M3 Ultra exposed: Inside Apple’s hybrid chip powerful enough to take on Nvidia

Just when we thought the M3 was in the ground and Apple had rapidly moved every single Mac model over to the fresh and new M4 processor, in rolls the high-end Mac Studio, equipped with the impossible-to-predict, never-before-seen M3 Ultra processor.

That’s right–in March 2025, a full 16 months after the M3 Pro and M3 Max chips arrived on the scene, Apple introduced a new chip in the old, nearly retired family. And it means that the new Mac Studio comes simultaneously in M3 and M4 varieties–with the M3 model being the high-end configuration

So confusing. I admit to being a bit baffled when I heard the news, too. But the more I think about it, the more I have come to grips with the fact that Apple’s chip strategy isn’t quite as straightforward as it would like us to believe. What if I told you that the M3 Ultra isn’t quite an M3 or an M4 chip? Follow me, friend, and let’s see how deep this rabbit hole goes.

M3 closeout sale

Here’s the story as we know it: Apple and its chipmaker, TSMC, were rightly proud when the M3 and A17 Pro chips arrived in 2023 as the first chips made by a new, state-of-the-art 3nm fabrication process. But as it turns out, TSMC was already working on a more efficient and economical second-generation 3nm process, making that first generation a bit of an architectural dead end. Therefore, everyone expected Apple and TSMC to rapidly turn from the M3 to the M4, which is based on the second-generation process and more efficient all around. That’s why Apple has spent the last few months updating almost every Mac from M3 to M4.

Petter Ahrnstedt

Except… there’s still the M3 iPad Air and the A17 Pro iPad mini, both of which use cut-rate versions of that older process. Nobody outside Apple and TSMC seems to know whether there are still portions of the old process running somewhere or if there’s a big bin of M3 and A17 parts that Apple knows will be enough to meet any demand for devices with the older chip architecture.

That was all weird enough, but the announcement of the M3 Ultra is something else. It’s a new chip, not some relic of the past. It’s a whole generation out of step with Apple’s current strategy, thawed out and dropped on an unsuspecting public like some sort of silicon Austin Powers.

Or at least that’s what you might believe if all you see is the big number 3 next to the capital M. But when you look closer at the M3 Ultra, you start to realize that all is not as it seems.

M3 or M3.5?

The M1 Ultra and M2 Ultra chips were essentially just two M1 Max and M2 Max processors, respectively, stuck together with Apple’s UltraFusion technology. Their specs were simply doubles of the smaller chips: twice the CPU cores, twice the GPU cores, twice the neural engines, and twice the maximum RAM.

But that’s not the case with the M3 Ultra, which is probably why it took Apple so much longer to release it than its sibling chips. (The M2 Ultra followed the M2 Pro and Max after just a few months, so 16 months is quite a wait.)

The M3 Ultra chip is very much modeled on the M3 Max and offers twice as many CPU and GPU cores, Neural Engine cores, and the like. But look closer: The M3 Ultra offers a maximum of 512GB of RAM, four times the maximum 128GB of the M3 Max. And the M3 Ultra supports the faster Thunderbolt 5 specification, not the older Thunderbolt 4 of the rest of the family.

That means the M3 Ultra is an in-between chip, mostly based on the older M3 generation, but with a few high-end additions that push it up toward the M4 in terms of capability. In my conversations with Apple representatives, I’ve gotten the impression that these changes, including the upgrade to Thunderbolt 5, had an impact on the release date of the chip.

So why call it an M3 and not an M4? I think Apple has some pretty strict internal rules that determine chip naming, and they probably start with the design of the CPU cores. One look at the single-core performance of the M3 Ultra and you’ll see it performs more or less like any other M3 chip. Of course, you would never buy an M3 Ultra and use just a single processor core–the whole point is that you’ve got 20 or 24 of them, and 60 or 80 GPU cores, too.

Foundry

Speaking of maxing things out, that’s what the M3 Ultra is designed to do. Because it’s mostly from the previous-generation chip architecture, each part is slower–but if you load up all those CPU and GPU cores and take advantage of an enormous heap of RAM and the high bandwidth of Thunderbolt 5, the M3 Ultra Mac Studio will run rings around an M4 Max Mac Studio. It’s a faster chip–when it’s used to its fullest.

That’s why the M3 Ultra Mac Studio really is a product designed for a relatively small audience. Even most professional users will find it hard to make the M4 Max chip feel slow, but for certain applications–video, rendering, science, and AI–the enormity of the M3 Ultra’s spec sheet will make it worth its very high price tag.

As Ben Thompson pointed out this week, an M3 Ultra Mac Studio with 512GB of RAM could run the four-bit quantized version of the state-of-the-art DeepSeek R1 model on device. Even at a price tag of tens of thousands of dollars, the highest-end M3 Ultra Mac Studio might be a bargain compared to expensive Nvidia arrays–if you find yourself in need of running an enormous AI model locally. Almost nobody wants that… but if you’re an AI researcher, you sure might.

Some open questions

Still, the M3 Ultra’s arrival is weird enough to make it feel like we’re all missing part of the bigger picture. If there’s no other shoe to drop, why did the M2 Ultra-based Mac Pro not get its own M3 Ultra chip bump?

Sure, it might mean the Mac Pro is just a laggard, and it’s going to show up in June with the same chip inside. Or it might mean that Apple doesn’t care about the Mac Pro and it’s pining for the fjords.

But consider this: Apple has suggested that there won’t necessarily be an Ultra chip in every M-series chip generation. Okay, but… what if there’s something else? A few years ago, there were rumors of an even higher-end chip destined for the Mac Pro, but it was reportedly canceled. I wonder if maybe the future of the Mac Pro is hiding in plain sight, and the M3 Ultra is what it is because Apple’s going to use its high-end chip design bandwidth in the M4 or M5 generation for something a little larger, designed just for the Mac Pro.

(A chip that huge would almost certainly need to be inside the Mac Pro, because I doubt the Mac Studio enclosure and its cooling system would be capable of handling a chip that large and hot.)

I don’t know if a fancy ultra-high-end chip for the Mac Pro is in the cards, in this generation or any other. But if the M3 Ultra Mac Studio already has some intriguing implications for running enormous AI models locally, imagine what might be enabled by an even larger chip with even more onboard memory. It would be a feature set worthy of the name Mac Pro.