Image: Apple
A report suggests Apple shelved plans to build the M4 Extreme Apple silicon chip. The company gave up on the SoC to divert resources to build an AI server chip.
The report points to Apple working with Broadcom to develop this AI chip.
Apple and Broadcom collaborating on an AI chip
All major tech companies have doubled down on AI. Apple is no different. Despite being late to the game with Apple Intelligence, the company is ramping up its efforts. While Apple uses Amazon’s AI chips to pre-train Apple Intelligence models, it relies on Apple silicon-powered servers for actual processing. It currently uses M2 Ultra-powered servers and plans to switch to M4 servers in 2025.
With AI use set to boom in the coming years, Apple is working on even more powerful chips for faster processing.
The Information reports that Apple’s AI chip is internally known as “Baltra.” It will seemingly be ready for mass production by 2026 and use TSMC’s advanced 3nm node. Apple will rely on Broadcom’s expertise to develop the chip’s networking technology, enabling it to connect to a network of devices for faster AI processing. Both companies have supposedly set a 12-month deadline to complete the SoC’s design.
In 2023, Apple signed a multi-year deal with Broadcom to develop 5G RF chips, indicating the close ties between the two companies.
AI chip takes priority over M4 Extreme
To focus on the development of this new AI chip, Apple supposedly canceled the M4 Extreme chip for its upcoming high-end Macs. The SoC would have presumably featured multiple M4 Max/Ultra chips glued together to deliver even better performance. Rumors point to the chip featuring a 64-core CPU and up to a 160-core GPU.
Nvidia is currently the world’s largest supplier of AI chips to other companies. However, Apple actively avoids using Nvidia’s GPUs to train its AI models due to its longstanding feud with the latter.
