Google is in preliminary talks with Samsung Electronics to manufacture a critical piece of its next-generation AI chip, a move that could reshape the pecking order in the semiconductor foundry business.
The negotiations center on a memory input-output die for Google’s upcoming Tensor Processing Unit, codenamed Icefish. Samsung would produce the component using its advanced 2-nanometer process technology, one of the most cutting-edge fabrication methods in existence.
Why Google is looking beyond TSMC
For years, TSMC has been the go-to manufacturer for virtually every major tech company designing custom silicon. Google’s TPUs have been no exception. But capacity constraints at TSMC are forcing Google to think creatively about its supply chain.
Google’s TPUs have been powering AI workloads across its data centers since 2016. Nearly a decade in, the company’s appetite for custom silicon has only grown as it pours resources into generative AI, search improvements, and cloud computing services.
