Apple and Google are reportedly discussing using Google's AI and cloud infrastructure to support the next version of Siri, Apple's digital assistant.

As reported by The Information, this builds on the two companies' existing agreement in which Apple uses Google's Gemini models for the assistant.

In addition to Apple, Google has also reportedly secured a major deal with Meta under which the social media company will be using Google's Tensor Processing Units (TPUs) to develop new models.

Apple already uses Google Cloud for online storage purposes, and for the training of Apple's AI models. An agreement, however, to support Siri, will deviate from the company's previous efforts to be self-sufficient in cloud computing.

In the past, Apple has said it sends AI queries to its "Private Cloud Compute," which runs on Apple servers with custom chips. Last year, the company announced it would be stepping up capex to invest in the platform.

Despite this aim, according to The Information, Apple's efforts to reduce reliance on external cloud providers have been "stymied" with executives resisting making large investments in its own infrastructure. Sources told The Information this has led to the "ongoing departures" of cloud experts in the company.

According to former employees, only around 10 percent of the Private Cloud Compute capacity is in use on average, and some servers are still waiting to be installed. However, should the new version of Siri prove to be a success, the company could see compute requirements dramatically increase.

Additionally, the sources said that Apple's custom chips are not designed with AI in mind, and aren't well-equipped to handle large models such as Google's Gemini.

Apple's relationship with cloud providers spans several years. With the launch of ChatGPT 3.5 in 2022, it became apparent that cloud-based models could offer a seismic shift, and, citing internal sources, Apple was in the process of decommissioning older Nvidia chips in its data centers that were starting to fail.

Thus, the company turned to cloud providers - starting with AWS, and later Google Cloud.

Meta goes TPU crazy

Last week, reports emerged that Meta Platforms had signed a major agreement to rent Google's custom AI chips known as TPUs.

Also, first reported by The Information, the deal was said to be worth billions of dollars, according to a person with knowledge of the matter.

The two companies are further discussing the possibility of Meta buying TPUs to deploy in its own data centers, though The Information was unsure how this was progressing.

This is in addition to an agreement Google has reportedly signed with an unnamed investment firm to create a joint venture that would lease access to TPUs to other customers. The company is also reported to be meeting with finance partners to borrow money for "special purpose vehicles" that would buy and lease out the TPUs.

Google's TPUs have typically been offered via Google's own cloud platform, hosted within Google data centers. Anthropic is a known user of the chips for training its Claude AI models.

In October 2025, Google’s VP and GM of AI and infrastructure, Amin Vahdat, said demand for the company’s TPUs was so oversubscribed that the company was having to turn customers away.

Just last month, Meta signed an agreement with Nvidia to deploy the chip designer's CPUs alongside “millions” of Blackwell and Rubin GPUs and the Spectrum-X Ethernet networking platform.

Financial terms of the deal were not disclosed, but in a statement, Nvidia said the agreement was multi-year and multi-generational, and would represent the first “large-scale” Nvidia Grace-only deployment.