Google built and maintains its AI leadership from cash, compute, and data accumulated illegally from its monopoly in internet search. Its control over internet search, advertisement, mobile phone operating systems, and cloud computing continues to give it an advantage in AI that its competitors lack. The U.S. Court of Appeals for the District of Columbia Circuit must consider this entrenched and vertically integrated market position when it revisits the lower court’s lax remedies, write Asad Ramzanali and Joel Thayer.
At the end of November 2025, OpenAI CEO Sam Altman declared a code red in reaction to Google launching Gemini 3, a surprisingly capable artificial intelligence model. OpenAI employees were directed to deprioritize anything that didn’t improve the company’s core offering, ChatGPT. The tables had turned from three years prior when Google declared its own code red after ChatGPT launched.
In many ways, this is exactly how market competition should work. One company releases a product that becomes popular, prompting another to improve its own product.
But when Google kicked its AI efforts into gear in 2022, it had two advantages that other market participants didn’t. First, it had long been investing in AI. The company’s search engine monopoly threw off cash, compute, and data, which it accumulated and which became the ingredients for building AI models. Second, Google could weave its Gemini AI products into its myriad digital products, including the most popular web address in the world, google.com.
In 2024, Judge Amit Mehta found Google liable for illegally maintaining a monopoly in the search engine market, but he later rejected remedying the monopoly through tools commensurate with the size of the problem, like banning its exclusive default agreements with Apple or divesting Chrome or Android. A central aspect of his rationale was that while Google violated federal antitrust law, the new-fangled market of generative AI could discipline that monopoly by introducing startups like OpenAI and Anthropic capable of eating into its share of the search market.
As we told an appeals court reviewing that remedy decision in an amicus brief filed recently, this logic ignores history, the market dynamics of AI today, and the role of judicial humility.
The history
First, the history. By the early 2000s, Google commanded the search engine market. That monopoly position allowed the company to accrue profits not available in more competitive markets, build its own cloud computing infrastructure capable of hosting vast amounts of data, and collect an unparalleled amount of information about user interactions on its web properties.
In 2011, Google employed its first deep-learning team. By 2012, it had a team codenamed Google Brain spending millions on compute. And then it caught the bug. Google’s head of engineering was “intent on cornering the worldwide market for deep learning researchers—or at least coming close,” and he was given “free rein to secure any and all of the leading researchers” in the field, according to Cade Metz’s Genius Makers: The Mavericks Who Brought AI to Google, Facebook, and the World. It spent tens of millions doing that. Then in 2014, the company acquired the market-leading AI research startup DeepMind for $650 million.
This yielded research and commercial success. In 2015, the company started using AI-based RankBrain to order results for Google Search, soon sorting results for 15% of queries. And 2017 marked perhaps the most notable breakthrough in modern AI research. Eight Google staff authored a technical paper entitled “Attention Is All You Need,” which introduced the transformer architecture for AI systems. Transformers quickly became a seminal technical development, and the architecture still undergirds modern AI systems (the “T” in ChatGPT stands for transformer).
Google’s search monopoly financed and seeded Google’s ability to establish the field of generative AI and thrive in it. The cash, compute, and data that entrenched Google in the search market became the essential inputs to its AI position.
Market dynamics of AI today
It’s more than just money. Google’s ecosystem of applications and services that comprise its full stack of technologies, and which helps it develop, run, and distribute its AI models, makes long term competition with Google a Herculean effort.
While Google had an early lead in AI research, it was caught flatfooted by ChatGPT’s unprecedented launch in November 2022, which surprised the rest of the world as well. ChatGPT’s virality sent Google into code red. ChatGPT was widely seen as the fastest growing piece of software ever, which resulted in the major milestone of reaching one billion monthly active users in early August this year. Though, external data suggests that it may have crossed that line in June.
However, after Google launched Gemini 3 in November 2025, it reconquered that ceded digital territory from OpenAI. Five days after OpenAI announced its milestone, Google announced that Gemini had also reached a billion active monthly users.
Indeed, Google now splits control over the market for AI foundation models with OpenAI and the other leading AI company, Anthropic. In addition to offering chatbots to consumers, these companies’ foundation models serve as digital infrastructure for third-party application developers, which access the models via API. When analyzing the current AI model market by API revenues, one estimate finds that Anthropic, OpenAI, and Google comprise 88% of API revenues, with Google commanding 21%, as of December 2025.
