After months of leaks, code analysis in iOS betas, and veiled statements at conferences, Apple's agreement with Google Gemini is now official and is beginning to take technical shape. Far from being a simple "Siri now uses Google AI," the move marks a profound shift in Apple's strategy to catch up in generative artificial intelligence.
The Cupertino company has accepted something that until recently would have seemed almost unthinkable: relying on Google's Gemini processors and cloud infrastructure to build the next generation of its Apple Foundation Models (AFM). New Apple Intelligence features, and in particular a revamped Siri, will be deployed on this foundation, with the update expected to reach users later this year.
What exactly has Apple signed with Google Gemini?
According to joint statements and reports from media outlets such as Bloomberg and CNN, Apple and Google have entered into a multi-year collaboration agreement. Under this agreement, the next generation of Apple Foundation Models will be developed using the Gemini models and some of the cloud computing technology from the Mountain View giant. This isn't simply a matter of "plugging" Gemini into Siri, but rather a transfer of the base model and tools so that Apple can build its own systems upon them.
Leaks suggest that Apple has licensed Gemini 3 as a foundational model , the equivalent of a "raw brain" that concentrates the knowledge acquired by Google's AI through its training on the internet and other data. This model arrives without a personality or defined rules of behavior. From there, Apple applies its own post-training phase to adapt it to its ecosystem.
Several analysts describe this approach as a " harness " system : Google provides the dictionary, but Apple writes the story . In practice, the result will be Apple Foundation Models that use Gemini's architecture and power, but with a tone, limitations, and capabilities tailored to iOS, iPadOS, and macOS, without any visible "Powered by Gemini" label on interfaces like Siri.
On the financial front, neither party has disclosed the figures, although Bloomberg has reported that Apple may be willing to pay around $1.000 billion annually for the use of Gemini in this new phase. Apple sees this investment as necessary to accelerate the rollout of AI, which had stalled in recent years.
Gemini under the hood: how Apple's new Intelligence architecture will work
One of the key points of the agreement is that Apple doesn't want the experience to be perceived as simply "Siri outsourced" to external services. The company has designed a three-tiered architecture that combines local processing, its own private cloud, and, at the top, Gemini-based models integrated into its servers.
At the first level, on-device processing , small modeling units (SLMs) will continue to operate, capable of handling everyday tasks without leaving the iPhone, iPad, or Mac. This includes functions such as summarizing notifications, quickly rewriting text, and managing simple actions within apps. All indications suggest that these local models will be distilled versions of Gemini's technology, compressed to take advantage of the Neural Engine in Apple's chips.
The second tier is occupied by Private Cloud Compute , Apple's private cloud. When the device runs out of processing power, requests are sent to these specially designed servers, which feature proprietary hardware and a privacy-focused architecture. Before accepting each request, a cryptographic validation of the device is performed, ensuring that it is a legitimate and up-to-date machine, and guaranteeing that the data is neither stored nor used to retrain models.
The major innovation comes at the third level, where the power of Gemini comes into play . Instead of sending queries to Google's data centers, Apple integrates Gemini's models into its own servers within that Private Cloud Compute "bunker." In other words, Google's models run on Apple's cloud, under its security rules, and without the data ever leaving its perimeter.
Thanks to this design, Apple argues that it's possible to leverage the foundation of a powerful LLM like Gemini without exposing personal information to third parties. Google's model provides the cutting-edge AI logic, while Apple maintains control over the environment, post-training, and the final user experience in Siri and other Apple Intelligence features.
Privacy and data control: the key argument for Apple
From Europe and Spain, where data protection regulators closely monitor every AI development, one of the biggest questions was obvious: what about privacy ? The agreement announcing the use of Gemini reignited fears about Google's potential access to the data of millions of iPhone, iPad, and Mac users.
In its public statements, Apple has attempted to preempt this concern by insisting that Apple Intelligence will continue to run, by default, on Apple devices or in the Apple cloud . The use of Google models is presented as an internal component of the training and inference process, but without enabling a direct data channel to the search engine company's servers.
This approach aligns with the strategy the company has been advocating for years: minimizing the information leaving the device , processing it in encrypted form when necessary, and preventing the storage of identifiable histories. In this context, Gemini's role is limited to providing the technological foundation upon which Apple builds customized and controlled models.
Internal sources cited by the international press indicate that, by design, Google will not have access to individual queries or personal data generated through everyday use of the new Siri or other Apple Intelligence tools. For European users, who are particularly sensitive to the implications of the General Data Protection Regulation (GDPR), this point will be one of the most closely watched when the new features are fully rolled out.
