The Antitrust Attorney Blog
The Gate to the Kingdom May Be Closing for Good: Apple Puts Google Inside Siri
Author: Luis Blanquez
On January 12, 2026, Apple and Google announced a multi-year deal—reported at roughly $1 billion a year—under which a custom Google Gemini model will run as the backend “brain” of the rebuilt Siri and the next generation of Apple Intelligence. Apple plans to unveil the new assistant at WWDC on June 8, 2026, and ship it with iOS 27 in September. Apple’s own on-device Foundation Models will handle simpler, privacy-sensitive tasks. Heavier queries will route to Gemini on Apple’s private cloud. Siri’s cognitive core will no longer be Apple’s. It will be Google’s.
Meanwhile, ChatGPT—the model Apple spent 18 months positioning as the future of iPhone AI, and the one still at the center of the xAI antitrust lawsuit pending in the Northern District of Texas—continues to sit where it has always sat: in a separate, opt-in, off-by-default layer that users have to turn on themselves. In iOS 27, Apple will expand that layer into an “Extensions” framework that lets users pick from Claude, Grok, Copilot, Perplexity, and a user-selectable Gemini chatbot app alongside ChatGPT.
On the surface, the Gemini deal looks like Apple conceding ground—admitting it cannot build a frontier AI model of its own and paying Google to fill the gap. It is not. Apple does not fear a better chatbot. It fears losing the user to one. By routing Siri’s cognition through a supplier Apple controls contractually, it keeps the interface, the invocation, the defaults, the billing, and the brand—everything that makes an operating system a platform—while outsourcing the only layer it never wanted to own: the AI model. It is classic Apple, but this time with Google as an ally instead of a threat.
Where The xAI Case Stands
Elon Musk’s X Corp. and xAI sued Apple and OpenAI on August 25, 2025, in the Northern District of Texas.
The theory of harm has two layers. First, ChatGPT’s integration into iOS, iPadOS, and macOS—announced in June 2024—delivers billions of iPhone-originated prompts to a single model. Second, the complaint alleges that Apple deprioritized rival chatbots in App Store rankings, making it harder for Grok and others to reach users even outside the system layer.
In late September 2025, Apple and OpenAI moved to dismiss. Apple argued the deal is “expressly not exclusive.” OpenAI called the suit part of a “lawfare” campaign. Judge Mark T. Pittman denied both motions on November 13, 2025, in a one-page order signaling the case is “more well-suited for adjudication through a motion(s) for summary judgment.” Discovery opened on October 10, 2025, and closes on May 22, 2026.
Two rulings tell us the real story.
The source-code ruling: Courts don’t want to become engineers, they just want evidence of platform foreclosure
On January 22, 2026, Magistrate Judge Hal R. Ray Jr. denied X’s motion to compel OpenAI to produce ChatGPT’s source code. X argued it needed the code to rebut any defense of lack of feasibility, to establish whether Apple could integrate Grok into the Apple iOS, and whether it would be feasible to integrate multiple AI products from various providers on the iPhone. The court disagreed on both relevance and proportionality.
On relevance, the order is blunt: “At this stage, it is unclear how the source code is relevant to whether Apple unlawfully excluded Grok from its products and conspired with the other defendants to create a monopoly.” On proportionality, the court noted that OpenAI had already produced API documentation and offered to stipulate that neither side would rely on proprietary source code. X refused the compromise. The court refused the production.
For a platform-exclusion case, that ruling is not a setback. It is a roadmap. Courts do not want plaintiffs dissecting competitor IP to infer what could have happened. They want direct evidence of platform misconduct—integration terms, routing logic, default settings, App Store treatment. That is where antitrust liability lives in a case like this. The court said so in plain terms, and future plaintiffs should take the hint.
The April 2026 motion to compel: Valuation and training data
X filed a second motion to compel on April 10, 2026, this one aimed squarely at OpenAI’s document production. The motion accuses OpenAI of having “persistently evaded its discovery obligations,” noting that OpenAI had produced 7,475 documents compared with X’s 32,116. The requested materials include valuation studies, financial projections through 2030, ChatGPT usage by country, and documents about OpenAI’s use of X data to train its models.
