The global chip manufacturing capacity remains tight, and major companies are seeking foundry partners other than TSMC. In this context, Google is considering collaborating with Samsung to produce its next-generation AI-specific chips.
According to reports, Google's new tensor processing unit (TPU), codenamed "Icefish," is its tenth-generation TPU product, mainly serving cloud data centers. According to the plan, the chip will adopt a divisional manufacturing model: the core computing engine will be produced by TSMC using a 1.4nm process; the memory input/output die will be manufactured by Samsung using a 2nm process. The "Icefish" chip is still in the design phase, and Google has joined forces with MediaTek in the chip design. The product is expected to be mass-produced as early as 2028, and the current cooperation plan still has room for adjustment.
In addition to securing Google's orders, Samsung has recently won other well-known companies' chip foundry business, including Tesla's new AI6 chip and a new language processor used in NVIDIA's "Vera Rubin" platform. Its foundry business is continuously expanding.
Key Points:
📊 Affected by the global chip shortage, Google has chosen to expand its foundry channels and plans to collaborate with Samsung to produce its next-generation TPU chip.
🔧 Google's tenth-generation TPU "Icefish" adopts a divisional production model, with the core developed by TSMC and some components outsourced to Samsung. It is expected to be mass-produced in 2028.
🤝 Samsung has recently secured multiple chip foundry orders, with clients including Tesla, NVIDIA, and other tech companies.