Apple itself has emphasized that its strategy involves demonstrating that advanced AI doesn't necessarily mean relinquishing control of information to large external server farms. The combination of local models distilled from Gemini and processing in Private Cloud Compute is, in theory, the way to square the circle between power, performance, and privacy.
Siri at the Center: A Race Against Time Rescue
The reality that has driven Apple to this agreement is that Siri had been lagging behind assistants and chatbots like ChatGPT, Perplexity, and other systems based on large language models for years. Apple's assistant had become stuck in rigid interaction patterns, struggled to understand complex requests, and had little ability to chain tasks together.
The big bet to reverse the situation was presented at WWDC 2024, when Apple unveiled Apple Intelligence as a cross-platform AI suite integrated into iOS, iPadOS, macOS, and its main applications. There, they promised a Siri with personal context understanding, the ability to "see" what's on the screen, coordinated access to different apps, and hundreds of possible automated actions.
However, the leap forward showcased at that conference was slow to materialize. At the end of 2024, Apple insisted that Siri's most advanced capabilities would arrive "in the coming months ," while simultaneously launching other Apple Intelligence features such as image generation tools (Image Playground) and Genmoji. The timeline was starting to feel tight.
In March 2025, the company itself clarified its message and acknowledged that some of those features needed more time , reducing the promise to a "more personalized" Siri with a phased rollout throughout the following year. WWDC 2025 confirmed this feeling of delay: the assistant's visible improvement fell short of the expectations Apple had fostered.
According to Craig Federighi, head of software, a "version 1" of the new Siri was ready for release between December 2024 and spring 2025, but its launch was halted because it was deemed not to meet internal standards or user expectations. This delay is what ultimately paved the way for the Gemini upgrade.
Calendar, iOS versions, and rollout in Europe
The agreement with Google not only involves a technical change, but also affects Apple Intelligence's release schedule . Apple has become accustomed to using iOS 0.4 versions as major milestones in its plans, and all indications are that it will repeat this strategy.
Documentation and leaks point to iOS 26.4 as the version destined to most visibly debut Siri's "new brain ." This update would bridge the gap between the more limited capabilities of the first wave of Apple Intelligence and the broader rollout planned for iOS 27, where Gemini should enhance the ecosystem more comprehensively.
Looking at past releases, the iOS 18.4, 17.4, and 16.4 betas arrived between late January and February, with final releases around March or April. If Apple maintains this pattern, developers could begin testing the new Siri based on Gemini-derived models around the end of winter , with a general rollout to the public before summer.
In Europe, Apple Intelligence arrived somewhat later than in other markets due to regulatory concerns and compatibility issues with EU law. With the Gemini agreement now in place, the company insists that its operations in the EU will comply with the existing legal framework , maintaining local execution and private cloud deployments under its control. It is possible that some features will be rolled out in phases or initially in "public beta" mode while privacy and competition aspects are validated.
In any case, Apple's objective is clear: to have a reinforced and operational Siri before WWDC in June , so that it can present iOS 27 and the rest of its systems with a more solid AI base, tested on real devices, including in the European market.
Relationship with ChatGPT and other models: a multi-sided AI ecosystem
The agreement with Google does not replace the relationships Apple had already established with other AI players. At WWDC 2024, the company announced a collaboration with OpenAI to integrate ChatGPT into certain Apple Intelligence features. Starting in late 2024, Siri can delegate to ChatGPT those queries that require real-time information or specially crafted responses.
This delegation is optional, activated only with the user's explicit consent , and complies with Apple's privacy policies. In practice, when Siri is unable to resolve a problem on its own, it offers to use ChatGPT, making it clear that this involves calling an external service.
Craig Federighi had already indicated that the underlying idea was to build an environment where users could choose between different models according to their needs: one stronger in creative writing, another more oriented towards programming, or another focused on up-to-date information. In that context, he openly mentioned the possibility of adding Google Gemini as one of those options.
With the new agreement, Gemini goes from being a mere optional alternative to becoming the primary technical foundation upon which Apple will train its future Apple Foundation Models. Even so, the company has given no indication that it will break with OpenAI. Rather, it seems to be betting on a scenario where ChatGPT continues to answer certain queries and Gemini is integrated as the internal "engine" of the new Siri and other Apple Intelligence features.
Meanwhile, projects like World Knowledge Answers have emerged , an AI-powered answer search engine that would combine Siri, Safari, and Spotlight to offer summaries generated by natural language models. In these types of tools, the use of Gemini as a technical foundation aligns with Apple's ambition to compete in the search engine arena without simply replicating the traditional link-based results model.
