December 12, 2024 – Reports from The Information indicate that Apple is collaborating with Broadcom on the development of its first server chip specifically designed for artificial intelligence. The AI chip, internally codenamed Baltra, is expected to be ready for mass production by 2026.
This development marks a significant milestone for Apple’s chip team if the production proceeds smoothly. Sources reveal that Apple is working with Broadcom on the chip’s networking technology, which is crucial for AI processing.
Recently, Benoit Dupin, Apple’s Senior Director of Machine Learning and Artificial Intelligence, stated that the company is evaluating the use of Amazon’s latest AI chip for pre-training its Apple Intelligence models.
According to Dupin, Apple has been utilizing AWS chips such as Graviton and Inferentia for over a decade to support services like Siri, Search, App Store, Apple Music, and Apple Maps. These chips have achieved a 40% efficiency improvement compared to x86 chips from Intel and AMD.
Furthermore, Dupin confirmed that Apple is currently assessing the use of AWS’s newest AI training chip, Trainium2, which is expected to increase efficiency by up to 50% during pre-training. However, it’s important to note that Apple’s use of Trainium2 is limited to the pre-training stage of AI models and will not be utilized for Apple Intelligence features. These features are powered by chips on Apple devices or Apple Silicon chips on the company’s proprietary cloud computing platform.
Earlier this year, Apple also acknowledged in a research paper the use of Google’s Tensor chips for training AI models, rather than relying on NVIDIA chips, which are preferred by other companies.
