OpenAI and Google Cloud said Wednesday, October 21, that OpenAI’s frontier models and its Codex coding agent are coming to Google Cloud’s Model Garden. That puts OpenAI on all three of the biggest US cloud platforms, after years of the two companies competing head-on.
Per Google’s announcement, GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna are rolling out now in private preview to select enterprise customers. OpenAI says general availability is planned “in the coming weeks,” and that its models will also show up in Google’s Antigravity coding tool and its Gemini Agent.
How we got here
For most of OpenAI’s commercial life, if you wanted its models through a big cloud provider, you went to Microsoft Azure. That was the deal: Microsoft invested billions and got an exclusive license to OpenAI’s technology.
That changed on April 27. In an amended agreement, Microsoft kept a license to OpenAI’s models and products through 2032, but it became non-exclusive. OpenAI can now serve its products “across any cloud provider.” Microsoft stopped paying OpenAI a revenue share, while OpenAI’s payments to Microsoft continue through 2030, subject to a cap.
Amazon moved within a day. AWS announced OpenAI models, Codex and OpenAI-powered agents in limited preview on its Bedrock platform, as CNBC noted. Google is the last of the three.
The two companies already had a supplier relationship. OpenAI said last year it was using Google’s cloud to help run ChatGPT and its API. This deal goes further: Google becomes a reseller.
What’s actually live
Today it’s limited. OpenAI’s access form describes an invite-only preview where select customers can evaluate the models through the Responses API “for supported development and internal testing use cases,” plus Codex through its app, CLI and supported IDEs.
Google adds that developers can point their local Codex command-line and desktop workflows at Google’s Gemini platform for inference. In other words, the coding agent runs on your laptop, and the model calls bill to Google.
The lineup, per Google:
- GPT-6 Astra: for demanding coding, computer use, professional tasks and scientific research.
- GPT-6.1 Sol: “near-Astra intelligence at lower cost.”
- GPT-6 Luna: the fast, cheap option for everyday work at scale.
Why enterprises care
The boring details are the point.
Customers buy through Google Cloud Marketplace. Usage lands on a single Google Cloud invoice, draws down existing Google Cloud spending commitments, and counts toward Google’s Flexible Savings Plans. Many big companies sign multi-year deals promising to spend a set amount with a cloud provider. If OpenAI usage counts toward that money, adopting it turns from “sign a new vendor contract and get it through procurement” into “flip a switch.”
Google is also pitching governance, which matters most in regulated industries such as banking and healthcare. It says prompts and outputs stay inside the customer’s data perimeter and aren’t used to train foundation models. The models run on dedicated Google Cloud infrastructure, behind isolated gateways, with Google’s identity, audit logging and network controls applied.
Two features are listed as “coming soon”: zero data retention, meaning nothing is stored after a request, and Confidential Compute, which keeps data encrypted even while it’s being processed. For some customers, those are the features that matter.
The frenemy math
These two companies compete directly on models. Google’s Gemini 4 Argon is a flagship rival to OpenAI’s GPT-6 line.
So why do it? For Google Cloud, the answer is that the cloud business wins either way. Google now gets to say it’s “the only hyperscale cloud provider offering access to all three leading closed source LLMs” (meaning Gemini, Anthropic’s Claude and OpenAI’s GPT), alongside more than 200 other models. Customers who want OpenAI no longer have a reason to leave.
The cloud business is on a run. CNBC reports Google’s cloud revenue growth accelerated to about 82% in the second quarter, versus 37% for Amazon’s and 42% for Microsoft’s.
For OpenAI, it’s distribution. CNBC notes the company is courting investors at a $1.4 trillion valuation and targeting an IPO, and enterprise revenue is central to that story. Per Ramp’s corporate card data cited by CNBC, most business AI spending already goes to OpenAI and Anthropic.
OpenAI’s post ends with a line clearly written with Redmond in mind: “Microsoft remains OpenAI’s primary cloud partner.” Under the April terms, OpenAI products still ship first on Azure unless Microsoft can’t or won’t support them.
Who it affects
- Google Cloud customers: easier access to GPT models without a new contract, once it’s generally available.
- Microsoft: loses a reason for customers to choose Azure. It keeps its OpenAI license and its stake, and still gets products first.
- Google’s Gemini models: now share a shelf with their most direct rival, inside Google’s own store.
The skeptical read
The catch: “coming to Google Cloud” and “available on Google Cloud” are different things. Right now it’s a private preview, for select customers, for development and internal testing. Production use waits for general availability.
Pricing is unannounced. Neither company said whether GPT-6 models will cost the same on Google as on OpenAI’s own API or on Azure. If Google charges a premium for the convenience of committed spend, the “flip a switch” argument weakens.
Two of the governance promises are still “coming soon.” Regulated buyers often need exactly those before they can say yes.
The takeaway
Our read: frontier models are turning into line items on whichever cloud bill a company already pays. That’s good for buyers and a slow squeeze on any model company that hoped exclusivity would be a moat. It also makes the model itself, not the cloud, the main thing labs compete on.
What to watch:
- When the preview opens to general availability, and what it costs compared with OpenAI’s own API.
- Whether zero data retention and Confidential Compute ship, and when.
- All three cloud giants were due to report quarterly results the week after the announcement. Listen for how much any of them credits third-party models.
Image: Lambtron / Wikimedia Commons, CC BY-SA 4.0
