Apple Intelligence will debut on select iPhone 15 Pro models later this year, with the company enlisting the help of an unexpected competitor.
Apple has revealed that its AI models for Apple Intelligence were trained using Google’s custom chips. According to a research paper released this week, Apple utilized Google-designed tensor processing units (TPUs) in two key components of its AI system. This move is noteworthy given the prevalent reliance on Nvidia’s GPUs by major tech companies.
Specifically, Apple employed 2,048 TPUv5p chips for training AI models used in iPhones and other devices, and 8,192 TPUv4 processors for server-based AI training. While Nvidia specializes in GPUs—widely used in AI research—Google’s TPUs offer an alternative approach to AI infrastructure.
Additionally, Apple has previewed its highly anticipated Apple Intelligence system. Nvidia currently dominates the AI chip market, holding over 80 percent of it, due to the high demand for its costly GPUs amidst the AI surge.
In response to the substantial costs of developing AI infrastructure, major companies like Google, Meta, Microsoft, and Amazon have increasingly turned to creating their own AI chips. Google’s TPUs are available through the Google Cloud Platform, requiring users to build their software on Google Cloud to access the chips. The research paper also notes that Google’s chips can support the development of more complex and larger models.
This publication is the second technical document Apple has released on its AI technology, following a more general overview in June. During its WWDC conference in June 2024, Apple showcased new AI tools, including the integration of OpenAI’s ChatGPT into its software. This week, some features of Apple Intelligence will be available to beta testers.
Apple is expected to announce its quarterly earnings later this week, which will provide insights into its strategies for AI and other products moving forward.
