Google Turns to Oracle and Other Partners to Widen Reach of Its AI Chips

Google has long argued that customers benefit from buying chips, models and cloud services from a single provider, a full-stack approach that rivals Amazon and Microsoft have tried to copy. Now the company is taking its hardware beyond its own data centers in an effort to build a larger business in artificial intelligence chips, a market Nvidia has dominated.

On Tuesday, a day before its third-quarter earnings report, Alphabet said Oracle will offer Google's tensor processing units, or TPUs, the custom chips Google has been developing since 2013. It is the first time customers will be able to reach TPUs outside Google's infrastructure. The move also shows that cloud chief Thomas Kurian, who spent more than 20 years at Oracle before joining Google in late 2018, is seeking growth through partners.

Kurian is working from a position of strength. Analysts expect Google Cloud revenue to rise about 90% from a year earlier, after an 82% increase in the second quarter, with TPUs helping drive that momentum. Google had already opened TPUs to outside data centers in the second quarter, ahead of the Oracle announcement.

"They're a very strong competitor, and it's very possible that they could eventually take market share from Nvidia," said Chris Ballard, managing director of Check Capital Management. Uday Cheruvu, a portfolio manager at Harding Loevner, which owns Alphabet shares, said Google "was great at doing its own thing, but it was never able to serve customers," adding that "that mindset has changed."

Estimates for the TPU business vary widely but all point to steep growth. TD Cowen models $9.4 billion in TPU sales for the second half of this year and $62.5 billion for 2027. Wells Fargo expects about $9 billion in third-quarter sales and $99 billion next year. Chief financial officer 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" arriving in 2027.

Google still trails its biggest rivals in scale. Its second-quarter cloud revenue was $25 billion, compared with $42 billion at Amazon Web Services and $29 billion at Microsoft Azure, though Google is growing the fastest. Selling chips externally brings trade-offs, including pressure on margins. TD Cowen estimates a 20% operating margin on TPU sales, against an overall cloud margin of about 35% last quarter. Cheruvu said he does not expect Google's cloud operating margin to widen much beyond the mid-30s. Google declined to comment.

Kurian must also weigh customer growth against Google's own needs. At a Goldman Sachs conference in September, he said Google's AI servers pay for themselves in under two years, and those running its own silicon reach payback in about half that time. "It's because we can optimize that whole stack," he said. SemiAnalysis estimates Google's latest Ironwood TPU costs about 44% less to own and operate than an Nvidia GB200 server.

Google's largest TPU customer is also one of its main AI competitors. Anthropic, whose Claude models compete with Gemini, agreed last October to use up to 1 million TPUs in a deal worth tens of billions of dollars. Mizuho says the agreement brings nearly a gigawatt of capacity online this year and describes it as Google "arming its closest model competitor." CNBC has also confirmed that OpenAI plans to buy TPUs from Google, following a report from The Information on Tuesday.

Anthropic is receiving outside help to pay for the chips. This week, Broadcom, which works with Google on TPUs, detailed a $63 billion vehicle backed by Blackstone and three large banks. It will buy TPUs and networking gear from Broadcom and lease them to Anthropic for five years. William Blair analysts say the plan covers about half of Anthropic's 2027 needs, suggesting more financing is coming.

In its IPO filing on Monday, Anthropic said it uses silicon from Google, Amazon, Nvidia and Advanced Micro Devices, letting it "select chips based on the best price-performance and efficiency characteristics for each workload." It said its partnerships with Google and Broadcom "provide us with access to compute capacity powered by Google TPUs across multiple generations of the TPU platform."