Google Unveils External TPU Sales via Oracle

Alphabet announced on Tuesday that Oracle will offer Google's tensor processing units to external customers. This marks the first time the custom chips, known as TPUs, will be available outside of Google's own data centers. The company made the disclosure the day before releasing its third-quarter earnings report.

The move represents a departure from a long-standing strategy. Google previously argued that customers benefited most from a full-stack approach where chips, models, and cloud services came from a single provider. Amazon and Microsoft have attempted to copy that model, but Google is now widening its hardware reach to compete in a market currently dominated by Nvidia.

Thomas Kurian, the head of Google Cloud, is driving this expansion. He joined Google in late 2018 after spending more than two decades at Oracle. Analysts expect his strategy to boost financial results, with Google Cloud revenue projected to rise approximately 90% from the same period last year. This follows an 82% increase in the second quarter.

Market reaction has been positive among investors. Chris Ballard, a managing director at Check Capital Management, noted that Google is a strong competitor capable of taking market share from Nvidia. Uday Cheruvu, a portfolio manager at Harding Loevner, said Google's mindset has shifted from serving only internal needs to serving external customers effectively.

Financial forecasts for the TPU business show steep growth. TD Cowen projects $9.4 billion in TPU sales for the second half of 2026 and $62.5 billion for 2027. Wells Fargo expects about $9 billion in sales for the current quarter and $99 billion for the following year. Anat Ashkenazi, the chief financial officer, stated in July that only a small portion of TPU revenue from existing agreements will be recognized this year, with the vast majority arriving in 2027.

Despite rapid growth, Google trails its largest rivals in total scale. Its second-quarter cloud revenue stood at $25 billion. Amazon Web Services generated $42 billion, while Microsoft Azure generated $29 billion in the same period. Google is growing faster, but selling chips externally introduces trade-offs, particularly regarding profit margins. TD Cowen estimates a 20% operating margin on TPU sales, compared with an overall cloud margin of about 35% last quarter.

Kurian must balance external sales with Google's internal efficiency. At a Goldman Sachs conference in September, he said AI servers pay for themselves in under two years. Servers running Google's own silicon reach payback in about half that time. SemiAnalysis estimates that the latest Ironwood TPU costs about 44% less to own and operate than an Nvidia GB200 server.

Anthropic is Google's largest TPU customer and a major AI competitor. The company agreed last October to use up to 1 million TPUs in a deal worth tens of billions of dollars. Mizuho states this agreement brings nearly a gigawatt of capacity online this year. CNBC has also confirmed that OpenAI plans to purchase TPUs from Google, following a report by The Information.

Anthropic is securing external financing to cover the cost of the chips. Broadcom detailed a $63 billion vehicle this week, backed by Blackstone and three large banks. The group will buy TPUs and networking gear from Broadcom and lease them to Anthropic for five years. William Blair analysts suggest this plan covers about half of Anthropic's 2027 needs, implying additional financing is likely.

In an IPO filing submitted on Monday, Anthropic stated it uses silicon from Google, Amazon, Nvidia, and Advanced Micro Devices. The company said it selects chips based on the best price-performance and efficiency characteristics for each workload. It added that its partnerships with Google and Broadcom provide access to compute capacity across multiple generations of the TPU platform.