EconWorks Podcast

Who Decides When AI Is Safe Enough?

Who should decide when an AI model is safe enough to release? AI companies have a responsibility to develop safe products. But does that responsibility allow competing developers to collectively restrict innovation? In this episode of EconWorks, we explore the tension between AI safety and antitrust law. We start with an unusual example: an agreement among competitors to lower prices. Even when consumers appear to benefit, such an agreement can violate antitrust law. We then examine two Supreme Court cases that help explain why some forms of cooperation are lawful while others restrict competition. In Broadcast Music, Inc. v. CBS, the Court recognized that collaboration among competitors can create new products and economic efficiencies. In *National Society of Professional Engineers v. United States*, the Court rejected an attempt to suppress competition on the grounds that competitive bidding could compromise safety. Finally, we consider the September 29, 2026, AI safety accord. Its commitments to internal controls, independent audits, and corporate oversight offer a possible approach to improving AI safety while preserving independent innovation decisions. Can companies take responsibility for safety without collectively determining the pace of technological progress? Read the full article and graphic analysis: https://blog.econworks.com/p/can-ai-companies-agree-to-slow-innovation?r=562wri Explore more visual economics content: https://econworks.com YouTube: https://www.youtube.com/@EconWorks-d3e Substack: https://blog.econworks.com This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit blog.econworks.com/subscribe

• Transcript

Can AI Stop Cartels Without Detecting Them?

Why do cartels continue to exist? And could AI make them less profitable without ever discovering an illegal agreement? In the final episode of How Cartels Really Work, EconWorks examines how artificial intelligence could change the incentives that make cartels profitable. We begin with a simple hypothetical: An economy contains 1,000 cartels. AI could make collusion easier, increasing their number to 10,000, or strengthen competitive pressures, reducing their number to 100. Either way, some cartels remain profitable. We explore why detecting illegal agreements is difficult and revisit the Nasdaq market-maker case, where unusual pricing patterns prompted an extensive antitrust investigation. But AI’s more significant contribution may come from buyers rather than competition authorities. Consider a marketing board whose members collectively sell one billion units annually. A price increase of just one cent per unit could generate $10 million in additional revenue. For an individual consumer, the same increase might cost only $1 annually—not enough to justify spending time searching for a better deal. AI could change that calculation. By reducing the costs of comparing offers, finding alternatives, and coordinating purchases, AI could allow smaller businesses and individual consumers to become more sophisticated buyers. As buyers respond more effectively to higher prices, some cartels may become less profitable or even collapse. And none of this necessarily requires buyers to identify an illegal agreement. The central question is not simply whether AI can detect cartels, but whether it can change the economic incentives that make collusion profitable. This episode concludes our ten-part exploration of cartel economics and introduces the next upcoming EconWorks series, When Your AI Meets Their AI. Read the full article and graphic analysis: https://blog.econworks.com/p/cartels-episode-10-the-other-side?r=562wri Explore more visual economics content: https://econworks.com YouTube: https://www.youtube.com/@EconWorks-d3e Substack: https://blog.econworks.com This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit blog.econworks.com/subscribe

• Transcript

AI Safety or Collusion? Who Gets to Decide?

Developers of AI have expressed concerns about the dangers of increasingly powerful artificial... But could their safety pact lead to antitrust problems if rival firms collectively cut back on innovation? On this edition of EconWorks, we take a look at the difference between legitimate cooperation and agreements that restrain competition. Let us begin with a simple question. If competitors cannot agree to fix prices, even at lower levels, when can they legitimately collaborate? Two landmark decisions in the Supreme Court provide some important insights: - Broadcast Music v. CBS (1979): Joint efforts by competitors that result in valuable goods and reduced transaction costs. - *National Society of Professional Engineers v. United States* (1978): Why safety issues don’t justify limiting competition as a matter of course. Next up we go to the AI safety pact of September 29, 2026, and its pledges on internal controls, independent evaluations, and corporate oversight. The deal creates an important opening: AI companies might be able to improve safety without collectively slamming on the brakes of innovation. The obvious questions are Who is responsible for the safety of AI? How much safety is enough? Who decides? And when does cooperation between competitors become an antitrust issue? EconWorks examines these questions through the lens of economics, antitrust law, and the incentives that shape AI markets. Read the full article and graphic analysis: https://blog.econworks.com/p/can-ai-companies-agree-to-slow-innovation?r=562wri Explore more visual economics content: https://econworks.com YouTube: https://www.youtube.com/@EconWorks-d3e Substack: https://blog.econworks.com This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit blog.econworks.com/subscribe

