Google is talking with Samsung Electronics about making part of its next-generation AI processor, a move that could trim the company’s reliance on TSMC and give Samsung a rare win in advanced chip manufacturing. The chip in question is Icefish, a future TPU used to run Google’s own AI systems and cloud services, and the split would keep the main die at TSMC while shifting a key memory-connectivity component to Samsung’s 2-nm process.
If that sounds like supply-chain hedging, that’s because it is. Google’s TPU program has become too important to leave vulnerable to a single fab partner, and the company also gets another quiet benefit: more room to reduce dependence on Nvidia’s GPUs, which still dominate AI compute. The Icefish talks are also being watched closely because they could reshape how Google sources its AI chip production.
What Samsung could make for the Google TPU Icefish
According to The Information, Samsung would not take over the whole chip. Instead, it could manufacture the component that links the processor to memory, a smaller but strategically valuable piece of the design. The project is being developed with MediaTek and is still at the design stage, with mass production not expected before 2028.
That detail matters because Samsung has spent years trying to prove its foundry business can compete with TSMC at the high end. It has invested heavily in new lines and cutting-edge nodes, but large marquee wins have been harder to land than the company would like. A Google TPU component would not close the gap overnight, but it would be the kind of customer logo Samsung can point to in every sales pitch for the next few years.
Google’s TPU strategy is getting wider
The Icefish talks also follow reports that Google is exploring Intel for future TPU production. Put together, that suggests a deliberate move toward a more distributed manufacturing setup rather than the old ”pick one giant supplier and hope for the best” model. In a market where AI capacity is becoming a competitive weapon, resilience is now part of the product plan.
- Chip family: TPU, code-named Icefish
- Main manufacturer: TSMC
- Possible Samsung role: memory-connectivity component
- Samsung process: 2-nm
- Mass production: not earlier than 2028
Why this would matter for Samsung
Samsung has long wanted to turn its foundry arm into a real alternative to TSMC, not just a promising backup option. A Google order would not just be a revenue line; it would be a signal that one of the most demanding buyers in AI hardware sees Samsung as good enough for advanced work. That is the sort of validation foundry businesses live and die on.
Neither company has confirmed the talks. Samsung declined to comment, and Google did not respond promptly, which leaves this in the familiar category of ”real enough to shape strategy, not real enough to invoice.” Still, if the deal lands, Google gets more flexibility for its AI infrastructure, and Samsung gets a chance to move from contender to serious option in the premium chip race.