Google is leaning on Advanced Micro Devices (AMD) to help develop a future version of its custom artificial intelligence chip, according to the research firm SemiAnalysis.
The chip in question is Google's tensor processing unit, or TPU, a processor the company designs specifically to train and run AI models.
A future version could draw on AMD's central processing unit designs, as well as its work in chip packaging and networking, the report said.
The collaboration would be notable because Google has more often been linked with Broadcom, Marvell or Intel for work of this kind.
Intel in particular has been the name most frequently mentioned for such ambitious projects in recent months.
AMD, best known as Nvidia's closest rival in AI accelerators, would gain a further foothold in the market for custom chips if the tie-up were confirmed.
The speculation underlines the growing value of two things: the ability to design application-specific integrated circuits, known as ASICs, and ownership of reusable chip building blocks, or IP blocks.
ASICs are chips tailored to a single task, in contrast to the general-purpose processors that dominated earlier computing.
In a market constrained by manufacturing capacity, analysts argue the race to build silicon is not a zero-sum contest, with room for several designers and suppliers to benefit at once.
Neither Google nor AMD has confirmed the report.
