Oracle to Sell Google Chips as Cloud Revenue Surges
Alphabet will let its partner offer tensor processing units to outside customers, a move analysts say could challenge Nvidia's dominance in AI hardware.
By Petra Murphy ·
Alphabet announced on Tuesday that Oracle will sell Google's tensor processing units, or TPUs, to external clients. This marks the first time the custom chips, developed since 2013, will be available outside Google's own cloud infrastructure. The announcement comes one day before Alphabet releases its third-quarter earnings report.
Google has traditionally sold its chips, models, and cloud services as a single stack, a strategy rivals Amazon and Microsoft have attempted to copy. Cloud Chief Thomas Kurian, who joined the company in late 2018 after two decades at Oracle, is now seeking new growth avenues by pushing hardware to partners. Analysts expect Google Cloud revenue to rise roughly 90% year over year, a pace that exceeds the 82% growth recorded in the second quarter.
The shift reflects a change in strategy, according to Uday Cheruvu, a portfolio manager at Harding Loevner. He noted that while Google was once self-contained, the company is now better positioned to serve external customers. Chris Ballard of Check Capital Management described Google as a strong competitor that could potentially take market share from Nvidia.
Revenue projections for TPU sales vary among firms. TD Cowen estimates $9.4 billion in TPU sales for the second half of 2026 and $62.5 billion for 2027. Wells Fargo anticipates approximately $9 billion in third-quarter sales and $99 billion next year. CFO Anat Ashkenazi told investors in July that most TPU revenue from existing contracts will be recognized in 2027, with only a small portion hitting the books this year.
Google opened its TPUs to outside data centers in the second quarter, a step taken before the Oracle deal. The company reported $25 billion in cloud revenue for that period, trailing Amazon Web Services at $42 billion and Microsoft Azure at $29 billion. Despite the gap, Google is the fastest-growing major cloud provider.
Kurian must balance external sales with internal needs. He stated in September that servers running Google silicon reach payback in half the time of standard AI servers, attributing this to full-stack optimization. SemiAnalysis estimates the latest Ironwood TPU costs 44% less to operate than an Nvidia GB200 server. However, external sales carry lower margins, with TD Cowen estimating a 20% operating margin for TPUs compared to Google's overall cloud margin of about 35%.
Major AI firms are already buying the hardware. Anthropic, which uses Claude models, agreed last October to use up to 1 million TPUs in a deal worth tens of billions of dollars. OpenAI is also planning to purchase TPUs, CNBC confirmed. Anthropic is financing part of its purchase through a $63 billion vehicle backed by Blackstone and three banks, a plan that covers about half of its 2027 compute needs.