Can AI Stop Cartels Without Detecting Them? | How Cartels Really Work — Ep. 10

Why are cartels still around when they are illegal? And could an artificial intelligence make them less profitable without even detecting them?In the last episode of How Cartels Really Work, EconWorks explores how AI may alter the economic incentives behind collusion.Suppose an economy with 1,000 cartels. AI could make coordination easier, and the number of cartels could rise to 10,000. Or AI could increase competition and take the number down to 100.But even if there are only 100 cartels left, the basic question remains: Why do firms keep colluding?We discuss the problem of detecting cartels, and we use the Nasdaq market-maker case to show the difference between detecting suspicious pricing patterns and proving an illegal agreement.We then turn to purchasers.Large companies such as Costco can shop around for suppliers and negotiate the best prices. Smaller businesses and individual consumers often don’t have the same tools.AI may change that.Let's say a marketing board represents producers that sell a billion units a year between them. Raising the price by one cent per unit could bring in an extra $10 million in revenue.So for a customer who buys 100 units a year, the same increase will cost just $1.In the past, the consumer had little motivation to spend time and effort searching for a better deal.But what if an AI agent could do the searching, comparing, and negotiating?As buyers become more sophisticated, it may become harder and harder for some cartels to keep prices high.The buyers wouldn't have to know about the cartel. They would just have to find better buying opportunities.In this episode, we explore how AI could alter the incentives for cartel members, buyers, and competition authorities—and why some cartels might still be profitable even if everyone is using AI. Read the full article and graphic analysis: https://blog.econworks.com/p/cartels-episode-10-the-other-side?r=562wri Explore more visual economics content: https://econworks.com YouTube: https://www.youtube.com/@EconWorks-d3e Substack: https://blog.econworks.com This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit blog.econworks.com/subscribe

DoorDash–Instacart: Two Shoppers, Two Competition Stories

Do you choose your supermarket first, or open a delivery app and then decide where to shop? That everyday decision provides a starting point for understanding DoorDash’s proposed acquisition of Instacart. In this EconWorks conversation, we follow two hypothetical neighbors whose grocery orders look identical but whose alternatives may differ. Could retailer delivery keep prices competitive? Would combining memberships benefit households? Could different discounts leave some shoppers more exposed to the loss of rivalry? And what evidence would show whether the merger’s promised savings outweigh its pricing incentives? We explore both the potential concerns and the efficiency case, explaining why a large market share does not answer every competitive question. EconWorks will follow developments and revisit the analysis as more evidence becomes public. Read the full article and graphic analysis: https://blog.econworks.com/p/doordashinstacart-whose-choices-would?r=562wri Explore more visual economics content: https://econworks.com YouTube: https://www.youtube.com/@EconWorks-d3e Substack: https://blog.econworks.com This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit blog.econworks.com/subscribe

• Transcript

Bid Rigging: Can AI Help Buyers Design Better Auctions?

