A new report has revealed technical details regarding the Apple Google AI deal, highlighting how the two companies plan to implement artificial intelligence features. The upcoming Worldwide Developers Conference (WWDC) is scheduled to showcase iOS 27 and an updated version of Siri. Meanwhile, details from industry sources indicate how the backend infrastructure will support these new capabilities.
Technical Details of the Apple Google AI deal
The Apple Google AI deal focuses on the core technologies behind the upcoming features rather than the user-facing applications. According to reports, Apple plans to prioritize on-device processing for its upcoming wave of artificial intelligence features. However, the company will still require external cloud infrastructure to handle complex queries that exceed local hardware capabilities.
To achieve local processing, Apple is reportedly using a version of Google’s Gemini large language model. Specifically, the company uses this model to train a smaller version that can run locally on mobile devices. This training methodology is known in the industry as distillation, which helps shrink the size of the model.
On-Device Processing and Model Distillation
In addition to model distillation, Apple is actively seeking to acquire smaller technology companies. These acquisitions aim to assist in the effort of shrinking artificial intelligence models to run efficiently on consumer hardware. Reports indicate that Apple has considered acquiring Liquid AI, a startup based in Cambridge, Massachusetts, which specializes in running local models.
Despite these efforts, many user queries will still require cloud support. The full Gemini model provided by Google contains trillions of parameters. Consequently, this model requires significant computing power, making it difficult for Apple to run entirely on its internal server infrastructure, known as Private Cloud Compute.
Cloud Infrastructure and Nvidia Hardware
To resolve these infrastructure limitations, Apple is turning to Google Cloud and Nvidia hardware. As part of the Apple Google AI deal, some user queries sent to Siri will run in Google Cloud on a licensed version of the Gemini model. This hybrid approach allows Apple to offload heavy computational tasks to external servers.
Furthermore, Apple recently approved the use of a specific privacy technology from Nvidia. This approval suggests that Apple will utilize Nvidia graphics processing units for at least some of its computing needs within Google Cloud. This decision was reportedly finalized in recent weeks as Apple finalized its launch strategy.
Privacy Protections and Confidential Compute
The integration of Nvidia hardware introduces a security feature known as confidential compute to enhance cybersecurity. This technology encrypts data and models while they are being processed inside the graphics processing units. Although this encryption slightly slows down the processing of queries, it helps Apple maintain its user privacy standards.
This aspect of the Apple Google AI deal highlights the balance between processing power and user privacy. Apple plans to continue using the Private Cloud Compute branding for its upcoming features. This branding will remain active even though the queries will no longer run exclusively on Apple’s proprietary servers.
