Sundar Pichai and Tim Cook
Source: Reuters; Apple
Apple said Monday that the artificial intelligence models underlying Apple Intelligence, its AI system, are pre-trained on processors designed by Google, a sign that big tech companies are looking for alternatives to Nvidia when it comes to training cutting-edge AI.
Apple’s choice of Google-developed Tensor Processing Unit (TPU) technology for training was detailed in a technical paper the company just published. Meanwhile, Apple on Monday released a preview version of Apple Intelligence for select devices.
Nvidia’s expensive graphics processing units (GPUs) dominate the market for high-end AI training chips and have been in such high demand in recent years that they’re hard to come by in the quantities needed. OpenAI, Microsoft, and Anthropic all use Nvidia’s GPUs for their models, while other tech companies, including Google, Meta, Oracle, and Tesla, buy them to develop their AI systems and offerings.
Meta CEO Mark Zuckerberg and Alphabet CEO Sundar Pichai both made comments last week suggesting their companies and others in the sector might overinvest in AI infrastructure, but acknowledged the business risk of doing otherwise was too high.
“The downside of being behind is that you’re not in a position to get to the technology that’s going to be most important for the next 10 to 15 years,” Zuckerberg said on a podcast with Bloomberg’s Emily Chang.
In its 47-page document, Apple doesn’t mention Google or Nvidia, but says that its Apple Foundation Model (AFM) and AFM server are trained on “Cloud TPU clusters.” That means Apple has leased servers from a cloud provider to do the computations.
“This system allows us to train AFM models efficiently and scalably, including on-device AFM, server-based AFM, and larger models,” Apple said in the document.
Representatives for Apple and Google did not respond to requests for comment.
Apple unveiled its AI plans later than many of its competitors, which have been embracing generative AI since OpenAI launched ChatGPT in late 2022. On Monday, Apple unveiled Apple Intelligence. The system includes several new features, such as a refreshed look for Siri, better natural language processing, and AI-generated summaries in text fields.
Over the next year, Apple plans to roll out generative AI features, including image generation, emoji generation, and an enhanced Siri that can access a user’s personal information and perform actions within apps.
In the document released Monday, Apple said the AFM on the device was trained on a single “slice” of 2,048 TPU v5p chips working together. This is the most advanced TPU, first released in December. The AFM server was trained on 8,192 TPU v4 chips that were configured to work together in eight slices over a data center network, according to the document.
Google’s latest TPUs cost less than $2 per hour of chip usage when reserved three years in advance, according to Google’s website. Google first introduced its TPUs in 2015 for internal workloads and made them available to the public in 2017. They are now among the most mature custom chips designed for artificial intelligence.
Google remains one of Nvidia’s largest customers, however. It uses Nvidia’s GPUs and its own TPUs to train AI systems, and also sells access to Nvidia’s technology on its cloud.
Apple had previously said that inference, which involves taking a pre-trained AI model and running it to generate content or make predictions, would happen partly on Apple’s own chips in its data centers.
This is the second technical paper on Apple’s AI system, following the release of a more general version in June, when Apple said it was using TPUs to develop its AI models.
Apple is expected to report quarterly results after markets close on Thursday.