Internal changes at Apple and market pressure
The decision to use Gemini cannot be understood without considering what has happened internally at Apple in recent years. The delay in the development of the new Siri led to significant changes in the organizational structure . In 2025, the company decided to remove direct responsibility for the assistant from the area headed by John Giannandrea, a star AI recruit from Google, and place it under the supervision of Mike Rockwell, involved in the development of Vision Pro, reporting directly to Federighi.
Months later, Apple officially announced that Giannandrea would be leaving his position, spending some time as an advisor, and retiring in the spring of 2026. Amar Subramanya took over as vice president of AI , with the goal of more closely aligning the artificial intelligence strategy with software and product decisions.
Meanwhile, Tim Cook had publicly acknowledged that entire teams were being reassigned to AI projects and that the company was open to acquisitions to accelerate its roadmap. However, visible results were not arriving at the expected pace, while competitors like Google, Microsoft, and OpenAI were releasing increasingly powerful and high-profile models.
For financial markets, the agreement with Google is seen as a sign that Apple has decided to rapidly close the gap in AI capabilities . Analysts like Dan Ives of Wedbush speak of it as a "springboard" for Apple's artificial intelligence strategy and a "major validation" for Gemini's technology. Not surprisingly, the announcement coincided with new market capitalization records for Alphabet, which surpassed $4 trillion in market value.
At the same time, Apple is confident that a much more capable Siri and truly useful AI features will help revive sales of iPhones and other devices after several years of more moderate growth. Analysts predict double-digit increases in units sold if the new experiences win over those who are on long upgrade cycles.
Europe, competition and the precedent of the Google search engine
In the realm of competition, Apple's agreement with Google Gemini inevitably brings to mind another landmark pact between the two companies: the one that made Google the default search engine on Apple devices . For years, that contract has generated billions of dollars annually for the Cupertino-based company and has been the subject of antitrust investigations in the United States and the European Union.
The US Department of Justice argued that these payments helped solidify Google's dominant position in online search. Despite this, a judge allowed the agreement to continue, although the debate about its impact on competition remains open and is likely to resurface as AI gradually replaces traditional search engines.
With Gemini, European regulators could focus on a new dimension: the risk of power concentration in the AI market . If the world's leading mobile operating systems—Android and iOS—rely heavily on models developed by the same company, the European Commission will want to analyze in detail what real room other AI providers will have to compete on a level playing field.
For now, Apple maintains that its approach remains one of an "open ecosystem" with regard to external models, and notes that it will continue to offer integration with services such as ChatGPT and, eventually, with other providers such as Anthropic or Perplexity if user demand justifies it.
In any case, the regulatory impact in Europe will be significant. How Apple implements Gemini within its private cloud, the transparency with which it explains data processing , and the degree of openness it allows to third-party models will be key factors in avoiding friction with Brussels and national data protection authorities.
What the user can expect: a different, but familiar Siri
Beyond technical documents, organizational chart changes, and regulatory debates, the big question for anyone who uses an iPhone, iPad, or Mac is simple: what will the agreement with Gemini actually change in their day-to-day lives? And, above all, when will they notice it?
In practice, the promise is a Siri capable of handling far more complex requests , better understanding natural language, reacting with more context, and coordinating actions between applications with less friction. Apple's idea is that the assistant will be able to read emails, messages, appointments, or documents and act on them more autonomously, in line with what some advanced chatbots already offer, but integrated directly into the system.
The user, in principle, will not have to decide whether their request is handled by a local model, Apple Foundation Models based on Gemini, or an optional call to ChatGPT. All of this routing will happen in the background , with Siri distributing queries according to their complexity, privacy requirements, and the computing power needed in each case.
The visual experience might also change, but not radically. Apple usually opts for subtle interface updates , so the feeling is that of a Siri "that finally understands and solves problems" rather than a completely new product. Where a more noticeable leap is expected is in the assistant's ability to maintain context throughout a conversation and to combine information from different apps into a single response.
For Spanish and European users, the added challenge lies in the quality of support in different languages and adaptation to local services. Apple will have to demonstrate that the models trained and refined from Gemini work just as well—or almost as well—in Spanish as in English, and that they understand the cultural, geographical, and usage nuances common in our environment.
All this activity paints a fairly clear picture: Apple has chosen to use Gemini's "brain," locked away in its own vault , to accelerate the evolution of Siri and Apple Intelligence, while maintaining its privacy rhetoric and control over the user experience. It remains to be seen whether, when the time comes to ask the iPhone to handle truly complex tasks, this combination of forces will finally translate into an assistant that lives up to what the company has been promising for years.
