Last September, US Judge Amit Mehta issued his judgement in United States v. Google and backpedalled on the much-awaited remedies sought by the Department of Justice involving Google’s divestiture of its Chrome browser. Despite the backlash on that front, that same decision imposed several new limits on the exclusive contracts that Google held with its competitors, see Apple and Mozilla, to install its proprietary apps on non-Google platforms.
The most salient contract was that between Apple and Google, where the former installed Google Search as the default search engine on Safari for all its devices in exchange for taking a cut of Google’s ad revenue on Safari and Chrome. In 2022, Google’s revenue share payment to Apple was an estimated $20 billion, which was equivalent to 17,5% of Apple’s operating profit in 2020. In early January, Apple and Google announced they entered into a multi-year collaboration under which the next generation of Apple foundation models will be based on Google’s Gemini models and cloud technology, to power its future Apple Intelligence features. The parallels between both cases are striking: both economic agents prefer to keep competition out and cut their pieces out of the digital cake.
A ground-breaking agreement that stifles innovation
The joint statement from Google and Apple does not provide much detail, insofar as it remarks that both companies will collaborate to integrate Google’s Gemini models into Apple’s foundation models and Apple Intelligence features. However, it does point to a salient aspect: Apple renounces to participate in the AI race entirely, as it had already let on whilst other competitors such as OpenAI, Meta, or Anthropic had advanced enormously in the powering of their LLMs.
Apple had been developing AI-related tools and features since 2015 with its Project Titan, which involved the very ambitious goal of manufacturing an autonomous driving unit running on AI. It’s not that Apple had given up on winning the AI race before it had even started, but rather that it was slow in adapting to the fast-paced rhythm of other competitors when they were launching their own LLMs and embedding them into their existing services. This set of circumstances led Apple to believe that it had to depend on others to launch its AI features in a reliable, sound, and safe way, abiding by its compromise to keep its proprietary devices as sophisticated and functional as possible. That’s the reason why it entered into negotiations with Anthropic, OpenAI, and Google to boost its features. To accelerate their delivery, it finally settled with Google, as it considers Gemini to be the most suitable LLM to provide the foundation for its AI models.
As ever, the devil is in the details. We will have to wait to learn about the nitty-gritty of Apple’s partnership with Google and whether it bears any resemblance to its previous exclusive contract over search defaults. As a matter of fact, the DOJ’s remedy order in the Google search case explicitly prohibits Google from entering or maintaining exclusive contracts relating to the distribution of the Gemini app.
At this point in time, it seems that the deal might bring more pains than gains to both players, since it seeks to reinforce Google’s and Apple’s access to user data that powers the LLM’s feedback loop whilst stifling innovation in AI deployment and development. Anything that Google integrates into Gemini, Apple’s own LLM will follow, and this simple step may potentially undermine consumer choice, reinforcing Google’s market power in the provision of its AI models.
Informal exclusivity is powering potential anti-competitive effects
Under antitrust rules, exclusivity provisions in contracts may serve many useful purposes, and not only an anti-competitive goal. Monopolisation (or abuse of dominance) allegations can fructify in the presence of market power. One cannot simply assert that either Google or Apple dominate the market for the provision of foundation models, since Microsoft is the economic player ahead, and the competitive rationale underlying the market is characterised by strong and dynamic competition (Schrepel & Pentland, 2025). A monopolisation claim against any one of them would, therefore, not hold enough ground because they cannot act independently from competitors and their consumers in the current configuration of the market.
Notwithstanding, the deal points to major challenges surrounding data advantages, content access, scale, distribution, and the setting of defaults that will force the competition authorities and regulators’ hand in, at least, having a look at the potential anti-competitive impacts of the deal. In the DOJ case, Judge Mehta already confirmed that exclusivity in contracts is not a strict and formal requirement, but rather an open-ended formulation that may capture more types of agreements than initially anticipated. All major distribution deals can be remarkably sticky in digital markets, and one must expect regulators to respond to the challenges, either via the application of ex post antitrust rules or ex ante competition policy instruments, such as the European Digital Markets Act or the UK’s Digital Markets, Competition and Consumers Act.
In the simplest of terms, the agreement between Google and Apple provides Gemini with access to more sets of eyes. AI agents are not yet there in terms of technological development, but they might be in a couple of years. And that’s when data access to the most prominent players in the market will make all the difference. Imagine how many users Google accesses via its suite of services on a daily basis, through its search, advertising, video, or email services. And now double it. That’s the number of additional users that Google will be able to get access to as a result of the deal: the whole user base of Apple’s proprietary devices.
AI models feed on this data, mainly through reinforcement learning from human feedback, and improve their performance significantly once they interact with humans. Apple’s integration of Gemini into its foundation models entails that it will immediately gain the acquired knowledge of Google’s proprietary LLMs, without the need to grow and iterate its AI models through time. In turn, Apple and Google will not compete in the market for foundation models, since there is no point in differentiating themselves from a rival whose service is fundamentally identical to their own. They may compete on other interrelated markets, in the spirit of co-opetition (Brandenburger & Nalebuff, 2021).
While this partnership provides Apple with a reliable shortcut to AI relevance and grants Google an unprecedented pipeline of user data from two billion devices, it risks replicating the stagnant default-biased landscape that Judge Mehta so sharply criticised in the search market. As regulators in the US and EU sharpen their ex ante tools, the central question remains whether a truly competitive market for foundation models can exist when the most vital distribution channels are locked behind multi-billion-dollar agreements.