Three companies bid on a government contract. One wins, the rest lose.But what if the companies agreed on the winner before they made their bids?In Episode 9 of How Cartels Really Work, EconWorks explores the economics of bid rigging using a real U.S. school bus cartel case.In United States v. Southwest Bus Sales (1994), the school bus distributors coordinated their bids and divided contracts amongst themselves. But the case also raises an interesting economic question: Were the school districts capable of designing better auctions?Bus manufacturers sold to the distributors and could offer them wholesale price breaks to help them compete. This meant that competition between manufacturers could influence the bids made by distributors.What if school districts held separate auctions for the manufacturing of buses and the distribution services?And what if they had AI?An AI buying assistant might help buyers evaluate alternative auction designs, identify additional suppliers, and monitor procurement outcomes over time.Buyers may not need to show that their suppliers are colluding before they improve their procurement processes.If suppliers realize that buyers are using AI to monitor and reverse-engineer their auctions, they may reconsider whether collusion is worthwhile.But there is another option: suppliers can also use AI.Could competing artificial intelligence systems learn to coordinate their bidding? And what happens when buyers and suppliers are constantly adapting their strategies to each other?In this episode we explore:How repeated procurement auctions enable bid rigging.The U.S. school bus cartel and the role of the manufacturers and distributors.The effect of auction design on competition is also important.How separate auctions for manufacturers and distributors could change procurement outcomes.How AI could help buyers design and monitor auctions at lower cost than traditional economic consulting.Why knowledge of buyers’ use of AI could deter suppliers from colluding.How supplier AI could create new competition challenges.The key question: Can buyers make cartels less profitable by designing better auctions rather than just trying to detect collusion? Read the full article and graphic analysis: Explore more visual economics content: https://econworks.com YouTube: https://www.youtube.com/@EconWorks-d3e Substack: https://blog.econworks.com This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit blog.econworks.com/subscribe

• Transcript

DoorDash–Instacart: Will Your Delivery Cost More?

Two neighbors order the same groceries from the same supermarket. One chooses the store first. The other opens a delivery app first. Could DoorDash’s proposed acquisition of Instacart affect them differently? This EconWorks episode explores how people choose grocery delivery, when a supermarket’s own service provides an alternative, and why switching between DoorDash and Instacart matters after common ownership. We also examine market shares, personalized offers, and the potential savings from combining delivery operations. The question is whether consumers would face better or worse terms than if the companies continued competing. Anna and Ben are illustrative shoppers, not measured consumer segments. The episode analyzes a proposed transaction and does not assume higher prices or a particular regulatory outcome. Read the full article and graphic analysis: https://blog.econworks.com/p/doordashinstacart-whose-choices-would?r=562wri Explore more visual economics content: https://econworks.com YouTube: https://www.youtube.com/@EconWorks-d3e Substack: https://blog.econworks.com This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit blog.econworks.com/subscribe

Bid Rigging: Can AI Help Buyers Fight Cartels? | How Cartels Really Work Ep. 9

Three companies bid on a government contract. One wins, the others lose. But what if the companies agreed on the winner before they placed their bids? In Episode 9 of How Cartels Really Work, EconWorks takes a look at the economics of bid rigging, using a real U.S. school bus cartel case. In United States v. Southwest Bus Sales (1994), the distributors of school busses had coordinated their bids and divided contracts among themselves. But the case also poses an interesting economic question: Could the school districts have designed better auctions? Bus manufacturers sold to the distributors and could provide them wholesale price breaks to help them compete. This meant that the bids made by distributors could be influenced by competition between manufacturers. What if school districts put separate auctions for the manufacturing of buses and the distribution of services? And what if they had AI? An AI buying assistant could help buyers assess alternative auction designs, find additional suppliers, and track procurement performance over time. Buyers may not need to prove that their suppliers are colluding if they want to improve their procurement processes. Suppliers may also reconsider collusion if they realize that buyers are using AI to monitor and reverse-engineer their auctions. But there is another option: suppliers can also utilize AI. Would competing artificial intelligence systems learn to coordinate their bidding? What happens when buyers and suppliers are always adapting their strategies to each other? In this episode we look at: How repeated procurement auctions enable bid rigging. The Role of the U.S. School Bus Cartel and the Manufacturers and Distributors. The impact of auction design on competition is significant. How separate auctions for manufacturers and distributors might affect procurement results. How AI could assist buyers in designing and monitoring auctions at lower cost than traditional economic consulting. Why knowing buyers' use of AI could discourage suppliers from colluding. How AI from suppliers could pose new competition challenges. The central question is whether buyers can design better auctions to make cartels less profitable rather than relying on detecting collusion. Read the full article and graphic analysis: https://blog.econworks.com/p/cartels-episode-9-bid-rigging-can?r=562wri Explore more visual economics content: https://econworks.com YouTube: https://www.youtube.com/@EconWorks-d3e Substack: https://blog.econworks.com This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit blog.econworks.com/subscribe

