Apple "Training AI with Google Chips"... Big Tech Seeking Alternatives to NVIDIA

Apple announced on the 29th (local time) that it has adopted Google chips instead of Nvidia for training the AI models that form the basis of its proprietary artificial intelligence (AI) system.

On the same day, Apple released a paper titled "Apple Intelligence Foundation Language Models (AFM)." AFM is the AI model that underpins Apple's AI system "Apple Intelligence," which was announced last month.

[Image source=Reuters Yonhap News]

Although Apple did not directly mention Google or Nvidia in the paper, it stated that the AFM on-device and AFM server models were trained on a "Cloud TPU cluster."

The Tensor Processing Unit (TPU) is a chip designed by Google specifically for AI operations. This means that Apple trained its AI models using cloud servers equipped with AI chips designed by Google.

Apple revealed that for training the on-device AI models running on devices like the iPhone, it used 2,048 of the latest TPUv5p chips introduced in December last year, and for server AI models, it used 8,192 units of the TPUv4, the 4th generation TPU version released last year.

CNBC interpreted Apple's use of Google chips instead of Nvidia for AI model training as "a sign that big tech companies are seeking alternatives to Nvidia for cutting-edge AI training."

Nvidia's graphics processing units (GPUs) currently hold an 80% share of the AI chip market. They are expensive, and due to a surge in demand over the past few years, procurement has become difficult. AI model developers such as OpenAI, Microsoft, and Anthropic all use GPUs. Google also remains one of Nvidia's major customers, training AI systems using both Nvidia GPUs and its own TPUs.

This content was produced with the assistance of AI translation services.