Apple taps TSMC for next-gen M5 chips

Everyone assumed Apple would be the first in line to eat the massive bill for TSMC's upcoming 2nm process. It turns out even the richest company on earth has a budget limit.

According to Korean outlet The Elec, citing Taiwanese supply chain reports, Apple has officially placed orders with TSMC for its next-generation M5 processor, kicking off early production development.

Instead of making the expensive leap to a 2nm node, Apple is sticking with TSMC's mature 3nm manufacturing process for the M5. But that doesn't mean it will be a boring upgrade.

Sticking with 3nm, betting big on 3D stacking

Apple's decision to skip 2nm for the M5 comes down primarily to skyrocketing manufacturing costs.

To make sure the M5 still delivers a noticeable leap over the current M4 generation, Apple is leaning heavily into packaging technology. Specifically, it is adopting TSMC's System on Integrated Chip (SoIC) 3D stacking architecture.

Unlike traditional 2D multi-chip designs, SoIC stacks chip components vertically. This structural shift drastically improves thermal performance and reduces electrical leakage, which has become one of the biggest bottlenecks in modern mobile processors.

Apple has also expanded its work with TSMC on hybrid SoIC packages that incorporate thermoplastic carbon fiber composite molding. That complex packaging setup reportedly entered small-scale trial production back in July.

From MacBook Pros to AI data centers

Production of the M5 is slated to begin as early as the second half of 2025, with hardware arriving between late 2025 and early 2026.

If historical cadence holds, the MacBook Pro could see M5 configurations by late 2025, while the iPad Pro could arrive around late 2025 or early-to-mid 2026. A refreshed MacBook Air is expected in early 2026, and an updated Apple Vision Pro headset could drop anywhere between fall 2025 and spring 2026.

M5 references have already surfaced in Apple's internal code. More notably, sources say Apple designed this modular SoIC architecture with a dual purpose: deploying M5 variants straight into its AI server infrastructure to power cloud-based Apple Intelligence features alongside consumer gear.

Why it matters

For power users and creative pros, better thermals and reduced leakage are the real prizes. The current M-series chips are fast, but keeping sustained peak performance without thermal throttling inside ultrathin chassis remains tricky. Still, if you are shopping for a Mac today, M4 machines are more than capable enough for almost any workflow; waiting a full year or more for M5 is only sensible for those planning their next multi-year upgrade cycle.

For the wider tech industry, this is a clear signal that the economics of the leading edge are breaking. When Apple balks at the cost of a new node and pivots to 3D packaging on existing nodes instead, it proves that Moore's Law is now officially an advanced packaging race.

We think using the same core silicon for consumer flagships and in-house AI servers is a brilliant hedge. It amortizes high R&D tooling costs across two massive deployment pipelines while cutting Apple's reliance on commercial AI accelerators. The real test will be TSMC's yield rates on hybrid SoIC, because packaging bottlenecks can easily throw an entire hardware roadmap off schedule.