June 12, 2026 – Google is poised to diversify its chip manufacturing strategy by tapping Samsung Electronics for a slice of its next-generation AI processor production, marking a significant shift away from relying solely on TSMC.
According to sources familiar with the matter, Samsung’s 2nm process node has been earmarked to produce select components of Google’s upcoming tenth-generation Tensor Processing Unit (TPU), internally codenamed “Icefish.” However, the most computationally intensive portion of the chip — the core compute engine — will remain in TSMC’s hands, fabricated using its more advanced 1.4nm technology. Samsung is expected to handle the memory I/O die, a division of labor that highlights TSMC’s continued dominance in the most cutting-edge semiconductor work.
The “Icefish” TPU, designed specifically for Google’s cloud data centers, is still in the design phase and may not enter mass production until 2028, though timelines remain fluid. Notably, Google has expanded its design partnerships beyond longtime collaborator Broadcom, bringing in Taiwan-based MediaTek to co-develop portions of the chip — a move that began last year.
For Samsung, securing even a partial role on the “Icefish” project carries outsized strategic value. The Korean chipmaker has been aggressively building its foundry credentials, having already won Tesla’s next-gen AI6 chip last year. Samsung is also slated to manufacture the novel language processor powering Nvidia’s forthcoming “Vera Rubin” platform, a chip engineered to boost inference performance and energy efficiency.
The broader backdrop is clear: with chipmaking capacity under persistent strain, major tech players are actively hedging their bets by spreading orders across multiple foundries — and Samsung is emerging as a key beneficiary of that realignment.
