Apple’s secret AI chip has been code-named ACDC (Apple Chips for Data Centers), and the Cupertino company has been working on this project under cloak and dagger for years with Taiwan Semiconductor Manufacturing Company (TSMC).
The Wall Street Journal broke the initial report on Project ACDC. This project has been in the works for years, which does not surprise me. Ever since Apple’s M-series chips emerged, both performance leaders and giant performance-per-watt leaders, it seemed somewhat obvious that Apple’s own data centers would begin to change at some point in the future.
AI for Inferencing
The ACDC AI chips are reportedly aimed at inferencing, not training AI models. This suggests that Apple is initially planning to partner with AI LLMs from the likes of OpenAI, Anthropic, or others. A month ago, Apple hired former Google, IBM, and Nvidia executive Sumit Gupta as director of products for Apple Cloud.
Apple’s latest M-4 chip delivers a next-generation GPU architecture with faster hardware-accelerated ray tracing and brings it to the iPad Pro ahead of future Macs. The M4 Neural Engine is Apple’s most powerful AI chip technology publicly announced.
Apple’s plans seem to aim at creating custom server systems for its data centers around the world. Bloomberg’s Mark Gurman refuted this idea, suggesting that Apple prefers on-device AI processing. But Apple has a long history of playing two hands of cards at once. A report suggests that Apple may never reveal these chips publicly.
Meanwhile, earlier this week, Apple released its M-4 chip inside the new iPad Pro.