Google Enlists Partners to Sell More AI Chips

Google is broadening its AI-chip strategy by selling its custom tensor processing units beyond its own cloud infrastructure, a shift that could expand its position in a market dominated by Nvidia. On Tuesday, one day before Alphabet's third-quarter earnings report, the company said Oracle will offer Google's TPUs, marking the first time customers can access the chips outside Google's data centers. The move follows Google opening TPUs to external data centers earlier this year and reflects cloud chief Thomas Kurian's push to pursue growth through partnerships, including with Oracle, where he previously worked for more than two decades. The strategy comes as Alphabet's cloud business accelerates. Analysts expect Google Cloud revenue to rise about 90% year over year in the third quarter, after growing 82% in the second quarter. Google still trails Amazon Web Services and Microsoft Azure in absolute revenue, with second-quarter cloud sales of $25 billion compared with $42 billion for AWS and $29 billion for Azure. Still, Alphabet's chip business is emerging as a key differentiator, with analysts warning that Google could eventually take market share from Nvidia. Financial estimates for the TPU business vary, but they point to significant growth. TD Cowen projects $9.4 billion in TPU sales in the second half of this year and $62.5 billion in 2027. Wells Fargo forecasts about $9 billion in third-quarter sales and $99 billion next year. Alphabet CFO Anat Ashkenazi said in July that only a small portion of TPU revenue from existing agreements will be recognized this year, with the vast majority expected in 2027. Google also plans to use more outside capacity during the third quarter while building out its own infrastructure. Google's largest TPU customer is Anthropic, one of the leading AI model developers competing directly with Alphabet's Gemini. Anthropic agreed in October to use up to 1 million TPUs in a deal worth tens of billions of dollars. Mizuho says the agreement will bring nearly a gigawatt of capacity online this year and describes the relationship as Google supplying silicon to a close model competitor. OpenAI is also planning to buy TPUs from Google, according to reports, reinforcing the chips' growing appeal among major AI developers. To help finance Anthropic's expansion, Broadcom, a key Google partner in TPU development, detailed a $63 billion financing vehicle backed by Blackstone and three large banks. The structure will buy TPUs and networking equipment from Broadcom and lease them to Anthropic for five years. Analysts estimate the financing covers roughly half of Anthropic's 2027 compute needs, suggesting further financing is likely. Anthropic, in its IPO filing, said it uses chips from Google, Amazon, Nvidia and AMD to select the best price-performance and efficiency characteristics for each workload. It also said its partnerships with Google and Broadcom provide access to compute capacity powered by Google TPUs across multiple generations. The broader shift shows Google moving from a vertically integrated cloud pitch toward a more open model in which its AI hardware, software and cloud services can be sold through partners and deployed outside its own cloud. By extending TPUs to Oracle and large AI labs, Alphabet aims to scale its chip business, diversify revenue and challenge Nvidia's dominance in AI infrastructure while still expanding its own capacity.