SpaceX–Tesla: A Big Merger, but a Weak Antitrust Case?

What would really change if Tesla and SpaceX merged? Rockets, vehicles, connectivity, energy storage, and AI. Those all sound consequential. But the size and the technological reach do not prove the merger would harm competition. In this episode of EconWorks we examine possible concerns about blocking rivals, bundling products, and killing future competition—and why those theories don’t seem to have much support at the moment. We also discuss the possible benefits of common ownership for better engineering, investment coordination, and cost reduction. The key question: What difference would ownership make that companies cannot achieve through contracts and cooperation? This episode is looking at a hypothetical merger; it’s not assuming a deal is done. Read the full article and graphic analysis: https://blog.econworks.com/p/spacex-and-tesla-what-would-a-merger?r=562wri Explore more visual economics content: https://econworks.com YouTube: https://www.youtube.com/@EconWorks-d3e Substack: https://blog.econworks.com This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit blog.econworks.com/subscribe

• Transcript

How Cartels Fool Sophisticated Buyers—and How AI Could Help

How can a cartel survive for years when its customers are sophisticated firms with substantial buying power? In episode 8 of How Cartels Really Work, we examine two large cases in history, the automobile parts cartels and the international vitamins cartel. Both involved suppliers selling to large businesses with professional purchasing departments. But some of these illegal agreements ran for a decade or more. Why didn’t their customers see the cartels? We begin with Costco’s buying practices. If suppliers want to raise their prices, Costco’s buyers examine the underlying costs and challenge any increases that don’t appear warranted. Sophisticated purchasing managers may have trouble figuring out if their suppliers are colluding. A company might hire economists to analyze the prices, costs, and bidding strategies of suppliers. However, it can be expensive to do detailed economic analysis on thousands of products. Artificial intelligence could alter that equation. AI may allow buyers to look at more transactions, identify suspect pricing patterns, and uncover alternative suppliers at a much lower cost. And here’s an intriguing possibility: Buyers might only have to show that a cartel is less profitable. AI could increase competitive pressure on suppliers and make some cartels harder to sustain if it helps customers identify better alternatives. AI could alter the economics of collusion by making buyers better at buying, right? Read the full article and graphic analysis: https://blog.econworks.com/p/cartels-episode-8-the-biggest-cartels?r=562wri Explore more visual economics content: https://econworks.com YouTube: https://www.youtube.com/@EconWorks-d3e Substack: https://blog.econworks.com This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit blog.econworks.com/subscribe

• Transcript

SpaceX + Tesla: Would a Merger Harm Competition?

What would really change if Tesla and SpaceX merge? Rockets, vehicles, connectivity, energy storage, and AI. That sounds consequential. But the size and the technological reach are not evidence that the merger would harm competition. In this episode of EconWorks we look at possible worries about blocking rivals, bundling products, and killing future competition—and why those theories don’t seem to have much support right now. We also discuss how common ownership might improve engineering, coordinate investment, and reduce costs. The key question: What difference would ownership make that the companies cannot achieve through contracts and cooperation? This episode is looking at a hypothetical merger; it’s not assuming a deal is done. Read the full article and graphic analysis: https://blog.econworks.com/p/spacex-and-tesla-what-would-a-merger?r=562wri Explore more visual economics content: https://econworks.com YouTube: https://www.youtube.com/@EconWorks-d3e Substack: https://blog.econworks.com This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit blog.econworks.com/subscribe

The Biggest Cartels You've Never Heard Of | Can AI Stop Them?

