Apple Does Fusion.

For the first time in five generations of Apple Silicon, these chips are not a single piece of silicon. The newly announced M5 Pro and M5 Max use what Apple calls Fusion Architecture. This is a big structural change, with long-term implications. And you can see this at work in the newly announced flagship Apple laptops. On the surface these are two third-generation 3-nanometer dies, bonded together into one system on a chip. But dig deeper, and with this modular, scalable silicon approach, Apple is setting itself up to cash in on the computing needs of the AI future.

To understand what Apple launched today, we have to go back just over five years, when Apple launched the M1 in November 2020. I wrote then:

“This approach to integration into a single chip, maximum throughput, rapid access to memory, optimal computing performance based on the task, and adaptation to machine learning algorithms is the future — not only for mobile chips, but also for desktop and laptop computers.”

That turned out to be right. The rest of the industry would embrace the new approach. Qualcomm built the Snapdragon X. AMD reshaped its laptop chips. Even Intel overhauled its architecture. The M1 didn’t just change Apple. It changed how chips were being conceived and built.

So the natural question about Fusion Architecture is this. Is Apple doing it again? Is this another moment where Apple redirects the industry? The honest answer is no. That, however, doesn’t mean it is not interesting.

---

For years, Apple’s narrative around its “M-series” chips was about integration. One chip. One die. Everything on the same piece of silicon. Unified memory so the CPU, GPU, and Neural Engine could all access the same data without copying it around. It worked beautifully for the M1 and M2. But now with the rise of AI, chips need to get bigger. AI demands more cores, more memory bandwidth, more compute. So, making one really big honking chip gets really expensive.

The larger a single die gets, the harder it is to manufacture. One tiny defect anywhere on the silicon and you toss the whole thing. Yields drop. Costs climb. AMD’s CEO Lisa Su recently showed that a design using four smaller chiplets delivered more total capability at 59 percent of the cost of one big chip.

Apple, too, faced a fork in the road. Keep building bigger and bigger single chips. Or break the big chip into smaller pieces and connect them together fast enough that software barely notices the split. They chose the second option, but made it their own. They call it Fusion Architecture.

This is not a new idea in the industry. AMD has been doing this with its chiplet strategy across Ryzen and EPYC for years. Intel has used 3D stacking and bridge interconnects. Nvidia builds massive AI accelerators using multi-die packaging. The chiplet market is now roughly $40 billion a year, and nearly all data-center AI products are built this way. The era of the giant monolithic die is ending. Chip-heads agree that the future is modular silicon.

This approach comes with its own tradeoffs. You split a chip into pieces, the pieces need to talk to each other. That means data traveling between dies, which adds latency. Memory gets divided up between chiplets. You solve the manufacturing problem but you compromise the architecture. Apple decided to do it its own way.

Johny Srouji, who has led Apple’s silicon efforts for years, says Apple kept unified memory intact. In the press release he said Fusion Architecture would “scale the capabilities of Apple silicon while preserving its core tenets of performance, power efficiency, and unified memory architecture.” While Apple says unified memory is preserved across both dies, the technical details of how memory actually works across two dies versus one die aren’t spelled out.

This is different from the UltraFusion approach Apple took with the M1 Ultra and M2 Ultra, where it duplicated the same chip and welded two copies together. With Fusion Architecture, the two dies are functionally different. An analysis of the M5 Pro and M5 Max chips by Ars Technica hints at how the new modular Fusion Architecture works.

The M5 Pro and M5 Max share the same first die. It handles the 18-core CPU, the 16-core Neural Engine, the SSD controller, and the Thunderbolt ports. The second die is where the two chips differ. The M5 Pro gets up to 20 GPU cores, a single media engine, and a memory controller with up to 307 GB/s of bandwidth. The M5 Max gets up to 40 GPU cores, two media engines, and up to 614 GB/s. So in the future, Apple could create a new M chip that has two of the second die, versus just one, and give it more AI-oomph.

Srouji told me in 2020

“We’re developing a custom silicon that is perfectly fit for the product and how the software will use it. When we design our chips, which are like three or four years ahead of time, Craig and I are sitting in the same room defining what we want to deliver, and then we work hand in hand. You cannot do this as an Intel or AMD or anyone else.”

---

The M1 was the revolution. M2, M3, and M4 were refinement. Each generation added cores, bumped bandwidth, squeezed more from the process node. M5 is different. It is the first proof that the original M1 idea is durable enough to survive a fundamental change in how the chips are built today and in the future. The capabilities of the new design reflect that.

For example, while the core counts didn’t change, Apple put a Neural Accelerator inside every GPU core. M5 Pro still has 20 GPU cores. M5 Max still has 40. Same as M4. But each core now does double duty. That is how Apple claims 4x the AI compute without adding more silicon. The GPU is becoming an AI processor that sometimes does graphics.

Memory bandwidth keeps climbing. M5 Max hits 614 GB/s, up from 546 on M4 Max. M5 Pro reaches 307 GB/s, up from 273 on M4 Pro. This is the number that matters for running large language models locally. Apple is building these chips for a world where your laptop runs a 70-billion-parameter model.

The CPU is also a departure. The M5 Pro and M5 Max dropped efficiency cores entirely. For the first time, Apple’s pro chips are an all-big-core design. At the top, six “super cores” handle peak single-threaded performance. Below them, 12 new performance cores are optimized for multithreaded throughput. As John Gruber notes, the naming is confusing. What Apple used to call “performance” cores from M1 through M4 are now “super cores.” The new “performance” cores are a genuinely new design, a middle tier that didn’t exist before. By the way, the approach is very similar to AMD’s strategy with its Zen 5 and Zen 5c cores.

---

This is why I think Fusion Architecture is the real story.

Not because of what M5 Pro and M5 Max can do today. Because of what it opens up. Once you’ve proven you can split the chip and keep unified memory working across the pieces, the question changes. It is no longer “how big can we make this chip?” It is “how many pieces can we connect, and in how many dimensions?”

And consider this. The packaging technology Apple is using to bond these dies together is the same interconnect technology that powers AI servers in data centers. Apple put it in a laptop.

It is well known across the chip industry that it is becoming increasingly difficult to put more transistors on a piece of silicon. Apple is using Fusion Architecture as a way to answer that problem. Not by fighting the physics but by figuring out clever ways to tame it.

March 3, 2026. San Francisco.

---

My Previous Coverage of M Chips

• “Steve Jobs’s Last Gambit: Apple’s M1 Chip,” November 17, 2020.
• “My Notes on Apple’s M1 Chip,” November 10, 2020.
• “Apple Launches New (M2) Chips,” January 17, 2023.
• “With Its New M3 Chips, Apple Joins the AI Party,” October 30, 2023.
• “Some Thoughts About the M4 Chip,”May 13, 2024.