Apple Building an AI App Store Through Siri Extensions in iOS 27
By Lisa Zhang
# Apple Building an AI App Store Through Siri Extensions in iOS 27
*By Lisa Zhang • March 30, 2026*
Apple is about to turn Siri into a marketplace. New details from Bloomberg's Mark Gurman reveal that iOS 27 will let users install third-party AI chatbots that run directly inside Siri. There's going to be a dedicated App Store section for these extensions. That's not an incremental update. That's Apple creating an entirely new platform category.
The company already lets users connect ChatGPT to Siri. But what's coming goes way beyond a single partnership. Apple is opening its AI assistant to any developer willing to build for it. Google Gemini, Anthropic Claude, Mistral, Perplexity, and any other chatbot provider can plug directly into Siri's interface. Users pick the ones they want, install them like apps, and access them through voice or text without ever leaving Apple's ecosystem.
## What Siri Extensions Actually Mean
The concept borrows from how browser extensions work. You've got a base application (Siri) and third-party additions that expand what it can do. But there's a critical difference. Browser extensions run inside a browser tab. Siri extensions run across your entire device.
Think about what that enables. You could ask Siri to analyze a photo using Claude's vision capabilities. Or have Gemini summarize a long email thread. Or use a specialized medical AI to interpret lab results. Each extension brings different capabilities, and they all work through the same natural language interface you already use to set timers and send texts.
Apple's approach solves a problem that's been bugging AI users for months. Right now, if you want to use Claude, you open the Claude app. Want Gemini? Different app. ChatGPT? Another one. Each has its own interface, its own conversation history, its own context window. It's fragmented and annoying.
Siri Extensions create a single entry point. You talk to Siri, and it routes your request to whichever AI model is best suited to handle it, or whichever one you've selected as your default for that type of task. The user experience gets dramatically simpler.
Apple's decision to build a dedicated App Store section for these extensions tells you they expect this to be big. They don't create new store categories for features they think will be niche. The [AI companies](/companies) building these models are about to compete for distribution through the most valuable consumer platform in technology.
## The Strategic Brilliance of Staying Model-Agnostic
Apple tried building its own AI models. Apple Intelligence launched with on-device capabilities that were, to put it diplomatically, underwhelming. The company's language models couldn't match what OpenAI, Anthropic, or Google were producing. Instead of dumping more billions into catching up, Apple made a smarter play: become the platform that connects users to the best models regardless of who built them.
This is classic Apple. They did the same thing with music. Instead of becoming a record label, they built iTunes and later Apple Music. They did it with apps. Instead of building every application themselves, they created the App Store and took a cut of every transaction.
The AI App Store follows identical logic. Let others spend billions training models. Apple provides distribution, a polished user experience, and probably takes 15-30% of subscription revenue that flows through the platform. Lower risk, guaranteed revenue, and they don't need to win the [AI models](/models) race to profit from it.
There's a defensive angle too. If Apple locks users into a single AI provider and that provider stumbles, Apple looks bad. By offering choice, they insulate themselves from any single model's failures. If Claude hallucinates badly, users can switch to Gemini. If GPT gets something wrong, try Mistral. Apple's Siri stays reliable because it's the router, not the source of answers.
## How Apple Gets Full Access to Gemini
A separate but related development fills in another piece of this strategy. As part of a deal announced in January, Apple now has what The Information describes as "complete access" to Google's Gemini models inside its data centers. This isn't just API calls. Apple can use Gemini to train smaller "student" models through a process called knowledge distillation.
Distillation works by having a large, capable model (Gemini) generate training data that a smaller model learns from. The student model won't be as capable as the teacher, but it can run efficiently on Apple's devices without needing a cloud connection. This is how Apple plans to improve its on-device AI without building massive foundation models from scratch.
The practical result: future iPhones and Macs will run AI features that were trained using Google's technology but don't require Google's servers. The models run locally, protecting user privacy while still delivering competitive performance. It's a way to close the gap with Android's on-device AI capabilities without the multi-billion dollar investment in training infrastructure.
For the [AI industry as a whole](/learn), this deal signals that the value chain is splitting. Some companies will specialize in training massive models. Others will specialize in deploying optimized versions of those models on consumer devices. The training companies and the deployment companies don't have to be the same entities.