How can a cartel survive for years when its customers are sophisticated firms with large buying power? Episode 8 of How Cartels Really Work looks at two big cases in history: automobile parts cartels and the international vitamins cartel. Both were suppliers selling to large businesses with professional purchasing departments. But some of these illegal deals lasted for a decade or more. Why didn't their customers see the cartels? Let's start with Costco's buying practices. Costco’s buyers look at the underlying costs when suppliers want to raise their prices and push back on any increases that don’t seem justified. For sophisticated purchasing managers, it may be difficult to tell if their suppliers are colluding. A firm might employ economists to analyze the prices, costs, and bidding strategies of suppliers. However, it can be costly to perform detailed economic analysis on thousands of products. Artificial intelligence could change that equation. AI might enable buyers to review more transactions, detect suspect pricing patterns, and identify alternative suppliers at a much lower cost. And here’s a tantalizing possibility: Buyers may be able to make a cartel less profitable without proving it exists. If AI enables customers to discover better alternatives, then that could increase competitive pressure on suppliers and make some cartels harder to sustain. Could AI change the economics of collusion by making buyers better at buying? Read the full article and graphic analysis: https://blog.econworks.com/p/cartels-episode-8-the-biggest-cartels?r=562wri Explore more visual economics content: https://econworks.com YouTube: https://www.youtube.com/@EconWorks-d3e Substack: https://blog.econworks.com This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit blog.econworks.com/subscribe

When Is a Discount a Wall?

A discount is usually the shape of competition. But can the structure of a discount render the purchase of a cheaper rival more expensive? This episode starts with a classic pricing problem in FTC v. Syngenta & Corteva. A company may charge more for units with high willingness to pay and less for additional units. It can be done through quantity discounts or nonlinear pricing. There is nothing wrong with that per se. The puzzle starts when the rebate depends on the customer’s proportion of purchases from the seller, rather than just the total amount bought. In a simple example, buying one more generic unit saves a mere 25 cents but risks the loss of a $1,250 rebate. The generic is cheaper, but switching is expensive. This leaves two explanations: simple price discrimination or exclusion. The episode asks what proof would separate them—and why internal pricing documents might be more important than the label on the discount. Read the full article and graphic analysis: https://blog.econworks.com/p/when-is-a-discount-a-wall?r=562wri Explore more visual economics content: https://econworks.com YouTube: https://www.youtube.com/@EconWorks-d3e Substack: https://blog.econworks.com This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit blog.econworks.com/subscribe

• Transcript

Why Cartels Collapse—and Why AI May Change That

Cartels are unstable by nature. Members cheat, managers change, companies are sold, and participants can get away with less punishment by reporting the cartel before everybody else. When AI encounters that familiar process, what happens next? Join us for episode 7. We start with the European MCAA cartel. Clariant acquired Hoechst’s chemicals business in 1997; the cartel continued for a while, and Clariant later discovered the conduct and reported it to the European Commission. The announcement triggered an investigation by the Commission. That history reveals a new source of cartel instability that has been overlooked: the firms may remain constant, but the people and organizations within them change. AI complicates the narrative. If business divisions can have their pricing systems, senior management may see the results but be unaware of the strategies the systems have learned. An internal investigation may uncover substantial amounts of information but not the telltale signs of a human cartel—meetings, emails, or an explicit agreement. In contrast, the situation is with the DOJ’s RealPage case. RealPage involved a visible common pricing system and an alleged sharing of competitors' nonpublic information. But what if every landlord had their own autonomous AI and no confidential information was shared? Would investigators be looking at an AI agreement, or just interdependence, if their prices stayed coordinated? The law has a lot of experience distinguishing agreement from parallel behavior among humans. It has much less experience making the same sort of distinction when autonomous systems interact. Read the full article and graphic analysis: https://blog.econworks.com/p/cartels-episode-7-why-cartels-collapseand?r=562wri Explore more visual economics content: https://econworks.com YouTube: https://www.youtube.com/@EconWorks-d3e Substack: https://blog.econworks.com This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit blog.econworks.com/subscribe

