Oracle to Resell Google TPUs, Targeting $9.4 Billion in Revenue

Alphabet announced on Tuesday that Oracle will sell Google's custom artificial intelligence chips directly to customers.

The move marks the first time tensor processing units, or TPUs, will be available outside of Google's own cloud infrastructure. Alphabet made the disclosure one day before its third-quarter earnings report. The strategy aims to capture a larger share of the AI chip market, a sector currently dominated by Nvidia. Cloud chief Thomas Kurian, who spent more than 20 years at Oracle before joining Google in 2018, is leading the push to expand hardware sales through partners.

Analysts view the shift as a departure from Google's traditional approach. The company previously sold a unified stack of chips, models, and cloud services. Chris Ballard, a managing director at Check Capital Management, noted that Google could take market share from Nvidia. Uday Cheruvu, a portfolio manager at Harding Loevner, said Google's previous focus on internal efficiency hindered customer service. That mindset has changed. Google Cloud revenue is expected to grow about 90% year over year, following an 82% increase in the second quarter. TPUs are a key driver of this growth.

Market estimates for TPU sales show wide variation but indicate rapid expansion. TD Cowen projects $9.4 billion in TPU sales for the second half of 2026 and $62.5 billion for 2027. Wells Fargo expects approximately $9 billion in third-quarter sales and $99 billion next year. CFO Anat Ashkenazi stated in July that most TPU revenue from existing deals will not be recognized until 2027. Google also plans to use outside capacity in the third quarter while it builds its own infrastructure.

Google Cloud generated $25 billion in revenue in the second quarter. That figure trails Amazon Web Services, which earned $42 billion, and Microsoft Azure, which brought in $29 billion. Despite the smaller base, Google is growing faster than its rivals. Ballard observed that Google is accelerating into an expanding market. His firm held about $236 million in Alphabet stock as of June 30.

The strategy carries financial trade-offs. TD Cowen estimates a 20% operating margin on TPU sales, lower than the roughly 35% cloud margin reported last quarter. Cheruvu does not expect cloud margins to rise much beyond the mid-30s. SemiAnalysis estimates that Google's latest Ironwood TPU costs about 44% less to own and operate than an Nvidia GB200 server. Kurian said at a September conference that Google's AI servers pay for themselves in under two years, with internal silicon reaching payback in half that time.

Major AI companies are central to this hardware demand. Anthropic, a competitor to Google's Gemini models, agreed last October to use up to 1 million TPUs in a deal worth tens of billions of dollars. Mizuho estimates this agreement brings nearly a gigawatt of capacity online this year. OpenAI is also planning to purchase TPUs from Google, CNBC confirmed after The Information reported the intentions. Anthropic uses silicon from Google, Amazon, Nvidia, and Advanced Micro Devices. The company selects chips based on price-performance and efficiency for specific workloads.

Anthropic is securing financing for its hardware needs. Broadcom, a partner in TPU development, detailed a $63 billion vehicle backed by Blackstone and three major banks. This fund will buy TPUs and networking gear from Broadcom and lease them to Anthropic for five years. William Blair analysts say this financing covers about half of Anthropic's 2027 needs. Anthropic stated in its IPO filing that partnerships with Google and Broadcom provide access to compute capacity across multiple generations of the TPU platform.

Google declined to comment on the new Oracle arrangement. The company will report its full third-quarter financial results on Wednesday. The announcement positions Google to compete directly with Nvidia through a new distribution channel. Oracle will handle sales to external clients while Google retains ownership of the technology. The expansion allows Google to serve customers who may not use its full cloud platform. This move could alter the competitive dynamics in the artificial intelligence infrastructure market.