A Promising Decision on Algorithmic Price Fixing

And a Circuit Split; Is Supreme Court Review Likely?

Hours after my recent post on algorithmic collusion the Third Circuit issued an important decision that reinstated a plaintiffs’ class action complaint. The defendants were Atlantic City hotels and casinos, and the plaintiffs were a class of guests who alleged they paid too much because the defendants used price coordination software that facilitated collusion. The plaintiffs complained that the defendant casinos sent detailed non-public data concerning hotel room rates and occupancy to the defendant Cendyn via its “Rainmaker” software. Rainmaker then computed optimal (i.e., monopoly) prices. Further:

• Plaintiffs allege that the resulting anticompetitive rates are automatically uploaded into each casino-hotel’s room-selling platform, causing consumers to pay anticompetitively high prices for guest rooms.

Further, the complaint alleged, the defendants collectively agreed to charge these room rates, even though they may not have communicated with one another directly. The district court had followed some other courts in concluding that the scheme described in the complaint did not allege an agreement among the casinos themselves.

As the court described the Rainmaker Software:

• These algorithmic programs are typically trained on data specific to a vendor or market to suggest (and in some cases implement) prices. They learn through an iterative process of “trial and error and through finding patterns from a great volume and variety of data.” They may use data “related to past, present, and future supply and demand conditions,” including data on competitors’ public prices, to allow vendors to adjust prices frequently at a lower transaction cost. Although dynamic pricing algorithms vary in settings, parameters, function, and sophistication, they share the common purpose of “optimiz[ing] business processes, [thereby] allowing businesses to gain a competitive advantage by . . . setting optimal prices that effectively respond to market circumstances.”

As a result, the court concluded:

• AI can enable participants in the marketplace to coordinate pricing and thereby collude in ways that reduce competition at the expense of consumers who bear the brunt of higher prices and reduced output or supply.

This decision creates a split between the Third and Ninth Circuits, which came out the opposite way in a very similar case involving the same defendant and substantially similar software.¹ That increases the chances of Supreme Court review, although it hardly guarantees it. The Supreme Court has not decided an antitrust case on the merits since 2021.² Given the rapidly increasing role of AI engines in potentially anticompetitive practices and the sharpness of this split, this case presents an opportunity to make some helpful law. Cutting against review, however, is the fact that this case was decided on a motion to dismiss, which means that it has not gone through discovery. The Supreme Court does decide some cases on motions to dismiss, but is less likely to do so if there are important disputed facts.

For starters, a couple of observations are important.

1. For a trained agent, such as an AI-assisted computer program, computing the maximizing (cartel) price and output from appropriate data is child’s play. It can do so much better than any individual market participant very likely could. It can also do so in close to real time, provided it obtains the information quickly enough.

2. Section 1 of the Sherman Act requires an agreement that restrains trade, but it does not specify who must be the parties to that agreement.

3. The line between simple information exchanges and price-fixing remains important.

4. There is pre-internet precedent.

AI programs that are properly trained can easily estimate cartel prices by looking at pricing history, vacancy rates, and output changes in response to price changes. In fact, they can very likely do this with far greater speed and accuracy than the individual market participants could do for themselves.

Several courts including the Third Circuit in this case have concluded that a horizontal agreement among the principal sellers is necessary. They then queried whether such an agreement can be inferred from parallel conduct or actions that are contrary to individual self-interest. However, the Sherman Act does not require competing sellers to agree with each other. Indeed, §1 of the Sherman Act is routinely applied to vertical agreements between suppliers and dealers. It requires merely that there be a “contract, combination, or conspiracy,” and that it be “in restraint of trade,” which in this setting means that it results in higher prices or reduced output or quality.

In a “hub and spoke” conspiracy the hub is the cartel decision maker, or manager. It may be, but need not be, an actual cartel seller itself. Indeed, in this setting a trained AI software program is much better qualified to make the necessary calculations than individual sellers would be. Further, the key is not that the individual sellers, or spokes, actually communicate with one another, but rather that they agree with the hub to an output reduction that the hub computed and relayed to each member individually.