## What Developers Need to Know
If you build AI applications, the Siri Extensions platform creates a new distribution channel worth paying attention to. Apple hasn't published developer documentation yet, but based on how they've handled previous platform launches, here's what to expect.
There will be an API specification that defines how extensions communicate with Siri. Extensions will need to handle natural language inputs and return structured responses. Apple will require privacy disclosures about what data each extension processes and whether it leaves the device.
The App Store review process will apply to AI extensions. That means Apple gets to set quality standards, reject extensions that don't meet them, and potentially block models that produce harmful content. Developers who've dealt with App Store review know this process can be slow and sometimes frustrating, but it also filters out low-quality submissions.
Revenue sharing will follow Apple's standard terms. For subscriptions processed through the App Store, Apple takes 15-30% depending on the developer's size and the subscription's duration. Large AI companies like OpenAI and Anthropic might negotiate custom terms, but smaller developers will likely pay standard rates.
The launch timing matters. iOS 27 is expected at WWDC in June with a public release in September. Developers should start planning their extension architecture now to be ready for the initial launch window, when user attention will be highest.
## Market Impact on AI Companies
This changes the competitive landscape for every AI model provider. Currently, distribution is one of the biggest challenges for AI companies. OpenAI has ChatGPT's brand recognition. Google has Android integration. But most other AI companies struggle to reach consumers directly.
Siri Extensions potentially put every AI model on equal footing in front of Apple's user base. That's over a billion active devices. A startup with a specialized AI model for cooking, fitness, or language learning suddenly has the same distribution channel as OpenAI or Google.
The question is whether Apple will curate aggressively or let the market decide. If they feature certain extensions prominently in the store, they become kingmakers. If they rank by user ratings and downloads, established brands with existing audiences have an advantage. Either way, being available through Siri becomes table stakes for any AI company targeting consumers.
For investors evaluating AI startups, Siri Extensions change the unit economics. Customer acquisition through the App Store is expensive but predictable. If an AI company can build an extension that users find genuinely useful, Apple's platform handles discovery, distribution, payment processing, and trust establishment. That's a lot of infrastructure a startup doesn't have to build.
## Privacy and Security Considerations
Apple's reputation is built on privacy. Any AI extension framework needs to maintain that reputation. Expect strict requirements around data handling.
Extensions that process data on-device will have an advantage in Apple's review process. Models that require sending user data to external servers will need clear, specific privacy disclosures. Apple may require extensions to offer an on-device processing option or limit the types of data they can access.
The company will likely implement a permission system similar to how apps currently request access to photos, location, or contacts. An AI extension might need explicit permission to read your emails, access your calendar, or use your photos. Users control what each extension can see.
This creates a tiered system. Extensions with broad data access can provide more personalized and useful responses but face higher privacy scrutiny. Extensions with limited access are easier to approve but less capable. Developers will need to find the right balance for their use case.
For more on how AI privacy works and why it matters, visit our [AI glossary](/glossary) and [comparison tools](/compare) that break down privacy practices across major AI platforms.
## Frequently Asked Questions
**When will Siri Extensions be available?**
Apple is expected to announce iOS 27 at WWDC in June 2026, with a public release in September. The AI Extensions App Store section should launch alongside the OS update.
**Which AI models will work with Siri Extensions?**
Apple is opening the platform to any developer. OpenAI's ChatGPT is already integrated, and Google Gemini, Anthropic Claude, Mistral, and others are expected to build extensions. Smaller specialized AI companies can also participate.
**Will Siri Extensions cost money?**
That depends on each developer. Some may offer free extensions with premium features. Others may require subscriptions. Apple will process payments through its standard App Store billing system.
**Does this replace Apple Intelligence?**
No. Apple Intelligence handles on-device tasks like text summarization and image editing. Siri Extensions add third-party AI capabilities on top of Apple's built-in features. They work together rather than replacing each other.
Get AI news in your inbox
Daily digest of what matters in AI.
Key Terms Explained
Anthropic
An AI safety company founded in 2021 by former OpenAI researchers, including Dario and Daniela Amodei.
Attention
A mechanism that lets neural networks focus on the most relevant parts of their input when producing output.
Chatbot
An AI system designed to have conversations with humans through text or voice.
Claude
Anthropic's family of AI assistants, including Claude Haiku, Sonnet, and Opus.