OpenAI Moves Into Google Cloud, Ending the Era of Cloud Monogamy

There was a time when a company’s cloud was a statement of loyalty. You picked one, you put your models in it, and you did not discuss the others at dinner.

That era ended on October 21, when Google Cloud announced it is bringing OpenAI’s models to its Model Garden. Techzine described OpenAI’s models as “finally fully supported” on Google Cloud, a headline with the weary relief of a long-distance relationship finally getting a shared address.

The facts, per Techzine and RuntimeWire: Model Garden is Google’s catalog of more than 200 proprietary, third-party, and open-source models. The rollout is a private preview for select enterprise customers, with general availability planned “in the coming weeks” and no date given. Three GPT-6 variants are on the menu.

GPT-6 Astra is the heavyweight, pitched at demanding work such as coding, computer use, and scientific research. GPT-6.1 Sol approaches Astra’s capability at a lower cost. GPT-6 Luna is the fast, cheap option for everyday work at scale. It is a good-better-best ladder, which is the oldest pricing structure in commerce and has never once failed to work.

Developers can already connect Codex CLI and Desktop workflows for inference, while support for Gemini for Business and Google Antigravity is coming later. So Google’s own coding environment will eventually let you call a rival’s model. Somewhere, a product manager is explaining this to a slide deck.

The part CFOs will care about is the billing. Sales go through Google Cloud Marketplace, usage lands on a single invoice, and it counts toward existing cloud commitments and Flexible Savings Plans. In other words: the money you promised Google can now be spent on OpenAI, through Google. It is a closed loop of spending that makes a procurement officer weep with joy. You have found a way to buy a competitor’s product without leaving your current vendor’s warm embrace.

Google’s pitch is that “modern enterprise AI requires that each workload be matched to the right model.” This is the sentence the whole industry has quietly adopted, because the alternative sentence, “no one model wins, please keep buying all of them,” is harder to put on a banner.

According to Google Cloud’s own social post as reported by RuntimeWire, it is now “the only hyperscale cloud provider offering access to all thre[e]” — the excerpt cuts off there, a fitting punctuation for a sentence about sharing. Per Techzine, the three leading closed-source families on Google Cloud are now OpenAI’s, Anthropic’s Claude, and Google’s own Gemini. The same piece notes Google has kept Gemini exclusive to its own cloud. Everybody is welcome at the party. The host just doesn’t leave.

History, briefly, because the lineage matters. OpenAI’s models were initially exclusive to Microsoft Azure, owing to Microsoft’s earlier investment. AWS added them in summer 2025, per Techzine. Before this week, Google’s deal with OpenAI covered computing capacity only. The models were not offered to Google Cloud customers. Now they are.

So what became of the exclusive relationship? Scalevise notes that Flexible Savings Plans “can apply across providers,” and that the rollout extends to Google Antigravity and the Gemini Agent. The exclusivity that once defined the Azure relationship is now, shall we say, historical.

Now the fine print, because there is always fine print. Google says prompts, context, and outputs stay within the customer’s data perimeter and are not used to train models. The models run on Google Cloud infrastructure, with queries passing through isolated gateways. But Zero Data Retention and Confidential Compute are described as expected to follow, with Techzine noting they need additional work. In other words, the lock on the door is coming. Please enjoy the door.

And availability is, once more: private preview. If you were hoping to log in Monday and select Astra from a dropdown, you will first need to be selected yourself, like a kid waiting to get picked for dodgeball, except the dodgeball is a three-year enterprise commitment.

What does it all mean? The cloud wars have quietly stopped being a war and become a trade show. Each hyperscaler needs every frontier model in the catalog, and each model maker needs every hyperscaler’s customers. Nobody is exclusive anymore. Everyone is strategic.

The customer, meanwhile, benefits in the way customers usually do, which is that there are now more menus. Pick a cloud, pick a model, apply your existing commitment, and receive one invoice. The invoice will be large. It will, at least, be consolidated.

One last observation, offered with affection. The product is intelligence; the business, increasingly, is where the intelligence is available to be rented. It is real estate with a context window.

Sources: