Arm's September 30 AMBA C2C announcement moves the chiplet boundary up the stack. The protocol carries familiar AMBA semantics across dies while remaining compatible with transports such as UCIe. The constraint being removed is not link bandwidth. It is the custom adaptation, coherence plumbing, and verification work required each time an SoC function moves onto a separate die.
That separation between function and transport gives design teams a more stable contract. CPU, accelerator, memory, and I/O chiplets can integrate around AMBA behavior while packaging teams choose the physical link. Arm also ties the approach to pre-silicon validation and off-the-shelf IP through Arm Total Design, which lets architecture and verification advance before every package decision is final. Chiplets become reusable components only when the interface above the PHY is reusable too.
Teams planning a disaggregated Arm design should prototype one partition against AMBA C2C before committing the package topology. The cost is an early modeling and interoperability exercise. By 2028, chiplet programs built around proprietary functional bridges will carry a visible schedule penalty against designs that can swap dies without rewriting the SoC contract.