Alphabet Rises More Than 3% as Google Announces Two New AI TPU Systems

Key Facts

Google announced two new TPU systems on November 9, TPU 8tx and TPU 8i Max, saying they bring to four the number of eighth-generation systems it has introduced in a year. Alphabet shares rose more than 3% in early trading that day, according to Reuters, which linked the announcement to the company’s efforts to challenge Nvidia’s dominance in the AI chip market. The announcement is part of Google’s push to expand use of its systems beyond its internal services and attract more demand to its cloud business. The company presented the systems as specialized additions to its TPU portfolio, not as evidence of two entirely new chip innovations. One focuses on confidential inference, while the other increases the resources of a system designed for real-time inference.

TPU 8tx targets regulated companies and government agencies that need to run inference while keeping data confidential. The system is designed to support agents performing multistep tasks, workloads that require continued processing while keeping data within more sensitive environments. Google says its latency can be up to two times lower than TPU 8t. The system is also paired with fifth-generation AMD EPYC host processors, according to the company’s announcement. This design gives 8tx a specialized role in the portfolio, focused on confidentiality and responsiveness in enterprise and government use cases.

TPU 8i Max combines two TPU 8i chips in a single package. Google says the design doubles compute capacity, SRAM memory and HBM bandwidth compared with TPU 8i. It is intended for cases requiring intensive inference processing and greater memory and compute resources. The company described the increase as simultaneous across these resources, so compute capacity does not rise without a corresponding increase in memory and bandwidth. The system thus offers a higher-capacity option within the 8i family Google unveiled earlier this year.

In April, Google introduced TPU 8t for large-scale training and TPU 8i for real-time inference, defining two distinct paths within the eighth generation. The company said at the time that 8t’s training performance was up to 2.7 times that of the previous generation, while 8i’s price-performance for inference had improved by 80% compared with the previous generation. 8tx adds a focus on confidentiality and latency to the portfolio, while 8i Max expands the resources of the system designed for inference. Google describes the four systems it launched during the year as a portfolio differentiated by workload type. These specializations put November’s announcement in the context of developing the existing TPU ecosystem and expanding the options available to customers.

Google says TPU systems have become a fast-growing business and that it had nearly 100 customers as of November 9. The customer base includes advanced research labs, trading firms, retailers and universities, according to the company. The list points to the breadth of use cases Google is targeting, from research lab needs to enterprise applications and commercial sectors. Reuters reported that TPUs were used mainly inside Google in products such as Search, YouTube and Gemini, while the company is working to make them available through its cloud business to attract enterprise customers. The shift means the commercial value of the systems is also tied to Google Cloud’s ability to turn demand for computing into paid services.

Results for the quarter ended in September offer an indication of the commercial impact TPU sales have begun to make. Alphabet said sales of the systems contributed materially to the quarter’s results, while Google Cloud revenue exceeded $30.71 billion after more than doubling, according to Reuters. The link between chips and cloud goes to the heart of Google’s strategy: using computing infrastructure to support its services, then making it available to customers through its cloud business. The information in the announcement does not specify how much TPU contributed to total cloud revenue, but it directly links sales to the quarter’s results. A customer base of nearly 100 provides a commercial foundation for the expansion the company announced in its products.

The market’s initial response was positive: Alphabet shares rose more than 3% in early trading on November 9, according to Reuters via Euronext. Reuters coverage via StreetInsider reported that the shares were up more than 2% in premarket trading, while they were down 9% year to date through the close of the previous session on November 6. The move remains an early-trading reading, not the final outcome for the session. EL7 data showed Alphabet closing at $284.78 and Nvidia at $186.64 on November 6, and did not include prices for the November 9 session. This places the premarket performance in the context of Alphabet’s decline since the start of the year, without settling the stock’s direction after the announcement during the session.