Originally published on The Daily Flare.
Google announced two additions to its eighth-generation Tensor Processing Unit lineup on Monday, November 9, targeting confidential computing and AI applications that need to process unusually large context windows.
The new systems, called TPU 8tx and TPU 8i Max, expand a range that already includes TPU 8t and TPU 8i. Google says the additions are designed for different workloads rather than as entirely new silicon designs. In an interview with The Straits Times, Google chip chief Amin Vahdat said the systems reconfigure existing processors to address requests from customers.
The 8tx is aimed at organisations handling sensitive information, including government agencies and companies in regulated industries. Google says the system encrypts data at the hardware level and pairs its TPU technology with processors from AMD. The company claims it can deliver up to twice the speed of the previous version with virtually no performance trade-off.
The 8i Max is designed for AI applications that need to consider large amounts of information in a single request, a capability known as a long context window. Google says it combines two chips in one package to provide more memory and computing power. Vahdat cited coding applications, mathematical reasoning and drug discovery as potential uses.
Google describes both announcements as new systems, a distinction it makes between the chip itself and the surrounding computing setup. The company says a system can bring together compute, networking and storage, with changes to packaging, memory, optical networking or security. It says its modular approach is intended to let engineers adapt configurations to emerging customer needs more quickly than a new silicon design can be produced.
The additions bring the eighth-generation portfolio to four systems, a first for Google in a single year, the company said. It introduced the 8t for large-scale training and the 8i for real-time reasoning at Google Cloud Next in April. Google said the 8t offers up to 2.7 times better training performance than its previous generation, while the 8i offers 80% better inference price performance.
Vahdat told The Straits Times that the lineup reflects a move towards more specialised AI hardware. “This really follows the thesis that we have here at Google, which is that the future is going to be about specialisation, and the upsides for specialisation continue to be significant.” Google said feedback from outside organisations has increasingly shaped its designs, including requests for systems suited to large context windows and specialised encryption.
Google says its TPUs now serve several roles: training and running its Gemini models, supporting agentic workloads, providing services to Google Cloud customers and, more recently, allowing customers to deploy the processors in their own data centres. The company says it has nearly 100 TPU customers, including frontier AI labs, trading firms, retailers and universities.
The products also strengthen Google’s effort to sell its custom processors beyond its own services. TPUs have mainly powered internal products such as Search, YouTube and Gemini, Reuters reported, while Google has been offering them through its cloud business to attract enterprise AI customers. Reuters said Alphabet shares rose more than 3% in early trading on November 9. The report also said TPU sales made a meaningful contribution to Alphabet’s September-quarter results, as Google Cloud revenue more than doubled to $30.71 billion.
The opportunity is drawing attention as Nvidia remains the dominant supplier of AI processors. The Straits Times reported that demand for Nvidia processors still exceeds the company’s ability to supply them, leaving room for Google and other competitors. Bloomberg Intelligence estimates, cited by The Straits Times, put Google TPU sales, including related software and hardware, at $12 billion in 2026 and project they could exceed $135 billion by 2028. That estimate excludes sales to customers using the chips through Google Cloud.
Google also says it is working to make TPUs easier for developers to use through support for PyTorch, open-source software and automated tools. The company presents that software effort alongside its modular hardware as a way to offer new configurations to developers and cloud customers.
Top comments (0)