Overview
Google is reported to be in negotiations with Samsung Electronics to manufacture a specific component of its next-generation artificial intelligence chip. The discussions focus on having Samsung produce the memory input-output die for Google’s 10th-generation Tensor Processing Unit, which carries the internal code name Icefish.
Technical split in production
According to people familiar with the matter, the current design framework would split manufacturing responsibilities between foundries. Taiwan Semiconductor Manufacturing Co. would remain responsible for the most demanding portion of the TPU - the compute engine - using its 1.4 nanometer process technology. Samsung would be tasked with making the die that links the computing element to memory, using its 2 nanometer production node.
Timeline and partners
Sources indicate the chip could move to mass production as early as 2028. Google is working with Taiwanese chip designer MediaTek on the project. The arrangement represents a notable change from Google’s historical reliance on a single foundry for its chipmaking needs.
Market context
The talks come amid a manufacturing environment in which TSMC is experiencing constrained capacity. That shortage has been attributed to heavy demand for AI chips, particularly from firms supplying data-center accelerators.
Key points
- Google is negotiating with Samsung to produce a memory I/O die for its 10th-generation TPU, code-named Icefish.
- Under the present plan, TSMC would continue producing the compute engine on a 1.4 nm node while Samsung would use 2 nm technology for the memory interface component.
- The chip could enter mass production as early as 2028, and Google is collaborating with MediaTek on the design effort.
Risks and uncertainties
- Negotiations are ongoing - the final manufacturing arrangement is not confirmed and could change.
- TSMC’s reported capacity shortages create uncertainty around supply timelines and could affect production pacing.
- The division of production between multiple foundries introduces integration and supply-chain coordination risks for high-performance chip assembly.
Impacted sectors
The developments touch the semiconductor manufacturing sector, cloud and AI hardware supply chains, and companies providing AI accelerators and data-center computing capacity.
Closing
The discussions mark a potential shift in Google’s chip production approach by engaging multiple foundries to deliver different functional elements of a single AI processor. How the negotiations resolve, and whether production timing holds to the 2028 target, will determine the practical effect on Google’s hardware roadmap and broader foundry demand dynamics.