The valuation request is the one to watch. X tells the court that OpenAI’s valuation grew by more than $500 billion in the past year—from $300 billion at the end of March 2025. X argues the jump “does not make sense unless OpenAI and its investors understand that ChatGPT’s exclusive deal with Apple is cementing OpenAI’s dominance.” That is a causation theory built on investor behavior, not technology.
The training-data request matters for a different reason. OpenAI has defended the case in part by arguing that Grok’s foreclosure from iPhone-originated prompts is insubstantial because xAI has asymmetrical access to X data. X’s response is direct: if OpenAI also trains on X data, the asymmetry disappears. That question—who gets to learn from where—sits at the center of every theory of harm in generative AI.
Both motions, taken together, ask the court to analyze platform foreclosure through evidence: deal terms, integration architecture, valuation impact, data flows. None of it is specific to OpenAI nor depends on getting inside OpenAI’s code.
That same framework applies to Gemini.
Gemini Isn’t Apple’s Competitor. It’s Apple’s Supplier.
Apple’s decision to build Siri on Gemini at the backend, while offering third-party chatbots through Extensions on top, reshapes the competitive picture more than any filing in Texas has.
Apple did not retreat from AI. It relocated AI. Every layer a platform needs—distribution, defaults, billing, invocation, user interface, identity—remains Apple’s. When a user says, “Hey Siri” and Gemini answers, Apple delivers the answer. The supplier is invisible. Apple owns the user relationship.
That is the Google Search playbook applied one floor up in the stack. The existing Apple–Google deal to make Google the default search engine in Safari has existed for over a decade, with payments roughly reaching $18–$20 billion a year in the early 2020s. Google delivered the search results. Apple kept the browser, the UI, the defaults, and the customer. Nobody ever thought of Safari as “the Google browser.” That deal survived the U.S. v. Google antitrust trial not because it was invisible, but because it proved how much Apple gained by making a dominant rival pay for access instead of competing on equal terms.
The Gemini arrangement reuses the same design. At the foundational layer, Gemini is Siri’s cognitive backend—contractual, invisible, and not subject to user choice. At the visible layer, the iOS 27 Extensions framework lets users swap among ChatGPT, Claude, Grok, Perplexity, and a Gemini chatbot app for certain requests. Apple will point to that second layer as evidence of openness. But the first layer is where platform control actually lives. Whatever model a user picks through Extensions runs on top of an infrastructure Apple has already built around Gemini. Apple does not compete with Gemini. Apple neutralizes it by absorbing it. History doesn’t repeat itself, but it often rhymes.
Forget the App Store. This is Microsoft All Over Again
The App Store line of cases is a distraction here. The real analogue is the Microsoft browser and search monopolization litigation, both in the United States and the EU.
In those cases, Microsoft did not win the browser wars because Internet Explorer was better. And it did not win search because Bing was better. Microsoft used the operating system as a default-distribution layer—embedding its browser and search entry points directly into Windows, controlling invocation and placement, steering user behavior at scale through defaults and friction. Microsoft’s core defense was that users could still download alternatives. Courts rejected that defense. Default placement and UI control at the moment of choice shape markets more than theoretical availability does.
Apple’s Gemini integration tracks that logic almost perfectly, but at a more consequential layer. Apple is not merely preinstalling an assistant. It is controlling the cognitive interface itself, the point where users express intent, ask questions, and issue commands. Siri is the AI-era equivalent of the Windows Start menu or the browser toolbar. And Apple is making sure it stays that way whether the underlying model is ChatGPT, Gemini, or whatever comes next.
The most important doctrinal lesson from Microsoft applies directly here: supplier substitution does not defeat monopoly power. Microsoft rotated technologies, features, and partners over time, and courts still found monopoly maintenance because Microsoft never gave up the interface layer. Apple’s move from ChatGPT to Gemini is the same pattern. Apple did not open the platform. It reaffirmed its control. In antitrust terms, the fact that Apple can swap AI “brains” without changing the user experience strengthens the inference of monopoly power—it shows that the binding competitive constraint is not OpenAI or Google, but Apple’s control over defaults, invocation, UI, and routing.
