Siri overhaul: Apple relies on Google and Nvidia to revolutionize its voice assistant

The project has dragged on for years. Apple promised a truly intelligent Siri, then postponed it for lack of a homegrown model up to the task.The chosen solution relies on its rivals: to power the new version of its assistant, the company would use Google’s cloud and NVIDIA’s chips.

Why did Apple choose Google?

The new Siri will be based on a custom-built Gemini model with 1.2 trillion parameters, under a multiyear agreement estimated at around $1 billion per year. Before reaching that point, however, Apple had other candidates: the company evaluated OpenAI’s ChatGPT and Anthropic’s Claude, the latter even being the favorite for a long time, with a proprietary version already tested on Apple’s servers. Discussions ran into trouble over the price: Anthropic was asking for several billion dollars per year, an amount expected to double over three years. Google proved considerably less expensive, in a more comfortable legal context, with the two companies already linked by the $20 billion agreement that makes Google the default search engine on iPhone. OpenAI, for its part, raised other concerns, ranging from hiring in-house engineers to a competing hardware project. Nevertheless, Apple continues to use Claude internally to develop its products, indicating that the disagreement was about the cost, not the quality of the model.

What dependencies is Apple willing to accept?

To absorb the load—that is, to train and run an assistant used by hundreds of millions of iPhones—Apple would also rely on Google’s infrastructure and NVIDIA’s accelerators, which are among the most powerful options available for large language models.These are the same chips that power the entire generative AI race today, and that no manufacturer yet truly knows how to replace. The standard-bearer of vertical integration, which designs its own chips and claims its independence, thus finds itself renting the brain of its assistant from one rival and its computing power from another.

The privacy promise, for its part, is starting to falter a little. Apple vows that processing remains on the device or in its Private Cloud Compute, but running a Google model raises questions, and for the deeper overhaul expected at its developers’ conference, the two companies would even be discussing hosting it on Google’s servers. A first Gemini-powered version is expected as early as spring, ahead of a complete revamp later in the year. Ultimately, it will be an assistant capable of multi-turn conversations and chained actions across apps, a far cry from today’s pushy Siri.

Apple thus ends up with a finally competitive Siri, but at the cost of its independence: the champion of vertical integration now rents its assistant’s intelligence from a rival and its computing power from a supplier. For the user, the deal is simple: a smarter assistant in exchange for one more dependency, carefully tucked under the hood.

👉🏻 Follow tech news in real time: add 01net to your sources on Google and subscribe to our WhatsApp channel.

Source: The Information