But Google controls products and services that OpenAI and Anthropic do not and which give Google a leg up as these companies continue to battle for market share. Specifically, Google’s control of distribution channels throughout its tech stack explains much of Gemini’s ability to catch up to ChatGPT and Anthropic’s Claude and why it could outstrip them, even if the dynamics seem more like a horse race today.
ChatGPT was new to the world in November 2022 and gained popularity from online buzz. When Google wanted to get its AI systems into the hands of users, it simply added AI Overviews atop Google Search results, reaching billions immediately. As of May 2026, over 43% of searches through Google Search returned an AI Overview, summarizing results and letting a user turn that summary into the start of a chat, akin to ChatGPT.
Other Google products, like Gmail, Maps, Drive, and Photos, have their own significant audiences, and Google has integrated Gemini tools there too. These apps encourage users to continue to use Google’s suite of services and also give Gemini unique data to train on.
Beyond controlling distribution channels, Google also benefits from being vertically integrated across the AI tech stack. An end user could access applications like AI Overviews in Google Search, powered by Google’s Gemini foundation models, which are built and accessed using Google Cloud infrastructure that includes Google TPU chips. But this is the simplified version of the stack. As our filing shows, an expanded version of the AI tech stack includes dozens of distinct product categories where Google is active. In a less simplified AI tech stack, a user accesses AI Overviews in Google Search in Chrome on a Pixel smartphone running Android’s operating system connected to a Nest home router accessing the internet through GFiber broadband that connects to Google Cloud data centers full of Google TPU chips, powered by Google Energy. Neither OpenAI nor Anthropic has access to this many resources across the AI tech stack, and thus fewer points to reach users and their data or achieve efficiencies.
Google also vertically integrates via financing, creating a win-win scenario for the company. Consider that Google is both among the largest investors in Anthropic and one of its largest suppliers of compute. More recently, OpenAI has also started using Google Cloud, among other providers, for its AI compute needs. This blunts the competitive market pressures that might discipline Google because it is not just competing for the AI market. Google’s Gemini might become the market winner in AI based on its Search monopoly advantage. If instead Anthropic or OpenAI end up winning the market, Google still benefits from AI companies’ need for cash and compute. Google wins either way.
In the United States and European Union, Google has been found to violate antitrust laws in cases on digital advertising, app stores, mobile operating systems, and search results with integrated shopping links. The problem with these cases is that, so far, they have relied on fines and requiring actions like Google sharing information with rivals. That clearly doesn’t work. When Gemini hit a billion monthly active users, it became the 14th Google product to cross that threshold. While product success is not a violation of antitrust laws, it’s Google’s anticompetitive tactics to maintain these monopolies that violate the law. As another amicus brief in the Google remedies appeal case describes, Google has a playbook of monopolizing a market and using its monopoly to reinforce power in other markets.
This is why we and so many others believe it is time to break Google up into smaller parts, whether that be the search service that underpinned its AI development or other products.
Judicial humility
Finally, while the Department of Justice sought major structural remedies, like forcing Google to divest Chrome, Judge Mehta opted to grant only modest, incremental remedies to the unlawful monopoly, because judges should exhibit “a healthy dose of judicial humility.” However, as we show, the humblest judicial posture is to apply and follow the law, not to forecast AI’s technological trajectory. American antitrust doctrine has never found that a monopolist entering a distinct, emerging market, where that firm has not yet achieved the same market power, excuses that firm’s unlawful maintenance of a monopoly in an older market. As the Federal Trade Commission described in a case against Meta, the only supermarket in town doesn’t stop being the only supermarket because it started selling pet food and now competes with Petco.
Consider what the remedy actually permits. Judge Mehta declined to bar Google’s payments to Apple of roughly $20 billion a year for default placement in Safari, requiring only that the arrangement not be exclusive, a condition easily satisfied. Four months later, in January 2026, Google and Apple announced that Gemini would power Siri. That’s the same tactic, same companies, same objectives: buy distribution before rivals can. Only this time it is in the market the court predicted would discipline Google.
In the coming months, the D.C. Circuit will consider an appeal of, in our view, the lower court’s error in letting Google off the hook with a slap on the wrist for unlawful actions to maintain its monopoly. We hope the appeals court notices that digital markets have long been under a code red, reflecting how extreme levels of concentration that hinder competition.
Authors’ Disclosures: The authors report no conflicts of interest. You can read our disclosure policy here.
Articles represent the opinions of their writers, not necessarily those of the University of Chicago, the Booth School of Business, or its faculty.
Subscribe here for ProMarket’s weekly newsletter, Special Interest, to stay up to date on ProMarket’s coverage of the political economy and other content from the Stigler Center.
