Apple is in the middle of revamping Siri with a new LLM-based backend, and a new report suggests Apple is in talks with Google to get Gemini integrated with the system.

There has been a lot of talk about Apple being behind in artificial intelligence, and so far, its offerings are very different from its competitors. Apple has internally hit a bit of a reset button and is actively rebuilding Siri with an LLM backend, but there's still a lot in question about implementation.

According to the latest report from Bloomberg, Apple is in talks with Google to use Gemini to "power a revamped version of the Siri voice assistant." The report takes this to mean that Apple could replace its backend with Google Gemini, but evidence increasingly suggests that isn't going to happen.

An anonymous tipster with the information either doesn't have the full picture, or the report is meant to support a previous narrative about Apple not knowing how to save Siri on its own. Either way, the reality is in the details and not the angle of the report.

The core of the leaked information is "Google has started training a model that could run on Apple's servers." That's nearly identical to what was shared previously about Apple's talks with Anthropic and OpenAI.

It isn't that Apple wants to replace the Siri backend with Gemini — it's that it wants to bolster its offerings with multiple external, but private, AI tools. The LLM-powered Siri that launches in early 2026 would run on Apple Intelligence, but could call out to other models as needed or per user request.

The report does acknowledge as much later, stating that the third-party models would run via Private Cloud Compute on external servers, not on-device. So if that's the case, it seems odd to suggest that Gemini would "serve as the foundation of the new Siri."

Apple's AI strategy

It certainly might be more tantalizing to suggest there's some kind of scramble at Apple internally to "fix" its AI efforts. Other reporting from Bloomberg has breathlessly suggested that each of the six departures from Apple's AI team would mean its imminent demise.

Everything else suggests Apple's AI efforts are going strong internally. Sure, the app intent-based intelligence upgrades were delayed by a year, but not because the system wasn't working.

Apple was clear that it had a working system the entire time, but it took a hybrid approach, blending ML and AI architectures. It led to higher hallucination rates than Apple desired, so it hit pause and reset.

What has emerged is a newly designed system with the core foundation Apple Intelligence model running Siri on device and in the cloud. Giving users the option to call out to third-party LLM models via Private Cloud Compute, which is private, running on renewable energy, and is incapable of training on user data, is just an add-on, not a replacement.

This is supported by separate reporting about project Linwood and Glenwood. These codenames reflect the two systems, Apple's onboard LLM and external third-party LLMs, working together in a joint system.

However, Bloomberg's reporting suggests that instead of these systems working together, Apple is running them against each other in some kind of "bake off." Given Apple's history aiming for vertical integration and its willingness to work with third parties, it seems more likely they are a cooperative effort, not a contest.

As Apple CEO Tim Cook said previously, it doesn't feel the need to be first to AI, but it does plan on being the best. And if Apple pulls off the implementation as described, it'll be at the forefront of the technology in the consumer space with an offering no other company can replicate.