Oracle Will Sell Google's AI Chips Directly to Customers

Alphabet announced on Tuesday that Oracle will sell Google's tensor processing units, or TPUs, to outside customers. This move breaks with the company's long-standing strategy of selling its chips, models, and cloud services as a single integrated package. By offering the hardware through a third-party vendor, Google is extending its reach beyond its own data centers to capture a larger share of the artificial intelligence market.

The partnership represents a shift in how Thomas Kurian, Google's cloud chief, approaches growth. Kurian, who joined the company in late 2018 after spending two decades at Oracle, is leveraging his background to expand the TPU business. Analysts project that Google Cloud revenue will increase by approximately 90% in the coming quarter, following an 82% rise in the previous period. This rapid expansion is largely attributed to the momentum generated by TPU sales.

Chris Ballard, managing director of Check Capital Management, noted that Google is a strong competitor with the potential to take market share from Nvidia. "They're a very strong competitor, and it's very possible that they could eventually take market share from Nvidia," Ballard said. His firm held about $236 million in Alphabet stock as of June 30. Uday Cheruvu, a portfolio manager at Harding Loevner, added that Google's previous focus on internal development has evolved. "Google was great at doing its own thing, but it was never able to serve customers," Cheruvu said. "That mindset has changed."

Google began opening TPUs to outside data centers in the second quarter of this year. Financial forecasts for the TPU business vary, but they all indicate substantial growth. TD Cowen estimates TPU sales at $9.4 billion for the second half of 2026 and $62.5 billion for 2027. Wells Fargo expects third-quarter sales of about $9 billion and total sales of $99 billion next year.

Anat Ashkenazi, Google's chief financial officer, stated in July that only a small portion of TPU revenue from existing agreements will be recognized this year. She indicated that the vast majority of that revenue will arrive in 2027. The company also plans to utilize more external capacity in the third quarter while it continues to build its own infrastructure. In the second quarter, Google generated $25 billion in cloud revenue, trailing Amazon Web Services' $42 billion and Microsoft Azure's $29 billion.

Despite the lower absolute revenue compared to competitors, Google is growing at the fastest rate. "It's almost like the bigger the pie is getting, the faster they're accelerating into it," Ballard said. Kurian must balance the goal of increasing market penetration with the company's internal technology needs. At a Goldman Sachs conference in September, he explained that Google's AI servers pay for themselves in under two years, while servers running its own silicon reach payback in roughly half that time.

"It's because we can optimize that whole stack," Kurian said. Industry research from SemiAnalysis estimates that Google's latest Ironwood TPU costs about 44% less to own and operate than an Nvidia GB200 server. However, selling TPUs externally affects profit margins. TD Cowen estimates a 20% operating margin on TPU sales, compared with Google's overall cloud margin of about 35% last quarter.

Google's largest TPU customer is Anthropic, a major competitor in the AI model space. Anthropic agreed last October to use up to 1 million TPUs in a deal worth tens of billions of dollars. Mizuho analysts say this agreement brings nearly a gigawatt of capacity online this year. OpenAI is also planning to purchase TPUs from Google. Anthropic is securing external financing to support this silicon purchase.

This week, Broadcom detailed a $63 billion financing vehicle backed by Blackstone and three major banks. The arrangement allows the entity to buy TPUs and networking gear from Broadcom and lease them to Anthropic for five years. William Blair analysts note that this financing covers about half of Anthropic's projected 2027 needs. In its IPO filing on Monday, Anthropic stated that it uses silicon from Google, Amazon, Nvidia, and Advanced Micro Devices to select chips based on price, performance, and efficiency.