Google TPUs Head to Oracle and a $5B Blackstone Venture

Oracle and Google said on 27 October 2026 that Google’s TPUs will be available through Oracle Cloud Infrastructure, “marking the first time Google TPUs will be offered on a hyperscale cloud platform outside Google.” Oracle’s release says: “Google TPUs on OCI will be available in calendar year 2027, ramping over a multi-year period to meet customer needs.”

A day later, on Alphabet’s third-quarter earnings call, Google Cloud chief executive Thomas Kurian described the change in plain terms: “So in the past, you could use TPUs in our cloud.” He said Google is now “making TPU systems available in their data centers” for customers who want them, and “also broadening availability to other providers.”

This piece traces, from Google’s and its partners’ own statements, how the chips moved from Google’s data centers to four ways for customers to get them. Call quotations come from a speech-to-text transcript of Alphabet’s webcast.

Business Pill · Distribution Advantage

An evergreen Business Engineer explainer on distribution advantage. It explains the general idea only and does not discuss TPUs, Oracle or Crux AI.

The key insight: By our arithmetic, about eight years separate Google Cloud’s 2018 TPU beta, when customers could only rent the chips, from Alphabet’s first disclosed agreements to supply TPU hardware for customers’ own sites, in the quarter ended 31 March 2026.

2016 to 2018: An In-House Chip, Then a Rental

Google first described the Tensor Processing Unit in a Google Cloud blog post dated 19 May 2016. It said: “We’ve been running TPUs inside our data centers for more than a year”, and that TPUs already powered applications including RankBrain and Street View.

On 12 February 2018, Google said: “Starting today, Cloud TPUs are available in beta on Google Cloud”. Customers could get “interactive, exclusive access to a network-attached Cloud TPU via a Google Compute Engine VM that you control and can customize.” The chips stayed in Google’s facilities; customers rented time on them.

Milestones in how customers get Google TPUs, 2016 to 2027, from Google's own posts, Alphabet's 10-Qs, the Crux
Milestones in how customers get Google TPUs, 2016 to 2027, from Google’s own posts, Alphabet’s 10-Qs, the Crux AI and Oracle releases and the Q3 2026 earnings call. Events are evenly spaced, not to scale. The count of four channels is our arithmetic.

2026: Hardware for Customers’ Own Sites

Alphabet’s Form 10-Q for the quarter ended 31 March 2026 disclosed: “Google Cloud has entered into a limited number of agreements to supply multiple gigawatts of TPU hardware to customers who require or provide on-premises infrastructure for specialized, high-scale workloads.”

The 10-Q for the quarter ended 30 June added: “In addition, in the second quarter of 2026, we began recognizing revenue from the sale of TPU systems.” It describes those sales as “the sale of TPU systems comprising hardware, software, installation, support, and extended warranty services.” We covered the first agreements in April and the first on-site revenue in July.

By our arithmetic, about eight years separate the Cloud TPU beta in February 2018 from the first disclosed agreements to supply TPU hardware for customers’ own sites, in the quarter ended 31 March 2026.

September: A Blackstone-Backed Venture

Crux AI launched on 9 September 2026 as “a U.S.-based AI infrastructure company built to deliver dedicated compute at industrial scale”, according to its release. It said it “has secured an initial $5 billion equity commitment from funds managed by Blackstone”, and that “for its launch platform, Google will provide TPUs, software and services.”

The release says Crux AI “plans to bring its first 500 MW of TPU capacity online in 2027, the initial phase of its multi-gigawatt scaling roadmap.” Its chief executive, Benjamin Treynor Sloss, “spent more than two decades at Google”, the release says.

On the call, Kurian said Google had “launched, along with Blackstone, an agreement for Crux AI, a joint venture to offer customers TPU systems.”

October: Oracle Becomes the First Outside Hyperscaler

Google’s 27 October blog post says the deal makes “OCI the first hyperscaler outside Google to offer TPUs”, and that “the global demand for AI compute exceeds any single cloud provider’s footprint.” Oracle’s release says “OCI will provide the power, cooling, and datacenter networking to deploy Google TPUs at scale within its infrastructure.”

Kurian said on the call: “Given continued strong and accelerating demand, we’ve recently entered into partnerships to deliver TPU systems with both Crux AI and Oracle, with more in the works.”

Why Google Says It Is Doing This

Kurian gave two reasons. Putting TPU systems “in third-party facilities and in customer data centers, it allows us, one, to broaden our addressable markets”, because some industries “have data in their premise that they cannot move to a cloud”. And: “Additionally, it also offsets the need for us to have to pre-build facilities to host these, uh, these TPU systems.”

