This blog is the latest coverage in a series on Artificial Intelligence and the Google cases. You can read more of our coverage, including on Artificial Intelligence and the Google cases here, here, here, here, here, and here.
As this blog has previously noted, the Google Search remedies decision represents a watershed moment for antitrust. But while the underlying case itself concerned only Google’s dominance in general search, the remedies decision signaled a keen interest in artificial intelligence (as we anticipated), to ensure Google’s “dominance in search does not carry over into the genAI space.”
This concern is not theoretical. Alphabet has strong positions for key inputs for AI products, including data and cloud computing. Moreover, many worry about a “flywheel effect” similar to Search; this allowed Google to use user interaction data from queries to develop a better search product, thus attracting more user interaction to build an ever-better product. The flywheel effect is itself not anticompetitive (in fact it improves product quality). But, it did give Google durable power to block competitors from building better alternatives.
As a result, a central theme of Judge Mehta’s Order was to ensure that Google’s competitors have access to key inputs like data. However, Judge Mehta rejected proposals that would prevent Google from aggressively competing in the AI space, effectively giving Google the green light. Google took the note, and has already moved aggressively to shape the burgeoning market. Other AI market participants are already sounding alarm bells as a result.
Building One Market on Another
Given its playbook in Search, it is not difficult to imagine how Google could leverage its dominance to exclude rivals in genAI. One could picture, for example, how exclusive dealing and default arrangements could deprive rivals, like ChatGPT, of key training data.
Landmark section 2 cases, from AT&T to Microsoft, had similar concerns. There, the companies’ monopoly power in the original market (telephone service, operating systems) allowed them to block rivals in different markets built on top, such as preventing interconnection of ancillary devices like modems, and default arrangements for web browsers. Modems and browsers, of course, soon became critical gateways for the Internet. It is no exaggeration to say that freeing these ancillary markets from monopolistic control was central to digital innovation in the 1990s and 2000s.
Perhaps learning from history, Judge Mehta’s decision seems directed at this concern. His remedies order requires Google to share categories of search data, barred exclusive contracts, and authorizes syndication to rivals. This allows Google’s rivals in Search—and AI—to access critical inputs to improve their models.
However, Judge Mehta controversially rejected more intrusive remedies, including the divestiture of Chrome and Android, and outright prohibitions on Google investing in the AI space. Judge Mehta also permitted—for now—Google to make payments to Apple and Samsung for default status, so long as they were not exclusive (reserving its rights to revisit). And the Court rejected proposals to stop Google from “self-preferencing” its own products.
As a result, critics argue that his remedies order does not go far enough to curtail Google’s monopoly power, and worry that Google now has a playbook to dominate genAI in the same way it dominated Search. But Judge Mehta cautioned that although he sought to stop Google’s illegal conduct, he did not intend to block Google from aggressively competing in the emerging market.
GenAI: What’s Next?
The market is quickly evolving, even beyond the famous chatbots. One example is product integration, combining non-public sources of data with new AI tools. This provides AI tools with crucial training data, and allows legacy companies to offer AI products to their customers. Recent announcements include integrations between Apple and ChatGPT, and Microsoft with Harvard Medical School.
Google has not ceded any territory. Just in the last month, Google launched a multiyear strategic partnership with PayPal to “combine their digital commerce capabilities and develop AI-based shopping experiences.” The companies intend to develop standards for agentic commerce, to allow AI agents to make transactions on behalf of users, and will integrate PayPal Payouts into products like Google Cloud, Google Ads and Google Play. The deal also means Google Cloud will undergird much of PayPal’s technological infrastructure. Beyond PayPal, Google has made similar deals with Klarna (another digital bank and payments provider), Figma (design software), Salesforce (CRM software), Atlassian (collaboration and productivity software), Oracle, and Samsung’s companion robot.
One could easily imagine how a more aggressive Remedies Order would have inhibited these integrations. For example, a broad restriction on Google’s ability to pay for default status could have restricted, or at least chilled, these deals. But the Court didn’t choose this path. Perhaps this choice was prescient – these innovative product developments have suddenly and dramatically expanded the scope of AI tools, a seemingly procompetitive result. Vigorous rivalry, including from legacy tech firms and new entrants alike, appears to reflect a market not currently dominated by one single firm.
Nonetheless, some have questioned whether these agreements could raise competition concerns. As the terms are confidential, it is unclear whether any of them are technically “exclusive.” But the deals certainly give at least preferred access to training data, and shift user interaction to the default AI tool. As a result, it is not a stretch to see how a firm with monopoly power could use integration deals to lock up sources of training data and foreclose rivals. (Elon Musk’s company xAI has already accused Apple and ChatGPT of doing just that). And it’s not just the data. OpenAI has grumbled to European enforcers about “lock-in” to legacy Big Tech platforms. OpenAI reportedly highlighted the control that Google and Microsoft have over key infrastructure, including cloud computing, to keep customers locked into their ecosystems, preventing competition in AI.
Another area where the Remedies decision has already had a direct impact is Google’s internal integration. Judge Mehta’s rejection of a “self-preferencing” ban freed Alphabet to leverage its wide portfolio of products to improve and expand AI offerings, giving it privileged, and almost certainly exclusive, opportunities to train its models.
One example is Google Overview, which offers an AI response to a Search query on the results page. Although many users seem to like the offering, publishers complain that Overview reduces clicks because the underlying information is simply regurgitated by the AI tool. This has already led to antitrust complaints in the United States and Europe, whereby content creators claim that Google has illegally tied access on the Search results page to the ability of Google to train its model on the underlying content.
Similarly, Judge Mehta’s refusal to order the divestiture of Chrome and Android has allowed Google to integrate with those products also. Given that Chrome has 71% of the global market for browsers, and Android has 75% of the global market for smartphones, these giant pools of training material are invaluable to developing Google’s AI products. And of course, Google has a many other important products as well: Gemini already integrates with Gmail, Maps, and YouTube.
ChatGPT has thus far been the most famous consumer-facing genAI tool. But Google’s integrations internally and externally mean that users are very likely to use Gemini in the near future—if they do not already. This is a direct (and apparently intended) result of Judge Mehta’s Remedies Order. However given the strong position Google has over training data and cloud computing, antitrust enforcers will almost certainly be hearing more on the issue in the years to come.
