Samsung is in talks to manufacture a portion of Google's upcoming Icefish tensor processing unit (TPU), according to Tech Times — a move that would mark a notable shift in how the world's largest technology companies source their custom AI chips.
For years, TSMC, the Taiwanese contract chipmaker, has been the dominant — and often only — choice for cutting-edge AI silicon. Google, Microsoft, Amazon, and other hyperscalers have relied on TSMC's advanced fabrication processes to build the chips that power their data centers and AI services. But that concentration carries risk: geopolitical tensions around Taiwan, capacity constraints, and pricing leverage have all pushed large cloud companies to look for alternatives.
By potentially splitting production of the Icefish TPU between TSMC and Samsung, Google would be following a supply-chain diversification playbook more common in consumer electronics than in advanced AI hardware. Samsung operates its own foundry business and has been aggressively courting hyperscaler customers, though it has historically trailed TSMC in yields and process maturity at the leading edge.
The Icefish TPU is Google's next-generation chip designed to train and run large AI models — the workloads that have become central to the company's products and its competition with Microsoft and Amazon in cloud services.
If the deal proceeds, it matters because it signals that the AI chip supply chain — long treated as a single-vendor dependency — may finally be diversifying, which could reshape pricing power, reduce geopolitical risk, and open the door for Samsung to become a serious player in the AI infrastructure race.