“Google Ruling Shows Antitrust Is Dead,” a Barron’s headline trumpeted. Perhaps a bit hyperbolic, it nonetheless reflects the inadequacy—especially in the dynamic artificial intelligence age—of relying on antitrust policies designed for the relatively static industrial age.
The court’s recent decision in United States of America et al v. Google LLC highlights the need for Congress to step up to its responsibility to define the relationship between AI and a competitive marketplace—quickly.
The background of the Google case
The U.S. Department of Justice, joined by a bipartisan group of 11 state attorneys general, filed suit in 2020 alleging that Google search was a violation of the Sherman Antitrust Act. After a 10-week trial, U.S. District Court Judge Amit Mehta ruled in 2024 that “Google is a monopolist, and it has acted as one to maintain its monopoly.” The case then proceeded to its second stage, where the court considered how best to remedy this situation.
The Department of Justice argued in favor of structural solutions, a traditional antitrust remedy. To stop the perpetuation of monopoly, the government proposed restructuring Google’s activities. This included undoing the defaults that make Google the search engine of choice for most browsers, including the purchase of exclusivity on devices, such as the 2022 payment of approximately $20 billion to Apple to make Google the default on iPhone. The Justice Department also raised the possibility of making the market more competitive by divesting Google-owned feeders to search, such as the Chrome browser or the Android mobile operating system.
Instead, the court imposed what Barron’s described as “almost a best-case scenario for parent Alphabet.” Rather than the requested structural solutions, the decision called for a series of behavioral requirements for Google. These included banning exclusive search deals for default placement (although allowing payments for non-exclusivity to remain) and requiring Google to share with rivals on “commercially reasonable terms” some, but not all, of the search results data that powers Google. To assist the Department of Justice in overseeing compliance, the court ordered Google to establish an independent technical committee.
The CEO of rival search engine DuckDuckGo, Gabriel Weinberg, called the court’s decision a “nothingburger.”
AI changed the game
The emergence of generative AI “changed the course of this case,” the judge wrote.
Between the time the lawsuit was originally filed and the judge’s decision, large language models (LLMs) moved out of the laboratory and became broadly accessible. It was a development that had a significant impact on the judge’s remedies ruling.
Thanks to AI, the court found online search had become competitive seemingly overnight. AI companies are now in a better position “to compete with Google than any traditional search company developer has been in decades,” he ruled.
What is the search market?
The arrival of AI has fractured online search into at least three identifiable markets. First is the traditional search market, where the user enters a query and receives a list of websites. As of August 2025, Google search had an 89.89% market share of the worldwide search engine market. Microsoft’s Bing was second with 3.92%. Both levels have been stable over the preceding 12 months.
The second category is AI search, where the query returns a summarized answer drawing from the top links. Such services include Google AI Overviews, ChatGPT, Perplexity AI, and Bing Copilot. A December 2024 study found Gemini and ChatGPT capturing 78% of all AI search traffic. OpenAI’s filing with the court suggested its AI search activities slightly exceeded those of Google’s overviews.
The third category is what apparently caught most of Judge Mehta’s attention: GenAI search. Such conversational agent-like dialog tools include ChatGPT, Claude, and Google’s Astra. A June 2025 study found that GenAI traffic was growing 165 times faster than organic search, yet it still accounted for less than 1% of total website traffic. Another study found Google search grew by over 20% in 2024, handling over 5 trillion searches—approximately 14 billion per day—a total that is 373 times bigger than ChatGPT search.
Gazing into a crystal ball
“[U]nlike the typical case where the court’s job is to resolve a dispute based on historic facts, here the court is asked to gaze into a crystal ball and look to the future,” Judge Mehta wrote.
Whether access to a limited amount of Google’s search data will make traditional search competitive is indeed a crystal ball issue. It is not difficult, for instance, to imagine Google prevaricating and procrastinating over just what data is covered; after all, each day of delay delivers not only the benefits of 14 billion more searches, but also the ability to use Google’s dominance to disadvantage competition in AI search and GenAI.
Judge Mehta’s ruling walked a tightrope between Google’s behavior in the past and the potential impact of a new technology on its future behavior. It is the essence of the competition policy challenge in an era of rapid-paced, AI-driven change. As well-intended as Judge Mehta’s decision may be, it is the reason why there is a need to move beyond trying to use antitrust litigation for behavioral outcomes.
The competitive dynamic of AI is beyond the vision of anyone’s crystal ball. The vagaries of such crystal ball forecasting emphasize the need for clearly delineated AI competition policy that is broader and more instructive than antitrust policy—and the reason why there is a need for risk-based and agile behavioral standards to promote and protect a competitive AI marketplace.
