AMD Says Fifth-Generation EPYC Processors Power Google’s TPU 8tx Infrastructure
Key Facts
AMD said on November 9, 2026, that fifth-generation EPYC processors power the host infrastructure for Google’s cloud TPU 8tx system. The announcement puts AMD processors in the host role supporting Google’s accelerators, rather than in the role of the accelerator itself. AMD said the system combines TPU units with EPYC processors to run AI workloads. The company describes TPU 8tx as the first cloud TPU system to offer hardware-enforced confidential computing. AMD’s premarket trading on November 11 could not be verified, so no share-price move that day is attributed to the announcement.
In TPU 8tx systems, EPYC processors coordinate workloads and prepare data, while managing high-throughput communication with TPU accelerators, according to AMD. This makes the processor central to organizing the flow of work between data and accelerators, while TPU units perform specialized computing tasks. The company said the system uses the Secure Encrypted Virtualization (SEV) capabilities built into EPYC processors. According to AMD, this extends protection from the CPU to data connected to the accelerators. The company links this design to confidential computing at the system level, rather than limiting it to the processor alone.
A Google Cloud official said combining fifth-generation AMD EPYC processors with eighth-generation TPU infrastructure enables hardware-enforced confidential computing in cloud TPU units for the first time. The statement indicates that the announcement concerns a security feature in the infrastructure’s design as well as its technical operation. The official described integration between the processors and TPU infrastructure as a way to bring this protection to the AI Hypercomputer environment. AMD says using SEV capabilities with TPU units extends protection to data connected to the accelerators. The announced partnership centers on integrating components from both companies within the cloud system.
Google said TPU 8tx uses the same chip as TPU 8t but is configured for secure, confidential inference and also supports small-scale training. It added that the system’s low-latency Boardfly network provides up to 2x lower latency than TPU 8t, with a near-zero performance penalty. These specifications place 8tx among the eighth-generation systems that Google dedicates in part to inference and training tasks. The company also said TPU 8t and TPU 8i deliver up to 2.7 times better training performance and 80% better inference price-performance, respectively, compared with the previous generation. The latter comparison figures apply to 8t and 8i together and do not refer to TPU 8tx alone.
Google said TPU units have become a fast-growing business for the company, with its customer base approaching 100. That base includes leading labs, trading firms, retailers and universities, according to the company. This places the AMD collaboration announcement within the broader expansion of TPU use among different types of organizations. Against a backdrop of demand for cloud computing services, Alphabet said TPU sales contributed materially to its September 2026 quarter results. Reuters reported that Google Cloud revenue more than doubled in the quarter to $30.71 billion.
As background to AMD trading, an Investing.com report on November 11 said Oppenheimer raised its rating on the stock to Outperform from Perform and set a $600 price target. The report said the development was part of momentum in the previous session. That alone does not establish AMD’s premarket performance on November 11. The technical collaboration announcement itself came from AMD on November 9 and detailed the roles of EPYC processors and TPU units, as well as confidential-computing features. The available evidence did not include a specific later date for the system’s launch or for an update from either company.