• Transcript

When Is a Discount a Wall? | FTC v. Syngenta & Corteva

Discounts always feel like competition. Lower prices should help the customer and make rivals respond. But what if a big rebate is contingent not just on how much a customer buys from a company but on how little it buys from its competitors? In this episode of EconWorks, we examine the economics of loyalty discounts through FTC v. Syngenta & Corteva. Let’s start from the perspective of a pricing specialist. Quantity discounts and nonlinear pricing are perfectly normal responses to differences in willingness to pay. The problem is not price discrimination per se. Then a generic competitor comes in. If all the customer does is to preserve a big rebate by buying 90% or more of its requirements from the incumbent, suddenly buying one extra cheaper generic unit can be costly. So the key question is not about whether there was a discount. The answer is: What was the discount intended to buy? More purchases from the incumbent, or fewer purchases from the rival? We also consider the evidence that may differentiate the two explanations, including pricing documents, loyalty thresholds, generic entry, and the magnitude of the "loyalty cliff." Read the full article and graphic analysis: https://blog.econworks.com/p/when-is-a-discount-a-wall?r=562wri Explore more visual economics content: https://econworks.com YouTube: https://www.youtube.com/@EconWorks-d3e Substack: https://blog.econworks.com This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit blog.econworks.com/subscribe

Why Cartels Collapse—and Why AI May Change That | Cartels Episode 7

Why do cartels fall apart? The normal answer is cheat. A cartel member secretly cuts prices, another firm ratifies the deal, or someone runs to the authorities for leniency. But there’s another source of instability: organizations evolve. In Episode 7 of *How Cartels Really Work* we examine the European MCAA cartel and Clariant’s takeover of Hoechst’s chemical business. The cartel survived the change of ownership for some time, but Clariant later uncovered the conduct through its compliance efforts and sought leniency, which sparked the European Commission’s investigation. And then, what if the same story has artificial intelligence involved? A multi-product company may have separate pricing AIs running in different divisions. The senior management might see the profits, but they may not understand the strategies that the systems have learned. RealPage offers a useful comparison. Regulators might spot common pricing schemes and track the movement of nonpublic information among competitors. But what if every landlord had a different autonomous AI, no confidential data was shared, and the AIs still came up with coordinated-looking prices? The evidence can be voluminous. The interpretation of the law may not. In this episode, we disentangle human collusion, algorithm-assisted coordination, AI-to-AI coordination, and good old-fashioned interdependence—and why AI may make one traditional cartel-breaking mechanism less reliable. Read the full article and graphic analysis: https://blog.econworks.com/p/cartels-episode-7-why-cartels-collapseand?r=562wri Explore more visual economics content: https://econworks.com YouTube: https://www.youtube.com/@EconWorks-d3e Substack: https://blog.econworks.com This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit blog.econworks.com/subscribe

Why Is Everyone Moving Across the AI Stack?

Why are AI companies moving across the stack? Nvidia is moving toward developer platforms through Hugging Face. AMD is moving toward model development through World Labs. Cursor is moving upstream from the application layer toward the models and computation. These moves all look like vertical integration, but the underlying economics may be entirely unique. Nvidia may be trying to expand demand for compute. AMD may be trying to become a stronger competitor. Cursor may be trying to improve the application by controlling more of the technology underneath it. This episode explores why firms integrate across the AI stack and why antitrust analysis should focus on what changes because of ownership—not simply on the fact that companies now operate at multiple levels. Read the full article and graphic analysis: https://blog.econworks.com/p/why-is-everyone-moving-across-the?r=562wri Explore more visual economics content: https://econworks.com YouTube: https://www.youtube.com/@EconWorks-d3e Substack: https://blog.econworks.com This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit blog.econworks.com/subscribe

• Transcript

Can AI Raise Prices Without Colluding?

