On September 2, Asad Ramzanali and Joel Thayer argued in ProMarket that Google is using search-monopoly cash, compute, and data to buy the AI market, and that the D.C. Circuit should treat Judge Amit Mehta’s search remedies as too light. I take the product-competition story seriously. The distribution numbers they cite already show how that check was supposed to work, and how it is working.

The steelman is not silly. In 2024, Mehta found Google liable for illegally maintaining a monopoly in general search. He later declined the big structural asks. No ban on the Apple default deal. No forced sale of Chrome or Android. A central piece of the rationale, as Ramzanali and Thayer reconstruct it, was that generative AI might discipline that monopoly. OpenAI and Anthropic looked like they could take queries away from the search box. If users start asking ChatGPT instead of Google, the old exclusive-default machinery matters less. A court that refuses to forecast a new industry, and that refuses to break up a company on a guess, can call that humility.

For a stretch the scoreboard even cooperated. ChatGPT’s launch in November 2022 put Google into its own “code red,” which the New York Times reported at the time. Three years later the tables flipped. The Information reported that Sam Altman declared a code red in late November 2025 after Google shipped Gemini 3, and told staff to deprioritize work that did not improve ChatGPT. ChatGPT later crossed a billion monthly active users. The Verge put that milestone in early August 2026. Reuters, citing external data, said the app may have crossed it in June. Five days after OpenAI announced the figure, Sundar Pichai said Gemini had also reached a billion monthly active users. That is a horse race. Horse races are what people mean when they say the market is working.

The ProMarket piece is an amicus brief in essay form, and it does not deny the horse race. It says the court treated the horse race as a substitute for a search remedy. That substitution only holds if Gemini has to win users the way ChatGPT did, by being chosen. Gemini does not have to do that. When Google wanted the model in front of people, it put AI Overviews on top of Google Search. Similarweb’s 2026 generative-AI landscape report, as cited by Ramzanali and Thayer, put AI Overviews on more than 43 percent of Google searches as of May 2026. The overview summarizes the results and can open into a chat. The user did not install an app. The user typed a query into the same box they have typed into for twenty years.

ChatGPT had to become a destination. Gemini inherited one. That is about placement, not about which lab writes a better eval. Gmail, Maps, Drive, and Photos do the same job at smaller scale. Each one is a place people already are. Each one can feed Gemini usage and, over time, training data that a chatbot born in 2022 has to buy, scrape, or partner for.

The rest of the stack repeats the placement advantage. An AI Overview can run on Gemini, served from Google Cloud, on TPU silicon Google designed, inside Chrome, on Android, on a Pixel, through a Google fiber line, in a data center Google owns. OpenAI and Anthropic can rent pieces of that stack. They cannot put a competing overview on google.com. Menlo Ventures’ December 2025 enterprise snapshot, again as cited in the ProMarket essay, had Anthropic, OpenAI, and Google at 88 percent of foundation-model API revenue, with Google at 21 percent. That is a concentrated API market. It is still the market in which Google is the only one of the three that also owns the default search surface.

Google also gets paid if it loses the chatbot contest. Ramzanali and Thayer note that Google is both a large investor in Anthropic and a large supplier of Anthropic’s compute, and that OpenAI has started using Google Cloud among other providers. I would not call that a conspiracy. It is a vertical. If Gemini takes the consumer surface, Search and Cloud keep the query and the inference. If Claude or ChatGPT takes more of the API wallet, Cloud still sells the GPUs and TPUs. New chat products were supposed to eat search. Search cash and cloud capacity can eat those products from both sides.

The Apple deal is the cleanest exhibit. Mehta declined to bar Google’s payments to Apple of roughly $20 billion a year for default placement in Safari. He required only that the arrangement not be exclusive. In January 2026, Apple and Google announced that Gemini would power a Siri and Apple Intelligence upgrade. CNBC, TechCrunch, and Apple’s own announcement put that date on the calendar. Four months after a non-exclusivity condition, the same two companies bought the next default, this time inside the assistant the court had treated as a potential search rival. Elon Musk called the concentration unreasonable. He has his own model to sell. The contract still happened.

I do not need Ramzanali and Thayer’s full breakup menu to find that sequence ugly. American antitrust has a long habit of fining conduct and leaving the default in place. Google has lost cases in the U.S. and EU on ad tech, app stores, Android, and shopping units. Fines and data-sharing orders did not stop Gemini from becoming the fourteenth Google product with a billion monthly users. Product success is legal. Using an illegal search monopoly to keep the default, then pouring the default into the next interface, is the fact pattern the liability finding already described.

Judicial humility is the authors’ third heading, and it is the part I am most willing to grant and then drop. Mehta said judges should not pretend they can see AI’s path. Ramzanali and Thayer answer that following the statute on an old monopoly is humbler than forecasting a new one. The FTC’s supermarket line, which they quote, holds even if you dislike the rhetoric. The only grocer in town does not stop being the only grocer because it added a pet-food aisle. Generative AI is a new aisle. The search default is still the storefront.

The D.C. Circuit will review the remedy. It can tighten default rules, demand more data sharing, or leave Mehta’s package alone. It cannot un-ship AI Overviews. By May 2026 those overviews already sat on two out of five Google queries. Making Safari’s search default non-exclusive leaves google.com as the box most people still type into. Advertisers still pay for the clicks and the AI summary impressions. Apple still cashes a search check and now a model deal. OpenAI and Anthropic still rent Google’s metal if they need it.

The appeals court can still rewrite the Safari condition. It cannot unwind the January Gemini-Siri contract, and it cannot make the search box a neutral shelf for someone else’s model.