the defensibility of those profits, and c) avoiding the DOJ, FTC, and NYT, the triumvirate of three-letter entities that can ruin a monopolist's day.

In the end, monopolists are a little bit better at coordinating this than cartels are, though improvements in information technology mean that the gap is shrinking. But if a monopolist behaves in a way that consistently improves their competitive position, it ends up looking like the behavior of a company that does face competition. To build a monopoly is to see that it was worth building before anyone else seizes the opportunity, and that correlates with being able to see what could threaten that status. Keynes once wrote of excess savings that:

[W]e are more concerned with the remote future results of our actions than with their own quality or their immediate effects on our own environment. The "purposive" man is always trying to secure a spurious and delusive immortality for his acts by pushing his interest in them forward into time. He does not love his cat, but his cat's kittens; nor, in truth, the kittens, but only the kittens' kittens, and so on forward forever to the end of cat-dom. For him jam is not jam unless it is a case of jam to-morrow and never jam to-day. Thus by pushing his jam always forward into the future, he strives to secure for his act of boiling it an immortality.[5]

The most durable monopolies do that, but with investment instead; they're always laying the groundwork for a moment when they could maximally exploit their customers and suppliers, but in practice they tend to only gradually and noisily exploit them, with plenty of backtracking and diversions along the way. They're still a problem when that distant future arrives, but typically a temporary one—running a monopoly like a theoretical monopolist would is the lazy, obvious choice, and it doesn't work very long.

  1. Though if you have much of a corpus online, there's a decent chance that if you paste something you wrote but never published into a temporary chat session with an LLM, and ask if who probably wrote it, they'll identify you or at least put you on the shortlist. Some of them have gotten pickier about this kind of thing, but the capabilities of the do-whatever-you-want models tend to catch up to the don't-do-that models pretty quickly. ↩︎

  2. One particular algorithmic collusion case, against Atlantic City hotels, was pretty weak ($, Diff) for another reason: casinos make their money from the bundle of gambling-plus-amenities, so tracking pricing for just one amenity means missing the big picture. Sometimes, the reason to have a high sticker price for a hotel room is so it feels like a bargain to lose some money gambling and then make it back on perks, even if those perks are excessively marked up. And it's harder for casinos to maintain a model where they subsidize some complements to bring in foot traffic and then monetize it in other ways; Vegas used to be a great vacation destination for people who didn't have any vices, because everything was priced as a loss-leader for vice. This collapse probably started when Sheldon Adelson came up with Comdex and started arbitraging this gap on his own. So if you see higher and more uniform prices for hotels owned by casino companies, part of what you're picking up is the fact that those companies are shifting to more general tourism businesses with a slight tilt towards casinos. ↩︎

  3. From a pure consumer welfare standpoint, the ideal arrangement for any given commodity industry is that there's one monopolist who controls almost the entire market, and they're overconfident that cutting prices to the point that they lose money will scare off their competitors. In that model, the monopoly is just a big company that willingly loses money, and continuously transfers wealth from itself to competitors. This model is a little too cute to have direct applications, but it's a good reminder that "cut prices until every competitor is wiped out, then jack them up" doesn't describe a one-time effort but instead describes the two poles an abusive monopolist oscillates between. ↩︎

You're on the free list for The Diff! Upgrade today for full access. You'll also unlock the full archive, including this week's posts about Michael Burry and Gamestop ($), how companies are reframing hiring, partly in light of AI ($), and the tradeoffs of Elon Musk merging all of his companies together ($).


Upgrade Today

Diff Jobs

Companies in the Diff network are actively looking for talent. See a sampling of current open roles below:

Even if you don't see an exact match for your skills and interests right now, we're happy to talk early so we can let you know if a good opportunity comes up.

Find a Role

If you’re at a company that's looking for talent, we should talk! Diff Jobs works with companies across fintech, hard tech, consumer software, enterprise software, and other areas—any company where finding unusually effective people is a top priority.

Find Talent with Diff Jobs

Elsewhere

Strategic Investments

In September, Nvidia and OpenAI announced a partnership in which Nvidia would invest "up to $100bn" in OpenAI. Even for a market that was used to record-setting headlines, this was still meaningful and moved Nvidia shares up about 4%. Now, Nvidia is having second thoughts, at least internally ($, WSJ). This story came out over the weekend, which was actually quite sporting of the WSJ in that it gave the companies involved a chance to set the record straight. Which Jensen Huang, at least, has done in a tepid way, by saying it might be "the largest investment we’ve ever made." Given that their largest ever M&A deal, for Groq, was worth $20bn, Nvidia could theoretically cut its planned spending by 80% without having to walk back that comment.

