In a surprising move, Apple recently announced that it had chosen Google’s Tensor Processing Unit (TPU) over Nvidia’s graphics processing units (GPUs) for training its cutting-edge AI models. This decision marks a significant shift in the industry, with big tech companies like Apple looking for alternatives to Nvidia’s GPUs for AI training.
Google’s TPUs have been gaining traction in the market as a cost-effective and efficient solution for training AI models. Apple revealed in a technical paper that its Apple Intelligence system was pre-trained on Google’s TPUs, showcasing the growing popularity of Google’s homegrown chips for AI training.
While Nvidia has long dominated the market for high-end AI training chips, the demand for its GPUs has skyrocketed in recent years, making them difficult to procure in the quantities needed. Tech giants like OpenAI, Microsoft, and Anthropic have been relying on Nvidia’s GPUs for their AI models, while others like Google, Meta, Oracle, and Tesla have been investing in building out their AI systems using Nvidia’s technology.
However, Apple’s choice of Google’s TPUs for training its AI models signals a shift in the industry towards more diverse options for AI infrastructure. Meta CEO Mark Zuckerberg and Alphabet CEO Sundar Pichai have both acknowledged the importance of investing in AI infrastructure, while also highlighting the risks of falling behind in the rapidly evolving technology landscape.
Apple’s Apple Intelligence system, which was introduced recently, features a range of new features including a refreshed look for Siri, improved 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 more advanced Siri that can utilize personal information to take actions inside apps.
In a 47-page technical paper, Apple detailed that its Apple Foundation Model (AFM) and AFM server were trained on “Cloud TPU clusters,” highlighting the efficiency and scalability of Google’s TPUs for AI training. Apple’s AFM on-device was trained on a single “slice” of Google’s most advanced TPU, while AFM-server was trained on a network of TPUs configured to work together in a data center setting.
Google’s TPU and Apple’s Future Directions
Google’s TPUs, which cost under $2 per hour when booked for three years in advance, have emerged as a mature custom chip designed specifically for artificial intelligence workloads. Google first introduced its TPUs in 2015 for internal use, before making them available to the public in 2017. Despite the success of its TPUs, Google remains one of Nvidia’s top customers, using a combination of Nvidia’s GPUs and its own TPUs for training AI systems.
Apple’s decision to use Google’s TPUs for training its AI models reflects a broader trend in the industry, where companies are exploring alternative options to Nvidia’s GPUs for AI infrastructure. With a focus on generative AI and advanced features for its Apple Intelligence system, Apple is positioning itself to compete in the rapidly evolving AI landscape.
Apple’s choice to deviate from the industry norm and adopt Google’s TPUs for AI training showcases a shift towards diversity and innovation in AI infrastructure. As tech giants continue to invest in advancing their AI capabilities, the competition between different chip manufacturers will drive progress and innovation in the field of artificial intelligence. Apple’s future plans for generative AI and advanced features indicate a strategic focus on staying ahead of the curve and delivering cutting-edge AI solutions to its users.
Leave a Reply