TPU Sales Projected at $9.4 Billion in 2026 as Google Cloud Reaches Outside Its Own Data Centers

Alphabet announced Tuesday that Oracle will offer Google's tensor processing units, or TPUs, to customers. This marks the first time the custom chips, developed since 2013, will be accessible outside Google's own data centers. The announcement came one day before the company's third-quarter earnings report.

Analysts anticipate Google Cloud revenue will increase by approximately 90% from the same period last year. This follows an 82% increase reported in the second quarter. Thomas Kurian, who joined Google in late 2018 after a two-decade tenure at Oracle, is leading this shift toward partner-driven growth. The company had previously opened TPUs to external data centers in the second quarter, ahead of the Oracle deal.

Market reactions focus on Google's ability to challenge Nvidia's dominance. Chris Ballard, a managing director at Check Capital Management, stated that Google is a strong competitor that could eventually take market share from Nvidia. Uday Cheruvu, a portfolio manager at Harding Loevner, noted that Google's mindset has changed from doing its own thing to serving customers. Estimates for the TPU business vary, but projections indicate steep growth. TD Cowen models $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 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. She described the "vast majority" of that revenue as arriving in 2027. Google still trails its rivals in total scale. Second-quarter cloud revenue was $25 billion, compared with $42 billion at Amazon Web Services and $29 billion at Microsoft Azure. However, Google is growing the fastest. Selling chips externally creates margin pressure. TD Cowen estimates a 20% operating margin on TPU sales, lower than the approximately 35% overall cloud margin from the previous quarter.

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

Anthropic, a major AI competitor to Google's Gemini models, is Google's largest TPU customer. 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. Anthropic is using external financing to pay for the chips. This week, Broadcom detailed a $63 billion vehicle backed by Blackstone and three large banks. The fund will buy TPUs and networking gear from Broadcom and lease them to Anthropic for five years. William Blair analysts say this plan covers about half of Anthropic's 2027 needs.

In its IPO filing on Monday, Anthropic said it uses silicon from Google, Amazon, Nvidia, and Advanced Micro Devices. The company stated it selects chips based on the best price-performance and efficiency characteristics for each workload. It noted that partnerships with Google and Broadcom provide access to compute capacity powered by Google TPUs across multiple generations. CNBC confirmed that OpenAI also plans to buy TPUs from Google, following a report from The Information on Tuesday.