Apple's M7 Ultra to offer up to 1.5 TB unified memory

Apple is redesigning its Mac‑chip roadmap to prioritize artificial‑intelligence performance. The company will skip the Pro, Max and Ultra variants of the upcoming M6 line and move directly to the M7 family. The base M7 chip is slated for the first half of 2027, with M7 Pro and M7 Max due in late 2027 and the M7 Ultra expected in 2028. The M7 Ultra is designed to support up to 1.5 TB of unified memory – roughly double the 768 GB cap of the M5 Ultra – matching the ceiling of the 2019 Intel‑based Mac Pro. Apple says the extra memory is intended to let developers run very large AI models locally on a Mac Studio without relying on cloud servers, and the chip could also form the core of a future Apple‑branded AI server projected for around 2029.

The roadmap shift is linked to a major upgrade of the Neural Engine, a component whose architecture originated from Apple’s cancelled self‑driving‑car project. By accelerating the M7 generation, Apple hopes to deliver “significant” AI performance gains that approach dedicated accelerator chips such as Nvidia’s Blackwell. The company is also exploring a partnership with Intel to produce some Mac and iPhone silicon, a deal that secured a tariff exemption for Apple after negotiations with the U.S. government. Supply‑chain constraints in the DRAM market could limit the availability of the 1.5 TB configuration and drive the price of a fully‑populated Mac Studio into the tens of thousands of dollars.