A rival's models, sold from Google's shelf

The strange part is the pairing, and both companies know it. OpenAI and Google compete directly in models, coding tools and search. Google is now reselling the competitor's flagship line inside its own enterprise platform, and presenting that as the point: in its announcement, Google Cloud argued that enterprise AI means matching each workload to a suitable model, and claimed the addition makes it the only hyperscaler offering access to all three leading closed-source model families, alongside more than 200 models in its garden. That superlative is Google's own claim, but the strategy underneath it is plain enough. A cloud that hosts every serious model keeps the customer, the data and the invoice, whichever model wins a given task.

For OpenAI, the gain is procurement. An enterprise that has standardised on Google Cloud can now buy OpenAI models through a contract, a billing relationship and a set of access controls it already has, instead of opening a separate vendor relationship. Reuters reported that regulated-industry customers are a specific target, because they can run the models under Google Cloud's governance and infrastructure controls. Since the April expiry of the Microsoft exclusivity, as Reuters reported it, OpenAI has added AWS and now Google Cloud to its distribution. The arrangements span all three hyperscalers. The availability does not: on Google Cloud, access begins with selected customers in a private preview.

Read the rollout terms before the headline

"Available on Google Cloud" needs its qualifiers restored. Google is introducing GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna in private preview for selected enterprise customers in Model Garden, with an interest form as the route in for everyone else. Customers with access can build through Google's APIs and connect their local Codex CLI and Desktop workflows directly to the Gemini platform for inference. The direction matters: the Codex clients stay on the customer's own machines and call Google-hosted inference. Google is not hosting Codex Desktop. Reuters put broader availability in the coming weeks. Google has announced no general-availability date, and no pricing was published in the announcement or the reporting, so there is currently no way to compare the cost of the same OpenAI model bought direct, through Azure, through AWS or through Google.

The governance split matters just as much. Google describes data-perimeter controls, a commitment that customer prompts, context and outputs are never used to train foundation models, authenticated access, VPC Service Controls and unified audit logging as part of the current offering. Two of the features an enterprise security review would ask about first are explicitly marked coming soon: Zero Data Retention options and Confidential Compute. Buyers should not treat those future features as protections the preview already has. OpenAI models inside Gemini for business and Google Antigravity sit in the same coming-soon column, with no date attached.

The channel is becoming the business

The most useful number in the story is not about OpenAI at all. Reuters reported that 47% of Anthropic's 2025 revenue came from Amazon and Google's cloud platforms, according to Anthropic's IPO filing as reviewed by Reuters. Read the figure for what it establishes, and no more: nearly half of one frontier lab's revenue already arrives via the two cloud platforms that back it and distribute its models. It does not establish how much of that revenue passed through cloud marketplaces, what fees the platforms retained, or that OpenAI's Google Cloud arrangement will carry the same economics.

What the figure does underline is where the leverage sits. Model quality still decides which name is on the purchase order. Increasingly, though, the cloud contract decides whose invoice it lands on, which commitments it burns down, and which governance stack wraps around it. That gives the hyperscalers leverage that does not depend on winning the model race outright, and it gives the labs a reason to be everywhere at once, including inside a direct rival's shop.

What to watch: a general-availability date, published pricing that can be compared across the three clouds, and whether the coming-soon governance features ship before broader availability does. Until then, this is a distribution agreement with a preview attached, and it should be budgeted as one.

Sources

  1. [1] Google Cloud, 'Google Cloud brings OpenAI models to customers', Google Cloud Blog, 21 October 2026 — the Model Garden launch and the model names as Google publishes them (GPT-6 Astra, GPT-6.1 Sol, GPT-6 Luna), the private preview for select enterprise customers, Marketplace transactions on a single invoice drawing down commitments and counting toward Flexible Savings Plans, the data-perimeter and no-training commitments, isolated gateways, authenticated identity with VPC Service Controls and unified audit logging, local Codex CLI and Desktop connections to the Gemini platform, and Zero Data Retention, Confidential Compute, Gemini for business and Antigravity listed as coming soonRead source
  2. [2] Reuters, 'OpenAI to sell its top models on Google Cloud in latest enterprise push', Krystal Hu, 21 October 2026 — availability in the coming weeks, purchases through Google Cloud Marketplace against existing cloud commitments, the regulated-industry governance angle, the April expiry of OpenAI's Microsoft exclusivity and the expansion to AWS, and Anthropic's 47% of 2025 revenue coming from Amazon and Google's cloud platforms according to its IPO filing as reviewed by Reuters. Read via a syndication mirror of the Reuters reportRead source
  3. [3] Unite.AI, 'Google Cloud Adds OpenAI's GPT-Astra and GPT-6 Models to Model Garden', 21 October 2026 — a detailed account of the Google announcement; note that it renders the first model's name as GPT-Astra, where Google's own post says GPT-6 Astra, and this article follows Google's namingRead source
  4. [4] PANews, 'OpenAI Coming to Google Cloud: Codex Coding Agent Listed on Google Cloud', 21 October 2026 — corroboration of the Model Garden, Marketplace and Codex termsRead source
  5. [5] AI Weekly (originally CNBC), 'OpenAI Puts Frontier Models on Google Cloud Model Garden and Antigravity Coding Tool', 21 October 2026 — corroboration that Antigravity and Gemini for business availability follow the initial rolloutRead source