To take a simple example. Suppose ten identical sellers are currently producing ten units each at a price of $100. Now the software hub obtains the needed data and computes that the monopoly output and price would occur if each seller reduced its output to seven units and charged $160 each. Under the software agreement, which is purely vertical between the program supplier and each seller, each reduces its output accordingly, whether by agreement or default. In that case we have achieved the perfect cartel result without any communication whatsoever among the individual spokes. To be sure, there may be cases when the individual sellers do in fact communicate to one another, but that finding should not be essential. As a matter of enforcement policy, the cost of not making such an arrangement an antitrust violation would facilitate rampant legal cartels, assisted by the massive processing power of AI-assisted software. The most important question is whether individual sellers obligated themselves actually to set the output or price that the software computed for them. That agreement, even if vertical, crosses the line from information provision to price fixing.

The first Supreme Court decision to condemn a roughly similar restraint was the American Linseed case in 1923, long before the rise of AI or the internet.³ The defendant linseed producers were all subscribers to the “Amstrong Bureau of Related Industries,” a business periodical requiring each subscriber to submit detailed information about its business so that “one subscribing manufacturer may obtain detailed information concerning the affairs of others doing like business.” Armstrong’s requirement for subscribers was that “directly at the close of each day’s business each subscriber shall mail by special delivery to the bureau a complete report of all its carload sales for that day of oil, cake, or meal, not covered by its previous daily sales reports, which report shall disclose the quantity and kind, price, and terms, and whether for immediate or future delivery, and, if no sale has been so made, this fact shall be likewise reported.” According to Armstrong’s business model, the individual linseed producers did not fix prices, but they did commit to making ongoing accurate reports. Further, any subscriber with good cause had the right to demand confirmation of the data produced by a different subscriber. In addition, if any buyer rejected a subscribing seller’s price offer, that seller was entitled to obtain verification of prices from other subscribing sellers. The Court gave an example of the company’s correspondence describing such a situation:

• Gentlemen: Our Chicago manager advises us that under date of February 1st the Enterprise Paint Mfg. Co. informed him that they had bought 10 barrels linseed oil at less than $1.46 from another crusher in the Chicago territory. Will you kindly bulletin the subscribers with a view to finding out if any of the crushers sold this lot under their published price?

The Court unanimously affirmed that the Armstrong system violated the Sherman Act and issued an injunction. “With intimate knowledge of the affairs of other producers, and obligated as stated, but proclaiming themselves competitors, the subscribers went forth to deal with widely separated and unorganized customers necessarily ignorant of the true conditions. Obviously they were not bona fide competitors; their claim in that regard is at war with common experience, and hardly compatible with fair dealing.”

Should the arrangement involving the Atlantic City casinos be illegal per se? The relevant issue is not whether the individual spokes agreed with each other but whether the agreement is “naked” – that is, did it serve simply to raise price, or reduce output or quality, without performing any integrative function? If so, the per se rule applies. The most important indicator of a per se restraint is that the software not only facilitated computation of higher rates, but it also set those rates with each subscriber, either by agreement or default. In this case, the complaint alleged that the software’s recommended rates were “automatically uploaded into each casino-hotel’s room-selling platform.” It does not state that the hotels agreed not to change it, so it was presumably a default. A default tells you that a particular starting position can be changed. However, that is true of all cartel agreements, none of which are legally enforceable. Section 1 bars anticompetitive agreements, even those that cannot be legally enforced.

The defendants objected that the market had a great deal of excess capacity and that demand had been declining. But the courts have repeatedly and consistently rejected defenses to collusion based on “ruinous competition.” The market has its ways of responding to overbuilding or weak demand, but collusion is not a lawful alternative. As the Supreme Court observed in the Socony decision, which also involved a distressed industry with substantial excess capacity:

• Fairer competitive prices, it is claimed, resulted when distress gasoline was removed from the market. But such defense is typical of the protestations usually made in price-fixing cases. Ruinous competition, financial disaster, evils of price cutting and the like appear throughout our history as ostensible justifications for price-fixing. If the so-called competitive abuses were to be appraised here, the reasonableness of prices would necessarily become an issue in every price-fixing case. In that event the Sherman Act would soon be emasculated….⁴

1. Gibson v. Cendyn Grp., LLC, 148 F.4th 1069 (9th Cir. 2025), cert. denied, 2026 WL 1052046 (U.S. Apr. 20, 2026).

2. NCAA v. Alston, 594 U.S. 69 (2021).

3. United States v. American Linseed Oil Co., 262 US. 371 (1923).

4. United States v. Socony-Vacuum Oil Co., 310 U.S. 150, 220-221 (1940).