OpenAI to Sell GPT-6 Astra and Codex via Google Cloud

OpenAI will sell its cutting-edge models and the Codex programming agent through Google Cloud, according to a joint announcement made on Wednesday (21) and reported by Reuters and Axios. Three models will be available in Google's Model Garden: GPT-Astra, GPT-6.1 Sol, and GPT-6 Luna. According to a Google post authored by Michael Gerstenhaber, they will be available "in private preview for select enterprise customers," with general availability expected in the coming weeks.

The agreement puts the main rival of Gemini on Google’s own platform. Purchases will go through the Google Cloud Marketplace, appear on the same cloud invoice, and count against existing spending commitments, as stated in the announcement. For a CIO who signed a multi-year contract with Google, the billing changes: using GPT will no longer require a second vendor, a second contract, and a second compliance track.

The End of Exclusivity in Three Stages

The sequence began in April when the exclusive distribution arrangement between OpenAI and Microsoft expired. On April 28, OpenAI’s models became available on Amazon Bedrock in limited preview. Prior to this, in February, OpenAI had already designated AWS as the exclusive third-party cloud distributor for Frontier, its agent management platform, in a deal where Amazon committed to invest $15 billion upfront and an additional $35 billion contingent upon goals, according to reports on the terms. With Google, OpenAI completes the cycle of the three major clouds within six months.

For OpenAI, the logic is about reach: large AI contracts are secured within already allocated cloud budgets. For Google, the logic is about retention: Google Cloud closed the second quarter of 2026 with revenues of approximately $24.8 billion, an 82% increase year-over-year, and a contracted portfolio of $514 billion, according to Alphabet's balance sheet from July 22. Each customer who needed GPT and left the platform to obtain it represented a crack in that portfolio.

What Changes for AI Buyers

Reuters emphasizes a point that matters more to banks, insurers, and healthcare than the content catalog itself: clients in regulated sectors will be able to access OpenAI’s models through Google Cloud's access controls, data governance tools, and infrastructure. In practice, the security layer already audited by the risk team will also cover the competing model. The models will also be available on Antigravity, Google's development environment, and on Gemini Agent, according to Axios.

The short-term effect is on negotiation. With GPT, Gemini, and Claude available under the same contract and contributing to Google's Flexible Savings Plans, switching models turns from a procurement project into an engineering decision. This reduces the bargaining power of any individual lab and shifts margin to those controlling the invoice. Cloud providers typically retain a share of the sales made on their platforms, recalls Reuters, and neither company has disclosed the percentage of this agreement.

There are limitations. For now, access is limited to chosen clients, the terminology still fluctuates between "GPT-Astra" and "GPT-6 Astra" in reports, and neither company has disclosed prices for the version served by Google.

Where the Effect Appears Outside the U.S.

In Europe, the regulatory argument carries more weight. Banks that have already standardized on Google Cloud, like BNP Paribas, which signed a five-year partnership with Google for agent-based AI in September, gain the option to test GPT without opening a new vendor under EBA and DORA outsourcing rules. Each new critical provider requires risk mapping, an exit plan, and notification to the supervisor, and purchasing a model under the existing contract shortens that process.

In India, the impact is felt through integrators. TCS, Infosys, and Wipro operate large Google Cloud practices and sell AI projects via the platform. A catalog featuring the three major labs under the same cloud shifts the value of model selection to orchestration and data engineering, where these firms charge by the hour. In Brazil, where major banks and retailers combine contracts with more than one hyperscaler, the change is likely to emerge first in renegotiating consumption commitments: the same spending now buys more model options.

The less discussed point is what the agreement indicates about Microsoft. Since 2019, privileged access to OpenAI has been Azure’s main commercial argument in AI. With GPT available on AWS and Google in less than six months, Azure now competes on price, integration with Microsoft 365, and delivery capacity, like any other cloud.