Apple Developing New AI Chip That Could Power Server-Based AI Features: Report

Apple is reportedly working on developing its artificial intelligence (AI) chipsets to run software in data centres. If true, this would mean a major change of plans for the Cupertino-based tech giant, as previous reports highlighted the company was focused on only on-device AI features for its devices. However, the development of AI chips hints that Apple is also considering server-based AI features. Some of these features could be unveiled at the Worldwide Developers Conference (WWDC) 2024, which is scheduled to be held on June 10.

According to a report by The Wall Street Journal, Apple has been working on its server project, which is internally known as Project ACDC for several years. Citing unnamed people familiar with the matter, the report highlighted that the tech giant is now planning to introduce specific chipsets for these data centres to make them capable of running AI computing.

The tech giant is reportedly working with Taiwan Semiconductor Manufacturing Co. (TSMC) to design and produce these AI chips, however, it is not clear if the manufactured processors showed a definitive result. These chips are said to be used for only inference purposes (running the AI model) and not to train large language models (LLMs). Notably, TSMC is a major chip-making partner for Apple and produces most of the silicon used in the company's devices.

If the claims are true, this could point to a major shift in Apple's AI strategy. Many reports earlier claimed that the company was interested in only bringing on-device AI features that could be processed entirely locally. The evidence of this plan was seen when it unveiled the new iPad Pro 2024 with the M4 chip, which comes with a new Neural Processing Unit (NPU) to run AI tasks.

However, so far no smartphone maker has been able to offer a suite of AI features that function entirely locally due to the high requirement for hardware and instead divide the processes between the server and the device. Notable examples are the AI features of Samsung's Galaxy AI and Oppo. It appears Apple has also realised this problem and now plans to combine both on-device and server-based AI features for its ecosystem.

This likely means that the features that are non-privacy intrusive might be run from servers while those features that access the user's data could still be processed locally. This combination could also allow Apple to expand these features to its older devices that do not have the hardware to support on-device AI features. While these are just speculations, we will know for sure if Apple unveils its AI features at the WWDC 2024.