AWS signs a multi-year, $1B+ Synopsys deal to design custom AI chips in-house

AWS is deepening its commitment to in-house silicon. Simply Wall St reported a multi-year deal with Synopsys worth over $1 billion that makes AWS a leading customer for Synopsys IP aimed at AI data center chips. The agreement has both licensing and royalty components, which signals ongoing production use of Synopsys technology inside AWS's chips rather than a one-off design engagement.

The mechanics matter. Licensing gives AWS's Annapurna Labs access to proven IP blocks such as interconnects, memory interfaces and SerDes, so it does not need to design them from scratch. Royalties scale with shipped volume, which means Synopsys profits as Trainium and Graviton deployments grow. That aligns both companies with AWS's aim of offering cheaper alternatives to NVIDIA GPUs inside its own cloud. The deal follows Amazon's planned acquisition of DuckDB maker DuckLabs, part of a wider pattern of buying or locking in core technology.

The context is a custom-silicon race. Google's TPUs anchor Anthropic's new $125.2 billion compute commitment, financed by Broadcom. Microsoft and Meta are building their own accelerators. Synopsys is also working with OpenAI on GPT-Synopsys for chip design (168 HN points) and is integrating NVIDIA's agent safety platform into its EDA tools. AI is now designing the chips that run AI.

Customers face rising costs in the meantime. AWS GPU reserve prices rise 15% on October 7, the fourth consecutive quarterly increase, which strengthens the economic case for pushing workloads onto Trainium. Watch Trainium roadmap updates at re:Invent and whether the deal's royalty structure shows up in Synopsys guidance.