Google May Tap Samsung Foundry for Its Next AI TPU

Samsung makes some of the world’s best semiconductor memory chips, and it may soon land a contract from Google. The latest rumors suggest that Samsung Foundry could be selected to manufacture Google’s next TPU, a deal that would bring significant revenue.

Google execs visit Samsung’s Texas fab to discuss TPU supply volumes

Analyst Jukan reported on X citing local media that Google executives visited Samsung’s Taylor, Texas, fab to discuss potential TPU supply volumes. Google developed the TPU in collaboration with Broadcom. The chip is expected to be roughly 80% cheaper than Nvidia’s H100 while delivering comparable, or possibly better, performance.

Google’s TPUs are among the most important custom AI accelerators in the industry. They are built to power everything from Search and YouTube to Gemini and internal cloud workloads. Winning that business would not only deliver volume for Samsung. It would also validate its foundry process at a time when it is struggling to match TSMC’s reputation for leading-edge yields.

Google has historically relied on TSMC to manufacture most of its custom silicon, including earlier TPU generations. Even a partial shift toward Samsung would point to meaningful improvements in Samsung Foundry’s advanced nodes. It would also align with Google’s broader strategy of diversifying suppliers to reduce reliance on any single chipmaker.

The TPU is co-developed by Google and Broadcom, combining Google’s AI architecture expertise with Broadcom’s experience in custom ASIC design. Reports suggest the upcoming TPU could be roughly 80% cheaper than Nvidia’s H100 while offering comparable, or in some workloads superior, performance. That cost advantage is increasingly important as hyperscalers look to rein in AI infrastructure spending without sacrificing compute capability.