Synopsys and Amazon sign $1bn multi-year IP agreement for custom chip development - Data Center Dynamics

This $1bn multi-year IP lock-in is Amazon closing the strategic loop on x86 decoupling — pushing the shift from procurement to design. Graviton already absorbs roughly 70% of AWS general-purpose compute, but the real margin lever sits in AI silicon. Next-gen Trainium and Inferentia SoCs demand HBM3 controllers, high-speed SerDes, and on-chip NoC — precisely where Synopsys holds its moat. The critical inference: Amazon is migrating from an "Arm core + proprietary accelerator" assembly model toward full-stack SoC ownership. Within 18 months, AWS will accelerate x86 server retirement, creating structural headwinds for Intel and AMD's cloud revenue base — not cyclical, but architectural. On the competitive front, Cadence's IP licensing book feels the most direct pressure. More subtly, Arm's per-core royalty model is being eroded by hyperscaler leverage; licensing terms will likely be renegotiated under volume pressure. The 12–24 month tail: Synopsys's anchor pricing will ratchet up industry-wide IP benchmarks. Microsoft's Maia and Google's TPU teams will almost certainly follow with comparable long-term agreements, re-rating the EDA/IP sector from "tooling vendor" to "strategic infrastructure." And every one of these custom dies ultimately converges on TSMC's N3/N2 lines in Taiwan, China — making advanced-node capacity, not IP, the true binding constraint.