Apple’s “We’re Not Exclusive” Defense Will Age Badly
Apple has told the Texas court that its ChatGPT deal is “expressly not exclusive” and that it plans to partner with others. The iOS 27 Extensions framework is how Apple plans to prove it. In that framework, users can plug ChatGPT, Claude, Grok, Copilot, Perplexity, or a Gemini chatbot app into Siri for certain requests. Apple will present that menu as the answer to any foreclosure claim.
That defense tracks Microsoft’s argument that alternative browsers were “available.” It works only if formal choice produces functional parity. But here it does not. Apple controls the default assistant, the invocation layer, the UI surface, the routing of tasks between Apple Intelligence and third parties, and the subscription economics through the App Store. A user who says “Hey Siri” gets whatever Apple decides to serve—from a cognitive engine already running on Gemini. Extensions sit on top of that engine. The choice among chatbots happens downstream, inside Apple’s frame, on Apple’s terms, and with no effect on the foundational model Apple has already selected. In other words, because Apple controls iOS’s integration chokepoints, it can privilege Gemini with first-class Siri/Apple Intelligence hooks while boxing rivals like OpenAI, Claude, or Grok into higher-friction, permission-gated, sandboxed “extensions” that can’t match system-level features.
The January 22 ruling in the xAI case fits the same pattern from a different angle. Feasibility, in a platform case, is about platform access on equal terms—not about whether a rival’s code could technically run somewhere. That is why the court sent X back to integration evidence. It is also why the “expressly not exclusive” defense will not do the doctrinal work Apple needs it to do. The relevant question is whether an AI model rival can reach a user without going through Apple’s interface. Under the current architecture, it cannot.
The Next Wave of Apple Antitrust Claims
The Gemini pivot does not close the xAI case. It widens the set of plaintiffs who have a theory.
An AI model company that powers Siri becomes a commodity. Its brand disappears. Its direct relationship with the user disappears. That is true for any model Apple selects next, whether OpenAI, Anthropic, xAI, or a company that does not yet exist. AI-native startups lose their distribution path. Vertical AI companies face Apple’s historical “Sherlocking” risk—once a feature proves useful, Apple absorbs it into the operating system. Device competitors without comparable partnerships face lock-in of their own. Content and app platforms whose users reach them through Siri lose direct control of the customer relationship.
Many of these companies will not see exposure until Siri/Apple Intelligence is the default entry point for user intent on more than a billion devices. By then, Apple—not the AI model, not the app—owns the interaction.
The theory of harm in each of these cases is Microsoft, not the App Store. It is not about commissions. It is about defaults, invocation, UI control, and routing—the chokepoints that decide whether competition can reach the user at all. The xAI case is the first to test that framework in the AI context and will not be the last.
What Companies In The AI Ecosystem Should Be Doing Now
Three things are worth doing in 2026, before the next round of filings.
First, map your exposure to Siri/Apple Intelligence as infrastructure, not as a platform partner. If Siri’s routing logic can send user intent to an Apple-selected supplier instead of your product, you have a platform-foreclosure risk whether or not Apple names a model that competes with yours.
Second, preserve the evidence of foreclosure the court in Texas has already told us matters: integration terms, API documentation, default behavior, routing tests, any communications with Apple about access, and any data reflecting pre- and post-integration user behavior. The January 22 ruling made clear that platform conduct evidence—not competitor code—is the currency of these cases.
Third, take the “formal choice” defense seriously. If Apple offers you access through an Extensions framework, the question that matters is whether that access produces functional parity at the point of user invocation. If it does not, your competitive position is worse than it looks, and your antitrust position is stronger than you think.
The xAI Case Is Just a Preview
The discovery fights in Texas will shape how platform foreclosure claims are tried in the AI era. But the xAI complaint was filed before the Gemini pivot. It was built around a specific ChatGPT integration and a specific exclusivity theory. The actual competitive problem—Apple consolidating control over the AI interface layer by absorbing rival models into its own stack—is bigger.
Apple did not give up ground when it picked Gemini. It locked it in. Under Microsoft precedent, and under the framework the Texas court is already applying, that is where the next generation of antitrust litigation will be fought. Apple is building the wall now. Companies whose access to customers run through Siri/Apple Intelligence can start building their case, or they can wait and explain to their board why they let Apple turn Siri into iGoogle.