Sundar Pichai and Tim Cook
Source: Reuters; Apple
Apple said on Monday that the artificial intelligence models underpinning Apple Intelligence, its AI system, were pretrained on processors designed by Google, a sign that Big Tech companies are looking for alternatives to Nvidia when it comes to the training of cutting-edge AI.
Apple’s choice of Google’s homegrown Tensor Processing Unit (TPU) for training was detailed in a technical paper just published by the company. Separately, Apple released a preview version of Apple Intelligence for some devices on Monday.
Nvidia’s pricey graphics processing units (GPUs) dominate the market for high-end AI training chips, and have been in such high demand over the past couple years that they’ve been difficult to procure in the required quantities. OpenAI, Microsoft, and Anthropic are all using Nvidia’s GPUs for their models, while other tech companies, including Google, Meta, Oracle and Tesla are snapping them up to build out their AI systems and offerings.
Meta CEO Mark Zuckerberg and Alphabet CEO Sundar Pichai both made comments last week suggesting that their companies and others in the industry may be overinvesting in AI infrastructure, but acknowledged the business risk of doing otherwise was too high.
“The downside of being behind is that you’re out of position for like the most important technology for the next 10 to 15 years,” Zuckerberg said on a podcast with Bloomberg’s Emily Chang.
Apple doesn’t name Google or Nvidia in its 47-page paper, but did note its Apple Foundation Model (AFM) and AFM server are trained on “Cloud TPU clusters.” That means Apple rented servers from a cloud provider to perform the calculations.
“This system allows us to train the AFM models efficiently and scalably, including AFM-on-device, AFM-server, and larger models,” Apple said in the paper.
Representatives for Apple and Google didn’t respond to requests for comment.
Apple was later to reveal its AI plans than many of its peers, which loudly embraced generative AI soon after OpenAI’s launch of ChatGPT in late 2022. On Monday, Apple introduced 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 functions based on generative AI, including image generation, emoji generation and a powered-up Siri that can access the user’s personal information and take actions inside of apps.
In Monday’s paper, Apple said that AFM on-device was trained on a single “slice” of 2048 TPU v5p chips working together. That’s the most advanced TPU, first launched in December. AFM-server was trained on 8192 TPU v4 chips that were configured to work together as eight slices through a data center network, according to the paper.
Google’s latest TPUs cost under $2 per hour the chip is in use when booked for 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’re now among the most mature custom chips designed for artificial intelligence.
Still, Google is one of Nvidia’s top customers. It uses Nvidia’s GPUs its own TPUs for training AI systems, and also sells access to Nvidia’s technology on its cloud.
Apple previously said that inferencing, which means taking a pretrained AI model and running it to generate content or make predictions, would partially happen on Apple’s own chips in its data centers.
This is the second technical paper about Apple’s AI system, after a more general version was published in June. Apple said at the time that it was using TPUs as it developed its AI models.
Apple is scheduled to report quarterly results after the close of trading on Thursday.
