Apple’s M7 Ultra could challenge Nvidia Blackwell with up to 1.5TB of memory

Apple is reportedly developing its next high-end processor with artificial intelligence at the centre of its design. The M7 Ultra, expected to arrive in 2028, could be one of the company’s most significant advances in custom silicon, bringing Mac systems closer to the capabilities of dedicated enterprise AI hardware.

According to Bloomberg’s Mark Gurman, the M7 Ultra is being designed to deliver a major increase in AI performance rather than serving as a routine upgrade focused mainly on faster processing and graphics. The chip could support workloads that are typically handled by specialised AI accelerators, potentially placing it in closer competition with powerful platforms such as Nvidia’s Blackwell.

The reported development suggests that Apple is changing the priorities behind its processor roadmap. While previous generations have focused on improving CPU speed, graphics performance, and energy efficiency, future Apple silicon could be built primarily to meet the growing demands of large AI models.

Unified memory could reach server-class levels

One of the most notable reported features of the M7 Ultra is support for up to 1.5 TB of unified memory. This would be an unusually large amount for a desktop processor and far beyond the memory capacity available in current consumer Mac computers.

The proposed capacity would also be roughly twice the amount reportedly planned for Apple’s upcoming M5 Ultra. If the development roadmap remains unchanged, the increase could enable future Mac systems to process much larger AI models directly on the device, without relying as heavily on cloud computing services or external storage.

Apple’s unified memory architecture allows the CPU, GPU and other parts of the processor to access the same pool of memory. This approach can reduce the need to copy data between separate memory systems, helping workloads run more efficiently. A much larger unified memory capacity could therefore provide important benefits when processing AI models that require access to large amounts of data.

Modern AI systems often need substantial memory to store model parameters and process complex requests. Memory limitations can restrict the size of models that can run locally or require parts of a workload to be moved between storage and memory. By supporting up to 1.5 TB, the M7 Ultra could reduce these restrictions and allow more demanding AI tasks to remain in the system’s main memory.

Such a capacity would place future Apple hardware in a category more commonly associated with high-end workstations and servers. However, the final specifications, product configurations and pricing have not been confirmed, and Apple has not officially announced the M7 Ultra.

AI could reshape Apple’s chip development strategy

The M7 Ultra reportedly reflects a broader shift in how Apple develops its processors. Rather than adding AI capabilities to chips designed mainly for general computing, the company may be allowing AI requirements to influence the overall architecture from an earlier stage.

Bloomberg reports that Apple has accelerated development of the M7 chip family as demand for more powerful AI computing continues to grow. The Ultra model is expected to sit at the top of the range and could offer AI performance closer to enterprise accelerators such as Nvidia’s Blackwell platform.

This would mark a change from Apple’s traditional focus on balancing processing speed, graphics performance and power efficiency across its Mac products. Those areas are still likely to remain important, but AI workloads could become a larger factor in decisions involving memory capacity, chip architecture and system design.

The move could also support Apple’s preference for processing more information directly on its devices. Running AI models locally may reduce delays caused by sending data to remote servers and could give users greater control over how information is processed. However, large and complex models require significant computing power and memory, necessitating more capable hardware.

Apple’s reported approach could provide advantages for developers and professional users working with generative AI, machine learning, scientific research and advanced media production. A Mac equipped with a high-capacity M7 Ultra could potentially handle larger models without requiring access to separate AI servers.

However, the processor remains several years away from its expected launch. Chip development plans can change due to manufacturing constraints, product decisions, or shifts in market demand. The reported 1.5 TB memory capacity may therefore differ from the specifications offered in commercial products.

Apple could enter a new level of AI competition

If the reported specifications are delivered, the M7 Ultra could move Apple into more direct competition with companies that provide specialised AI computing hardware. Nvidia has established a strong position in the AI accelerator market, with its Blackwell architecture designed for demanding generative AI and large-scale computing workloads.

Apple’s strategy would be different from Nvidia’s approach. Rather than selling dedicated accelerators primarily for data centres, Apple could integrate advanced AI performance into its own Mac hardware through a tightly integrated combination of custom processors, operating systems, and software tools.

The large unified memory capacity could be a significant advantage for certain AI workloads. Developers may be able to run larger models on a single system without dividing tasks across several devices. This could make high-end Macs more attractive to researchers, software developers and creative professionals who want powerful local AI capabilities.

The M7 Ultra could also expand the role of Apple’s most powerful desktop computers. High-end Mac systems have traditionally targeted professionals working in areas such as video production, software development and digital design. Stronger AI performance could add machine learning, generative media and local model development to that list.

Despite the reported comparison with Nvidia Blackwell, performance will depend on more than memory capacity. Processing speed, memory bandwidth, software support, power use and compatibility with popular AI development tools will also affect how the M7 Ultra performs in real-world applications.

Apple has not publicly confirmed the processor, its memory capacity or its expected release schedule. With a possible launch not expected until 2028, further details may emerge as the company develops future generations of Apple silicon.

If Bloomberg’s reported roadmap proves accurate, the M7 Ultra may become more than another yearly processor upgrade. It could signal a new phase in Apple’s chip strategy in which artificial intelligence becomes a central part of hardware development and places future Macs in closer competition with enterprise AI systems.