Apple considers Anthropic or OpenAI to power Siri

Apple might finally be ready to admit it can't fix Siri alone.

When Cupertino launched the iPhone 16, the entire marketing pitch revolved around being "built for Apple Intelligence." But behind closed doors, Apple has reportedly been struggling so badly that it is now weighing a massive pivot: replacing Siri's in-house AI engine with models from Anthropic or OpenAI.

According to Bloomberg, Apple has held talks with both AI giants about training their frontier models specifically to deploy them directly onto Apple's private cloud servers.

Delays, executive shuffles, and delayed hardware

This is fundamentally different from Apple's current setup. Right now, Siri merely hands off tricky queries to ChatGPT as an opt-in extension. What Apple is exploring now is ripping out the core brains of Siri and substituting them with Claude or GPT.

Apple has not made a final call yet and is still keeping an in-house version of Siri powered by Apple Foundation Models in development. But the internal timeline has already slipped hard. A fully capable AI version of Siri, originally expected across iOS 18, iPadOS 18, and macOS 15, is not expected until 2026.

The delays have triggered chaos internally. Apple stripped AI chief John Giannandrea of Siri responsibilities, handing the reins over to software boss Craig Federighi and Mike Rockwell, the executive who led Vision Pro. Giannandrea also lost control of Apple's robotics division.

The fallout has even hit the hardware roadmap. An unreleased smart home display combining an iPad and a HomePod, designed to rival the Amazon Echo Show, has reportedly been delayed because Siri isn't ready. There have even been whispers that Apple mulled outright buying AI search startup Perplexity.

Meanwhile, Bloomberg reports that the prospect of outsourcing Siri's brain is cratering morale across Apple's AI division, right when rivals are throwing blank checks at top researchers.

Local privacy meets borrowed cloud smarts

Apple's on-device models aren't disappearing entirely. The company pledged at WWDC to open its local models to third-party developers this year, keeping features like Genmoji and Writing Tools native to the silicon.

Yet on-device models simply lack the raw reasoning power of massive cloud models. Running an external partner's model on Apple's own cloud hardware is Apple's attempt to split the difference: preserving its legendary privacy narrative while finally giving Siri an IQ upgrade.

Why it matters

For anyone holding an iPhone, this would be a massive practical upgrade. Users have endured Siri being a glorified timer for over a decade. Whether it gets powered by Anthropic's Claude or OpenAI's GPT, iPhone owners will finally get a conversational assistant that actually works, even if they have to wait until 2026.

For the broader tech landscape, this looks like Apple quietly conceding the frontier model race. Apple built its empire on vertically integrated hardware and software. Outsourcing the primary voice and intelligence layer of iOS puts Apple in the exact same position as Samsung, which relies heavily on Google Gemini to power Galaxy AI.

We think licensing an external model and hosting it on Apple infrastructure is the smartest triage move Tim Cook can make to avoid shipping a half-baked assistant. But it carries enormous risks. It signals to top AI talent that Cupertino is no longer the place to build frontier breakthroughs, and it leaves Apple dependent on external roadmaps for the most critical platform shift in mobile history.