Apple’s most powerful Mac has arrived, and its most interesting target may no longer be Hollywood.
The new Mac Studio is available starting Sept. 22, with buyers choosing between M5 Max and M5 Ultra. The M5 Ultra version starts at $5,499. Apple is positioning the machine for demanding creative work, but its support for up to 512GB of unified memory, massive local AI models, and clusters of multiple Mac Studios reveals a bigger ambition: turning the desktop into serious AI infrastructure.
Apple says the M5 Ultra delivers up to 4.3 times the peak AI compute performance of the M3 Ultra and up to 1.8 times faster graphics performance than the prior generation, depending on the workload.
Much of today’s enterprise AI infrastructure still revolves around discrete GPUs, servers, and cloud compute. Apple is making a different bet: some increasingly sophisticated AI workloads can stay on the desk.
M5 Ultra gives Apple a bigger local AI machine
The M5 Ultra is where that argument gets serious.
The base M5 Ultra Mac Studio combines a 30-core CPU and 64-core GPU, while the top configuration scales to a 36-core CPU and 80-core GPU. More important for large AI models, M5 Ultra supports up to 512GB of unified memory and 1.2TB/s of memory bandwidth.
Apple says that memory capacity allows users to run enormous large language models entirely on the device. Its GPU also brings Neural Accelerators to the Ultra chip for the first time, contributing to Apple's claim of up to 4.3 times the peak AI compute performance of M3 Ultra.
The specifications sharpen a story eWeek identified after WWDC: Mac Studio is evolving beyond its traditional role as a compact workstation for video editors, designers, and other creative professionals. It is becoming Apple's clearest attempt to make high-end local AI a reason to buy a Mac.
There is now substantially more horsepower behind that pitch.
An early, unverified Geekbench 7 result for the 36-core M5 Ultra produced scores of 3,774 in single-core testing and 52,516 in multi-core testing. MacRumors calculated that at roughly 30% and 35% higher, respectively, than average results for the high-end 32-core M3 Ultra.
The result remains unverified and should be treated as an early indicator rather than definitive performance testing.
Four Mac Studios turn Apple’s desktop into AI infrastructure
The bigger shift becomes apparent when one Mac Studio isn't enough.
Thunderbolt 5 and Remote Direct Memory Access, or RDMA, allow multiple Mac Studio systems to be connected into a cluster. Apple says the machines can create a shared memory pool for particularly large open-weight models.
A cluster of four Mac Studios can deliver up to three times faster distributed AI inference than one machine, according to Apple. That changes the Mac Studio story.
Apple isn't merely selling a faster desktop for an individual developer or filmmaker. It is providing a way for enterprises, researchers, and AI teams to scale local compute by connecting multiple Macs.
That doesn't suddenly make Mac Studio a substitute for GPU-based data center infrastructure. Software ecosystems, compatibility, scalability, and cost remain critical considerations for enterprise AI deployments.
But clustering gives Apple a more credible position than either a single AI workstation or the cloud. Teams that want models and sensitive datasets to remain local can potentially add compute without immediately sending those workloads to external infrastructure.
That local-compute proposition is already showing up in enterprise AI deployments. Latham & Watkins recently bought Nvidia servers to keep some sensitive AI workloads out of the cloud, illustrating how organizations are weighing local infrastructure against cloud compute as their AI requirements grow.
For Apple, keeping more inference on-premises could become a differentiator rather than merely a technical feature.
The surprise upgrade isn't necessarily M5 Ultra
For all the attention on Apple's new processor, early reviews suggest one of the biggest everyday improvements may be hiding inside the storage subsystem.
Apple advertises up to twice the storage performance of the previous Mac Studio. But Macworld testing highlighted by MacRumors produced an even larger result in one configuration.
Macworld tested an M5 Max Mac Studio with 128GB of unified memory and a 4TB SSD using Blackmagic's Disk Speed Test. It measured roughly 2.6 times faster read speeds and 2.1 times faster write speeds than the previous-generation M4 Max Mac Studio used for comparison.
Those results come from one machine and benchmark, so they shouldn't be treated as universal. They do illustrate something important about Apple's workstation strategy, however: not every meaningful performance gain requires the $5,499 M5 Ultra.
The M5 Max Mac Studio starts at $2,499 and still gets the new storage architecture, Thunderbolt 5, Wi-Fi 7, Bluetooth 6, and up to 128GB of unified memory.
What eWeek found: Apple is turning the Mac into an AI endpoint
The $5,499 starting price makes M5 Ultra a specialized machine. Its significance is bigger than the number of people likely to buy one.
Apple is building an increasingly complete local AI stack around its own silicon, unified memory, Neural Accelerators, macOS, and Apple Intelligence. Mac Studio is the extreme end of that strategy, where models that would overwhelm ordinary personal computers can potentially remain on-device.
The addition of clustering pushes the idea further. One Mac can function as an AI workstation. Four can begin behaving more like local AI infrastructure.
That gives enterprises another architecture to consider as they decide which workloads belong in public clouds, private infrastructure, or directly on high-end endpoints. Privacy-sensitive development, model experimentation, inference, and work involving proprietary datasets are areas where Apple's approach could attract attention.
There are limits. Apple still needs its software ecosystem to keep pace with the hardware, and its own performance claims need broader independent testing. A four-machine M5 Ultra deployment also turns a $5,499 workstation into a much more expensive infrastructure decision.
There is also an availability wrinkle for teams eyeing Apple's maximum memory configuration. The 512GB unified-memory option, which requires the higher-end M5 Ultra with a 36-core CPU and 80-core GPU, won't be available until late October.
But the arrival of M5 Ultra makes the direction clearer. Apple has been increasing its focus on AI across its hardware and software ecosystem, a strategy that has also led to changes in the leadership overseeing its AI efforts. Mac Studio now represents the extreme hardware end of that push.
The question surrounding Apple's most powerful Mac is therefore becoming less about whether a desktop can replace a traditional workstation. It's whether Apple can carve out a meaningful place for the Mac in an AI market built overwhelmingly around GPUs, servers, and the cloud.
Also read: Apple Reportedly Accelerates M7 Chips for Major AI Upgrade for a look at how Apple’s future silicon could push local AI even further.


