Google is rolling out two fresh iterations of its proprietary AI accelerators, the tensor processing units (TPUs), according to reporting from Bloomberg News on Monday. The new models carry the designations TPU 8tx and TPU 8i Max. According to Amin Vahdat, Google's chief technologist for AI infrastructure, these represent adaptations of proven chip designs, engineered in response to specific workload requirements voiced by the company's customers.
The TPU 8tx targets applications demanding security and confidentiality, particularly those serving government bodies and enterprises operating under strict regulatory frameworks. This variant operates in tandem with processors manufactured by AMD, reflecting a collaborative arrangement between the two firms, Bloomberg reported.
The TPU 8i Max takes a different approach, concentrating on AI systems requiring extended context windows—the ability to process substantial volumes of text or information within single queries.
Both variants descend from Google's eighth-generation TPU family, comprising the TPU 8t and TPU 8i, which debuted at the company's Cloud Next conference in April. The 8t specializes in model training, whereas the 8i handles inference on already-trained models. Google indicated both would reach general availability by year's end.
Historically, Google released a single new AI processor biennially, Bloomberg noted. The company shipped two this year, and these fresh variants bring the eighth-generation family to four distinct chips. According to Vahdat, "each one is tuned for different needs."
TPUs originated as internal infrastructure for Google's own operations before becoming a major competitor to Nvidia's graphics processors in the AI computing arena. Rather than selling TPUs as discrete products, Google distributes them via its cloud platform. Last year, Anthropic committed to deploying up to 1 million units.
Google is simultaneously broadening its manufacturing ecosystem. The company partners with four design firms, including Broadcom and MediaTek, in the creation of its TPUs.
These new variants align with industry movement toward AI agents—autonomous systems executing tasks on behalf of users—while addressing mounting security concerns as organizations deploy such technologies at scale, Bloomberg reported.
Source: The Next Web
