OpenAI's GPT models on Google Cloud: one invoice, and usage counts toward existing cloud commitments
Google Cloud launched OpenAI's GPT-6 models in its model catalog on October 21, 2026, in private preview for select enterprise customers. That means Google's own Gemini is now sold on the same platform as its biggest competitor — and Google, by its own account, is the only hyperscale cloud provider offering all three leading closed language models on a single platform.
What launched
The announcement came in a blog post by Michael Gerstenhaber, Vice President of Product Management at Google Cloud, dated October 21, 2026 (Google Cloud). Three OpenAI models are available in Google's model garden, with descriptions worth quoting as Google itself presents them:
- GPT-6 Astra: "Designed for demanding work in coding, computer use, professional tasks and scientific research," according to Google.
- GPT-6.1 Sol: "Near-Astra intelligence at lower cost" for agentic coding, computer use and professional work — Google's description positions it as a cheaper, near-peer alternative to Astra.
- GPT-6 Luna: "A fast, cost-effective model for everyday work at scale."
Access is for now limited: private preview for select enterprise customers. OpenAI plans general availability "in the coming weeks," the company said in a blog post — known via The Next Web's coverage of OpenAI's post, not directly from OpenAI.
How it works for buyers
The practical move lies in the billing. Because the OpenAI models are transacted directly through Google Cloud Marketplace, all usage lands on a single Google Cloud invoice, draws on existing cloud commitments, and counts toward Flexible Savings Plans across model providers, according to Google (Google Cloud).
For enterprise buyers, this changes a concrete mechanism: large companies typically hold multi-year commitment agreements with a cloud platform, and model usage that previously had to be purchased separately from OpenAI can now in practice be covered by the commitment they already have with Google. That reduces the friction of trying and mixing models across providers — and it is this mechanism that makes distribution deals economically significant for the cloud platforms.
Data handling and governance
Google claims that customers' messages, context and outputs remain within the customer's agreed data perimeter and are never used to train foundation models, and that the OpenAI models run on dedicated Google Cloud infrastructure. Two additional offerings — Zero Data Retention (ZDR) and Confidential Compute — are, however, only "coming soon," meaning not available at launch. That is a detail that may matter for regulated customers in finance, healthcare and the public sector, who often require precisely these mechanisms before moving sensitive workloads.
These points are the company's own claims, not independently verified.
The context: OpenAI on all three major clouds
With this, OpenAI's models are available from all three largest US cloud providers. Amazon Web Services began offering them in April, after Microsoft's exclusive license expired, and OpenAI added its agents to Amazon Bedrock last month — per The Next Web, which cites OpenAI.
At the same time, OpenAI emphasizes that Microsoft remains the company's primary cloud partner. That is an important nuance: the Google deal expands distribution, but does not change the core of OpenAI and Microsoft's relationship, according to OpenAI via TNW.
OpenAI's president and co-founder Greg Brockman called the partnership "an important step toward making our most capable models and Codex available where customers are already building," in the announcement, as reported by TNW. Google Cloud chief Thomas Kurian framed it as a matter of model choice for customers.
The competitive effect: Gemini on the same shelf
The obvious tension is that Google is now distributing the models of its own most capable competitor. Analyst Mike Leone of Moor Insights & Strategy pointed to this to Axios: "Gemini now has to win those customers with OpenAI sitting right next to it," he said (TNW).
The economics of such deals are not public — neither pricing, revenue sharing nor other terms are known from any of the sources. But the industry norm is that cloud providers take a share of the model sales on their platforms, reported Krystal Hu for Reuters. One earlier example of how much distribution can matter: Anthropic, according to its IPO filing, accounted for 47 percent of its revenue in 2025 through Amazon's and Google's cloud platforms, per Reuters.
Google's own formulation of being "the only hyperscale cloud provider offering all three leading closed source models, alongside more than 200" other models, is a company claim that should be read with precision: the claim depends on how "the three leading closed models" is defined, and none of the sources document whether AWS or Azure offer comparable combinations.
An internal tension at Google
The Decoder frames the move as a recurring tension within Google: compute capacity spent on cloud customers strengthens the biggest rivals of the company's own Gemini team. The outlet also claims, with the caveat "reportedly," that this resource conflict is said to have pushed researcher Noam Shazeer back to OpenAI — a detail that is single-sourced and unconfirmed by other outlets, and should therefore be treated with great caution.
What remains to be seen
Several questions are open. The financial terms between Google and OpenAI are not known. The models are only in private preview, so performance, reliability and actual adoption on Google's infrastructure remain unresolved. ZDR and Confidential Compute are still missing. And Google's "only hyperscaler" claim is independently unverified.
The timing is nonetheless not coincidental: earnings reports from Alphabet, Microsoft and Amazon come next week, and enterprise customers' model choices — and how the cloud platforms monetize them — will likely be central. General availability is supposed to come "in the coming weeks," according to OpenAI via TNW.