Antitrust enforcement is important but inadequate
Antitrust is an important tool in protecting a competitive marketplace. In the AI era, however, it cannot be relied upon as the only tool.
As Judge Mehta wisely and humbly observed, keeping pace with technological change and its impact on the market is “not exactly a judge’s forte.” The complexities of antitrust cases, in fact, are exceedingly rare in a federal judge’s career. A 2012 study estimated that, “[i]n each of the past five years, antitrust cases accounted for less than half of one percent” of all civil filings. Combining this relative lack of experience for even the most talented jurist with the economic and technological complexity of the issues in such cases ensures that decisions are inherently uncertain.
Antitrust cases are also reliably lengthy, as the Google case itself exemplifies. The case was filed in October 2020, challenging two decades of alleged bad behavior by Google. The court issued its monopoly decision almost four years later in August 2024. The remedies decision required another year. Four years from filing to an initial decision, followed by an additional year for the remedies ruling, is not a criticism of Judge Mehta, but a recognition of the complexity of such litigation.
And this isn’t the end of the process; subsequent appeals all the way up to the Supreme Court of the United States, where antitrust law is ultimately made, probably mean there will be no final decision until possibly 2027 or 2028. Such delay is an eternity in the exponential pace of digital technology, as new technologies change the landscape of the marketplace, including the potential of making the initial complaint moot.
Antitrust law, by design, is an after-the-fact review of past actions. The Sherman Act, which Google was found guilty of violating, was designed to be backward-looking. Section 2 of the act makes it illegal to “monopolize, or attempt to monopolize, or combine or conspire…to monopolize any part of…trade or commerce.” When the court found Google “has violated Section 2 of the Sherman Act,” it was a decision about Google’s past practices. Then the court rejected the government’s recommended remedies on the grounds that they “overreached in seeking forced divestiture of these key assets, which Google did not use to effect any illegal restraints.”
The problem with relying solely on antitrust enforcement to address the competitive challenges of the AI era is directional. While antitrust is designed to eliminate illegal past practices, as Judge Mehta’s opinion demonstrates, it is not a vehicle for the promotion of competition going forward.
The need for forward-looking AI competition policy
The week before arguments began in the remedies portion of the case, Google began to reposition the scope of the discussion from looking at past activity to looking toward the future. In an April 20 blog, Google described the lawsuit as “a backwards-looking case at a time of intense competition and unprecedented innovation” that would “hurt America’s consumers, economy, and technological leadership.”
There is, however, no established forward-looking competition policy for the AI marketplace. Continued reliance on antitrust statutes means continued reliance on the mitigation of already existing harms rather than the establishment of policies that would encourage innovation through the protection and promotion of competition going forward.
Competition policy is about more than anti-monopoly. A goal of public policy should be the ex ante promotion of competitive behavior, not just the ex post redress of its absence. This means augmenting antitrust’s backward-looking, company-specific, and behavior-specific litigation with forward-looking regulatory oversight broadly applicable to the dominant providers of services.
The companies seeking to avoid such oversight frequently claim that regulation hurts innovation and investment. They are correct—except that they are referencing old-style industrial micromanagement rather than a new form of oversight for a new era. The 21st-century regulatory model must be one of protecting the public interest while promoting the expansion of innovative advancements.
In place of top-down regulatory micromanagement of old utility-style regulation, the AI era requires agile risk management. This means replacing utility-style regulatory mandates with a new oversight model that focuses on competitive market outcomes using risk-based and agile oversight of expectations, not the regulatory dictation of management practices.
In this regard, Judge Mehta’s decision was directionally on course for as far as he thought the law would let him go. His effort was to fundamentally alter competitive market dynamics by addressing the behavior of the dominant company regarding its control of an asset necessary for rivals to compete. But addressing a behavior from 2020 by embracing something that didn’t exist at the time begs the question as to a going-forward solution for protecting that forecasted AI competition.
The court did very little to assure that GenAI would itself remain a competitive marketplace capable of providing its hoped-for solution.
What is needed for there to be a competitive AI marketplace is a similar—but forward-looking—behavior-oriented set of expectations for the essential assets of AI. The data that populates LLMs is, of course, one of those assets—but the ruling does little to overcome the control of this essential asset by the dominant AI companies. So is access to another essential AI input: computing power. It is no accident that the Big AI companies, such as Google, Microsoft, and Amazon, are also the three largest cloud computing platforms.
Perhaps Judge Mehta’s decision will fundamentally alter the competitive landscape for search. For it to be successful, however, requires there to be a competitive GenAI marketplace. For that market to exist requires a forward-looking policy establishing the expectation of open and fair access to the inputs necessary for AI innovation and diffusion. To ensure that the court’s decision will not just kick the competition can down the road requires acting today to establish policy that protects AI competition going forward.