Finance chief Anat Ashkenazi said: “Uh, to date, we’ve had some system sales recorded as revenue, but expect more of it to come in through, uh, twenty twenty-seven.” Alphabet’s inventory build and the trade-off between selling and using the chips are covered separately.

Four Routes to a TPU by 2027

By our count from these statements, there are four routes to Google’s TPUs once the 2027 capacity arrives, against one in 2018: renting them on Google Cloud; buying TPU systems for customers’ own data centers; and, from 2027 by the companies’ plans, capacity from Crux AI and Oracle Cloud Infrastructure. The count is our arithmetic. Google has not given revenue by channel.

Our earlier pieces explain what a TPU system contains and which customers Google named on the call.

The Structural Read

As we read the releases and the call, Google is moving TPUs through partners that bring their own buildings, power and capital: Oracle’s release says OCI will provide the power, cooling and datacenter networking, and Crux AI’s release cites an initial $5 billion equity commitment from Blackstone-managed funds.

Kurian said on the call that placing systems in third-party facilities and customer data centers “also offsets the need for us to have to pre-build facilities to host these, uh, these TPU systems.”

Google’s 27 October post put the reason in its own words: “the global demand for AI compute exceeds any single cloud provider’s footprint.”

Thomas Kurian, Google Cloud CEO, Q3 2026 earnings call, 28 October 2026

“So in the past, you could use TPUs in our cloud”

Three Implications

FOR CUSTOMERS Oracle’s release says Google TPUs on OCI will be available in calendar year 2027, and Crux AI’s release says it plans its first 500 MW of TPU capacity in 2027.

FOR GOOGLE’S FACILITIES Kurian said partner and customer sites offset the need for Google to pre-build facilities to host the systems.

FOR REVENUE TIMING Ashkenazi said some system sales have been recorded as revenue and that Google expects more of it to come in through 2027.

The Business Engineer Lens

This story maps onto the Business Engineer framework The State of AI Data Centers.

The framework argues: “The structural bottleneck isn’t computing, it’s electrons.”

The partners in this story supply the physical side: Oracle’s release says OCI will provide the power, cooling and datacenter networking for the TPUs. The framework does not discuss TPUs or these partnerships.

What Is Not Established

Neither Google nor Oracle disclosed pricing, volumes or revenue sharing for TPUs on OCI. Crux AI’s release gives no customer names and no figure for Google’s stake, if any, in the venture. Alphabet does not report TPU revenue by channel.

The Oracle and Crux AI capacity is scheduled for 2027 in the companies’ own releases; neither has said how many chips that involves.

This Week In The AI Supercycle

Execution Fault Lines, Concentrated Demand, and Geopolitical Firewalls

This week’s issue: the AI build is not short of demand for its parts. It is short of cheap money, and its end revenue rests on a narrow base of large, often venture-funded buyers.

Read the full issue →

Business Engineer Framework

Google’s TPU AI Compute Arbitrage

A Business Engineer analysis of how Google uses its own chips to change the economics of AI compute.

Read the Map of AI →

The Bottom Line

Google first described its TPU in 2016 as a chip running inside its own data centers, and from 2018 customers could rent it on Google Cloud. In 2026 Alphabet began supplying TPU systems for customers’ own sites, Crux AI launched with an initial $5 billion equity commitment from Blackstone-managed funds and Google TPUs, and Oracle said it will offer TPUs on its cloud from 2027.

Business Engineer Premium
Premium: Alphabet & Microsoft on the Map of AI

Cash conversion, the lengthening backlog, the arms-dealer turn at Layer 3: the full Business Engineer read of both prints.

Read the premium analysis →

95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.

A note on sourcing. The 2016 and 2018 milestones come from Google Cloud blog posts; the 2026 supply agreements and first revenue come from Alphabet’s Form 10-Qs for the first and second quarters; the Crux AI details come from its 9 September release; the Oracle details come from Oracle’s and Google’s 27 October announcements; call quotations come from a speech-to-text transcript of the 28 October webcast. The year count and the four-route count are our arithmetic.

Sources: Oracle release (27 Oct 2026) · Google Cloud blog, TPUs on OCI (27 Oct 2026) · Crux AI release (9 Sep 2026) · Alphabet Form 10-Qs, Q1 and Q2 2026 (SEC EDGAR) · Alphabet Q3 2026 earnings call webcast (28 Oct 2026) · Google Cloud blog, TPU (19 May 2016) · Google Cloud blog, Cloud TPU beta (12 Feb 2018)

Scroll to Top
FourWeekMBA