Apple's M7 Ultra Chip Targets 1.5TB Unified Memory and Blackwell-Class AI Performance
Apple is designing the M7 Ultra chip to feature up to 1.5TB of unified memory and dramatically upgraded AI performance, according to a report.
Apple is currently designing its planned M7 Ultra chip to support up to 1.5 terabytes of unified memory while pushing AI performance closer to the class of dedicated accelerators from Nvidia, according to a report published by Mark Gurman and cited by Tom's Hardware. The ambitious memory configuration represents roughly twice the capacity that Apple has planned for the M5 Ultra. However, whether this maximum memory configuration can ultimately ship depends entirely on the state of the broader memory market and supply availability.
The realization of the 1.5TB memory target is constrained by ongoing component availability. Tom's Hardware notes that Apple already pulled the 128GB Mac Studio this year due to supply constraints as DRAM prices climbed. Delivering a 1.5 terabyte part would require significantly larger volumes of scarce and high-cost memory components, making the rollout conditional on supply chain improvements.
In terms of architecture and scheduling, Apple has reportedly compressed its Mac silicon timeline by taping out the M7 roughly six months after starting the same process for the M6. While a base M6 chip is expected this fall for entry-level Macs—with the company reportedly skipping Pro, Max, and Ultra variants for that generation—the base M7 is anticipated in the first half of 2027. The M7 Pro and M7 Max are slated for the end of 2027, followed by the M7 Ultra in 2028.
The M7 Ultra's performance focus centers on handling heavy local AI workloads. Gurman describes the processor as delivering a dramatic upgrade to AI performance that brings it closer to Nvidia's Blackwell-class accelerators, rather than achieving strict parity with data-center silicon. By comparison, Apple's current M3 Ultra achieves 819 gigabytes per second of memory bandwidth by fusing two Max dies together.
Beyond client hardware, Apple is preparing server deployments utilizing its silicon. According to the report, the company is building an AI server using the M5 Ultra under the code name J246 for deployment soon, with a second server chip based on the M7 Ultra planned for 2029. Additionally, the 2028 generation will move to a 1.4-nanometer process, aligning with TSMC's A14 node scheduled for mass production in the second half of 2028. None of these dates or specifications have been officially confirmed by Apple.