This sets up a tricky situation for both companies. Nvidia's share price and revenue growth are ultimately downstream from the operating cashflows of a handful of companies and the venture checks into a larger but riskier cohort. So if Nvidia cuts its investment, and this reads as Nvidia getting marginally less bullish, Nvidia's market cap probably declines by more than the amount by which they cut it. But if Nvidia is bearish on OpenAI, they're telling a key customer that it might want to take a second look at backup plans. On one hand, custom silicon takes a long time to get right, even if you buy rather than build it from scratch. But the more pessimistic you are on timelines, the higher sell side consensus Nvidia revenue is for the period in which those chips would be delivered.

Most of the cross-shareholdings and weird deal structures in AI have a reasonable explanation; Nvidia wants a fragmented end market, and generates cash flow that could be deployed by companies that are buying compute upfront and using it over time. But those financial deals have also put Nvidia in a position where its strategic investments are a direct input into the sentiment that ultimately funds those investments. It's potentially expensive to buy into a growth company at peak valuation when it has multiple competitors who are gaining share (though the status ranking for AI labs is pretty volatile). In this case, it's also pretty expensive not to do it.

Disclosure: long NVDA.

Media Form Factors

The mass-market paperback may be going away. Which is not a decline-of-reading story, because the physical book business overall seems fine. Their problem is basically:

  1. They were already more long-tail than other categories, partly because you could fit more discrete titles on a single shelf, and partly because the fixed cost was lower so there was more room for experimentation. But that meant that they competed against digital formats that were much more long-tail, and where searches or purchase-based recommendations could beat browsing. (That's partly offset by a loss of serendipity; it would be nice if a major online bookstore experimented with a recommendation algorithm that deliberately sought out obscure books on the same topic, essentially telling you what else you might spot on the same shelf in a used bookstore.)
  2. They're portable, but there's another portable entertainment source that's lighter, smaller, and offers a wide selection of genre fiction in digital format in addition to other kinds of entertainment.
  3. They don't look impressive on a shelf, and physical books remain both an information storage medium and a decorative object.

There are always categories of the economy that seem like a necessity but turn out, after some technological advance, to be entirely circumstantial. But this is usually hard to see until they're on the way to disappearing.

Fees

If you pay someone a fee to invest in a private company, you're paying for some combination of access and selection. But these can be unbundled in the form of special-purpose vehicles, which are purely based on access. In one sense, you could say that getting into hot deals is less than the totality of a venture capitalist's skill, so they should get paid a lower cut. And a single-stock investment vehicle doesn't have the ongoing costs that an actively-managed fund would, so there's no reason for an ongoing management fee. But you could also say that access to investment opportunities in private companies probably correlates with access to people who have lots of money but are somewhat cost-insensitive. That last one wins at least some of the time, because there are some pretty lopsided deals with high upfront markups coupled with carry ($, WSJ). There are plenty of deals that are incentive-aligned, where an SPV solves the problem that access and liquidity are not perfectly correlated, and generates fees to compensate for that. But those will start to get crowded out by easier-to-sell opportunities. The highest-fee vehicles seem to be in high-profile companies, and presumably part of the return investors are getting and paying for is that they get to say they're already a shareholder of SpaceX, OpenAI, etc. It ends up being another way to motivate the biggest late-stage startups to go public, so the retail investor demand for their stock doesn't get absorbed by private company brokers first.

Critical Minerals

The US plans to spend $12bn on a critical minerals stockpile. The practical argument for this is that the US doesn't always have consistent access to these materials, but sometimes needs them, and that we ought to have a backup plan. There's also a financial argument for government stockpiles of commodities: systematically buying low and selling high produces returns similar to selling options, and governments have the balance sheet to harvest this risk premium. Mining companies tend to trade at low multiples when demand is high, but the other way to look at that is that they're much more sensitive to short-term profits than to some long-term shift in their value. They won't necessarily invest in more capacity, since they know that a one-time stockpile is exactly that. But their shareholders will appreciate the elevated earnings while it happens.

Brand Names

OpenAI is finally retiring GPT-4o, after trying to do so last year. Their announcement is basically begging their users to upgrade to a model that has the same qualitative features, but works better. It's interesting that naming LLMs has been a notoriously hard problem (they improve too quickly to give them some kind of gold/platinum/diamond branding without eventually running out of superlatives, and it is kind of hard to describe the difference between various flavors of the same model without just telling people to try them out). Model numbers tend to be forgettable, but when they catch on—747, F-150, AK-47—it's hard for similar brand names without any connotations to compete.