OpenAI Models Now Count Toward Google Cloud Budgets
Google Cloud is rolling out OpenAI models in its Model Garden as a private preview. Existing spending commitments should count toward usage.
Since 21 October, Google Cloud has been rolling out OpenAI models in its Model Garden, initially as a private preview for selected enterprise customers. Billing runs through the Google Cloud Marketplace on a single Google Cloud invoice. Existing spending commitments with Google should count toward the purchase.
Key takeaways
- Google Cloud is rolling out three OpenAI models in the Model Garden. Access is available as a private preview only and is limited to selected enterprise customers.
- Usage runs through Google invoicing and spending commitments. Companies should be able to apply their existing spending commitments with Google Cloud toward the purchase of the models.
- Some data protection options have only been announced so far. Zero Data Retention and Confidential Compute are still outstanding, and general availability is planned within the coming weeks.
Related:Neoclouds: Gartner expects a wave of acquisitions / GPU sharing without a subscription: NVIDIA makes Run:ai open source
Three OpenAI models at Google
The Model Garden, Google Cloud’s model catalog, will now also carry models from OpenAI. Google Cloud is rolling out three of them at launch: GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna. They are being switched on as a private preview for selected enterprise customers. A private preview is a pre-release version for a small circle of customers selected by the provider.
OpenAI justifies the move with more freedom of choice for companies. They are set to gain more options to find and use the most powerful models. For Google Cloud, the agreement means that its own catalog now includes an additional model provider. Customers reach this provider through the contracts and tools they already use with Google.
What is the Model Garden? The Model Garden is the model catalog of Google Cloud, through which companies find AI models and integrate them into their own applications. Since 21 October, Google Cloud has also been rolling out models from OpenAI there, initially for selected enterprise customers. Billing for these models runs through the Google Cloud Marketplace on a single Google Cloud invoice.
Billing on a single Google invoice
Google Cloud handles the purchase through its Marketplace. OpenAI usage appears on a single Google Cloud invoice, together with the rest of a company’s consumption on the platform.
A rule for budgets that have already been negotiated comes on top. Companies should be able to apply their existing spending commitments with Google Cloud toward the purchase of the OpenAI models. Such spending commitments are contractually agreed minimum expenditures with a cloud provider over a fixed term. Money that a company has already committed at Google can therefore also flow into OpenAI usage.
Data boundary and dedicated infrastructure
Google Cloud guarantees that customer inputs, context and outputs stay within the defined data boundary. The data is also not used to train foundation models. According to Google Cloud, the OpenAI models run on dedicated Google Cloud infrastructure, meaning resources reserved specifically for this offering.
Two components have so far only been announced. Options without data retention, referred to by Google as Zero Data Retention, are to follow soon. Confidential Compute is also still outstanding. It refers to processing in which data remains protected even during computation. Google Cloud names no date for either component.
Preview by invitation only
Access to the preview is granted by invitation only. Participants must confirm that they use the models only for approved development and internal use cases. The preview is therefore not intended for production use with customer data.
The partners plan to make the models generally available in the coming weeks. From then on, access is to be available without an invitation. Until then, the circle of users remains small and geared toward testing.
Microsoft remains the primary cloud partner
According to OpenAI, the move changes nothing about the existing partnership. Microsoft remains the company’s primary cloud partner. The models are therefore available at Google Cloud in addition and do not replace the existing offering.
Another detail concerns developers. As of the announcement, local workflows with the Codex CLI and the desktop application can be connected to Google’s Gemini platform for inference. The Codex CLI is the command line version of Codex. Inference refers to running a trained model on new inputs. The compute work for these workflows can therefore run through Google.
Frequently asked questions
How do companies get access to the preview?
Interested parties can register through a form. Registration leads to being considered for early access and is not a commitment. The providers themselves continue to grant access by invitation only.
Which interface do companies use to test the models?
Selected customers evaluate the OpenAI models through the Responses API. Approved uses are supported development and internal testing.
Does the usage also count toward Google Cloud savings programs?
Yes, according to Google Cloud, the OpenAI usage counts toward the Flexible Savings Plans. These savings plans apply across providers.
Editor’s Picks
- PHP2Shell cripples authorities: Who needs to patch now
- Anthropic and Nscale negotiating a second billion-euro contract
- Hijacked backup app takes down ConnectWise Cloud Backup
Image source: AI-generated (October 2026)
Translated from the German original using artificial intelligence. The German version is authoritative.