Suppose every business provides its AI with the same instruction: maximize our long-run profits, but obey the law. The AI estimates demand, studies customer behavior, learns how competitors respond, and gradually discovers that some forms of aggressive competition are not very profitable. A price cut may trigger another price cut. A promotion may trigger a competing promotion. Attempts to steal customers may provoke retaliation. Eventually, each firm’s AI may independently decide to compete less aggressively. But there is no agreement. The AIs never communicate. Nobody fixes a price. In Episode 6 of How Cartels Really Work, we look at why this problem predates artificial intelligence. U.S. cartel law has long distinguished unlawful agreements from lawful interdependence among firms that independently recognize how competitors are likely to respond. AI may not change that legal rule. What AI could change is how cheaply, quickly, and accurately firms learn to maximize profits. The result raises a provocative question: what happens if AI makes firms much better at producing less aggressive competition without ever forming a cartel? Read the full article and graphic analysis: https://blog.econworks.com/p/cartels-episode-6-what-happens-when?r=562wri Explore more visual economics content: https://econworks.com YouTube: https://www.youtube.com/@EconWorks-d3e Substack: https://blog.econworks.com This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit blog.econworks.com/subscribe

• Transcript

Why Is Everyone Moving Across the AI Stack?

In this episode of EconWorks, we look at why some of the biggest AI companies are jumping across different layers of the AI stack. Nvidia transitions from compute to developer platforms via Hugging Face. World Labs is the pivot from compute to model development for AMD. Cursor sits near the application layer and flows upstream toward models and compute. All three moves look like vertical integration at first glance. But the economic logic is different in each of these cases. Nvidia may be trying to build demand for AI compute. Maybe AMD is trying to obtain better hardware, software, and systems so it can compete better. Cursor might be trying to take more control over the models and compute underneath it to improve the application. The antitrust question isn't just whether firms are moving up and down the AI stack. But that is different when you own it. Does integration enhance output, improve products, and enhance competition? Or does it result in foreclosure, steering, higher rivals’ costs, or control over an important bottleneck? Read the full article and graphic analysis: https://blog.econworks.com/p/why-is-everyone-moving-across-the?r=562wri Explore more visual economics content: https://econworks.com YouTube: https://www.youtube.com/@EconWorks-d3e Substack: https://blog.econworks.com This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit blog.econworks.com/subscribe

What Happens When Every Firm Gets Better at Maximizing Profit?

What if every business told an AI one simple thing: **maximize long-term profit without breaking the law**? A company today needs economists, salespeople, marketers, lawyers, finance departments, and managers. They have different information, different incentives, and often conflicting advice. AI could well embed a lot of those functions into a much more disciplined profit-maximizing system. The result raises an odd antitrust issue. An artificial intelligence can learn that aggressive price cuts tend to trigger competitive responses. It may discover that promotions, expansions of capacity, or efforts to lure away customers provoke retaliation. It may therefore be able to recommend less aggressive competition on its own. There was no secret meeting. No agreement to fix prices. No competitor communication. So where is the cartel? "In Episode 6 of How Cartels Really Work, EconWorks talks about the role of interdependence, the historic legal distinction between agreement and independent conduct, and why AI may make an old boundary in cartel law much more important. The unexpected problem could be that AI gets good at breaking antitrust law. It could be that AI gets much better at maximizing profits while obeying it. Read the full article and graphic analysis: https://blog.econworks.com/p/cartels-episode-6-what-happens-when?r=562wri Explore more visual economics content: https://econworks.com YouTube: https://www.youtube.com/@EconWorks-d3e Substack: https://blog.econworks.com This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit blog.econworks.com/subscribe