Google licenses custom chips to Oracle in shift away from exclusive cloud strategy

Google is abandoning its traditional strategy of keeping AI chips exclusively within its cloud ecosystem, instead licensing custom processors to rivals and competitors through external partners. This shift signals a fundamental change in how the search giant competes in AI infrastructure, trading vertical integration for market share in a sector where Nvidia has long dominated.

• Oracle will offer Google’s tensor processing units to customers, marking the first time TPUs are available outside Google’s own infrastructure.
• TD Cowen projects $9.4 billion in TPU sales for the second half of this year and $62.5 billion for 2027, compared with current cloud revenue of $25 billion in Q2.
• Google’s largest TPU customer is Anthropic, which agreed last October to use up to 1 million TPUs in a deal worth tens of billions, despite competing directly with Google in AI models.

• $9.4B TD Cowen’s estimated TPU sales for second half of this year
• $62.5B Projected TPU sales for 2027 according to TD Cowen model
• 44% Cost advantage of Google’s Ironwood TPU versus Nvidia GB200 server
• $63B Broadcom-backed financing vehicle for Anthropic’s TPU and networking purchases

According to reporting by CNBC, Google announced on Wednesday (October 28) that Oracle will distribute its custom tensor processing units, a shift away from the company’s historical approach of selling TPUs exclusively through its own cloud infrastructure. The move comes one day before Alphabet’s third-quarter earnings report and represents a strategic reversal under the leadership of Thomas Kurian, Google Cloud’s chief executive, who previously spent more than two decades at Oracle.

Oracle becomes the first external distributor of Google’s custom chips

By licensing TPUs to Oracle, Google is taking its chips outside its own data centers for only the second time. Google opened TPU access to outside data centers in the second quarter before this week’s announcement, signaling a deliberate pivot in strategy. The shift reflects Kurian’s belief that Google needs to compete for market share rather than rely solely on its integrated stack of chips, models and cloud services.

Chris Ballard, managing director of Check Capital Management, said that Google’s move positions it to capture Nvidia’s market dominance in AI chips. “They’re a very strong competitor, and it’s very possible that they could eventually take market share from Nvidia,” Ballard said. Uday Cheruvu, portfolio manager at Harding Loevner, which holds Alphabet shares, attributed the strategic change to a shift in company thinking. “Google was great at doing its own thing, but it was never able to serve customers. That mindset has changed,” Cheruvu said.

Anthropic emerges as Google’s largest TPU customer despite direct AI competition

Google’s biggest consumer of TPUs is Anthropic, an AI model developer that competes directly with Google’s Gemini. Last October, Anthropic agreed to use up to 1 million TPUs in a deal valued at tens of billions of dollars. Industry analyst Mizuho characterized the relationship as Google “arming its closest model competitor,” while noting the arrangement brings nearly a gigawatt of capacity online this year.

Broadcom has structured financing to support Anthropic’s chip purchases. On this week, the semiconductor company detailed a $63 billion vehicle backed by Blackstone and three major banks that will buy TPUs and networking equipment from Broadcom and lease them to Anthropic for five years. Analysts from William Blair estimate that the financing plan covers approximately half of Anthropic’s 2027 needs, signaling that additional funding arrangements will likely follow. OpenAI also plans to buy TPUs from Google, expanding the list of Google’s AI competitors purchasing its custom silicon.

Google faces margin pressure and capacity constraints as it scales TPU distribution

Google Cloud revenue is expected to jump roughly 90% from a year earlier, according to analyst forecasts, after growing 82% in the second quarter.

The shift toward external distribution introduces financial trade-offs for Alphabet. TD Cowen estimates that TPUs carry an operating margin of approximately 20%, substantially lower than Google Cloud’s overall margin of about 35% in the third quarter. Cheruvu noted that Google’s cloud operating margin is unlikely to widen far beyond the mid-30s range, suggesting limited upside from margin expansion as TPU sales grow.

At a Goldman Sachs conference in September, Kurian acknowledged the tension between distributing TPUs externally and retaining capacity for Google’s own operations. He said Google’s AI servers pay for themselves in under two years, while servers using Google’s own silicon reach payback in roughly half that time. “It’s because we can optimize that whole stack,” Kurian said. According to SemiAnalysis, Google’s latest Ironwood TPU costs about 44% less to own and operate than an Nvidia GB200 server, giving Google a significant cost advantage in its own infrastructure. To manage capacity constraints, CFO Anat Ashkenazi told investors on the July earnings call that Google plans to rely on more outside capacity in the third quarter while the company builds out its own infrastructure. Ashkenazi also said that only a small portion of TPU revenue from existing agreements will be recognized this year, with the “vast majority” arriving in 2027.

The BlockWest read. Google’s TPU strategy now hinges on whether external distribution can generate sufficient unit volume to offset lower margins and constrain its own infrastructure investments. Investors should focus on how quickly Google scales partner capacity relative to competing chip vendors, and whether the financing vehicle structured by Broadcom and Blackstone becomes a template for future customers seeking access to Google’s silicon without upfront capital commitments.

With Google’s third-quarter earnings arriving Thursday (October 29), investors will scrutinize management’s updated guidance on TPU sales growth, capacity allocation between internal and external customers, and the timing of when the “vast majority” of TPU revenue kicks in during 2027.

BlockWest is a news publication. Nothing here is investment advice. Read our disclaimer and editorial policy.