Apple is contemplating the use of the latest custom Amazon AI chips to pre-train Apple Intelligence models in the future, the company’s senior machine learning director confirmed in a public appearance on Tuesday (via CNBC).
Benoit Dupin, Apple’s senior director of machine learning and AI, made a rare public appearance on behalf of the company at the Amazon Web Services re:Invent conference in Las Vegas. Dupin discusses how Apple has been relying on Amazon’s servers and custom chips to power web services like Siri, Apple Music, and Apple Maps. Dupin points out that Amazon’s chips like Trainium and Graviton have led to a 40 percent efficiency gain compared to the competition.
Perhaps surprisingly, Dupin also delved into Apple’s plans to rely on Amazon AI chips for pre-train future Apple Intelligence models on the newly announced Trainium2 processor. “In early stages of evaluating Trainium2 we expect early numbers up to 50% improvement in efficiency with pretraining,” Dupin added.
Discover new horizons, always connected with eSIM
Travel the world stress and hassle-free with the best eSIM service available. Enjoy unlimited data, 5G speeds, and global coverage for affordable prices with Holafly. And, enjoy an exclusive 5% discount.
Apple relies on different vendors to pre-train its Apple Intelligence models. For example, a research paper published by the company earlier this year disclosed the use of Google Cloud’s TPU chips to train some Apple Intelligence models. However, Apple has yet to disclose its reliance on any offerings from Nvidia, which has become the industry standard for AI training.
It’s important to note that the use of Google and Amazon AI chips is limited to training Apple Intelligence models only. Production versions of the company’s AI features continue to operate on custom Private Cloud Compute servers that Apple runs itself and are exclusively powered by Apple silicon chips.
