OpenAI Brings GPT-6 Models and Codex to Google Cloud's Model Garden

OpenAI's frontier models are coming to Google Cloud. In coordinated announcements on Oct. 21, OpenAI and Google Cloud said OpenAI's newest models and its Codex agentic coding workflows will be offered in Google Cloud's Model Garden and sold through Google Cloud Marketplace, letting enterprises that already run workloads on Google's platform buy OpenAI models on their existing cloud bill. Google Cloud said GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna are rolling out first in private preview to select enterprise customers, while OpenAI said general availability is planned in the coming weeks.

Industry background

Model Garden is the catalog inside Google Cloud's AI platform where enterprises select, test and deploy foundation models. Google said the catalog now holds more than 200 first-party, third-party and open models, and claimed that adding OpenAI makes Google Cloud the only hyperscale provider offering access to all three leading closed-source large language models. Google had already hosted OpenAI's open-weight gpt-oss models; this agreement covers OpenAI's closed, API-only frontier models for the first time.

The announcement fits a broader shift in how enterprise AI is bought. Large companies increasingly route different tasks to different models based on cost, latency and capability rather than standardizing on one vendor. Google framed the deal in those terms, saying customers can select and swap models for each use case without separate vendor contracts. Greg Brockman, OpenAI's president and co-founder, called Google Cloud a leading platform for developers and businesses building AI agents, while Google Cloud CEO Thomas Kurian said the company is focused on giving customers a broad selection of models, including OpenAI's.

Industry implications

The most immediate change is commercial rather than technical. Enterprises with large committed-spend contracts on Google Cloud can now direct part of that budget to OpenAI models instead of negotiating a separate agreement, a friction point that has slowed adoption in procurement-heavy sectors such as financial services, healthcare and government contracting. Because usage counts against existing commitments, chief information officers gain a way to test OpenAI models without adding a new supplier to their vendor list.

Governance is the second lever. Google said customer prompts, context and outputs stay inside the customer's designated data perimeter and are not used to train foundation models, that OpenAI models run on dedicated Google Cloud infrastructure behind isolated gateways, and that every model interaction operates under an authenticated identity with VPC Service Controls and unified audit logging. Zero Data Retention options and Confidential Compute are listed as coming soon, so regulated buyers will likely wait for those features before moving sensitive workloads. Box, an early named customer, said it plans to give its users OpenAI models backed by Google Cloud's security and compliance controls.

Competitive and supply chain implications

For OpenAI, the agreement widens distribution beyond its own API and Microsoft's platform into a hyperscaler that also builds a competing model family, Gemini. That tension is notable: Google is choosing to sell a rival's flagship models because enterprise customers want choice inside one governed environment, and keeping those customers on Google infrastructure matters more than which model answers a given query. Expect pressure on other model developers to secure comparable multi-cloud placement and on cloud providers to match Google's billing and commitment terms.

On the infrastructure side, Google said the OpenAI models run on dedicated Google Cloud infrastructure, adding inference demand to its data center fleet. Neither company disclosed the hardware, capacity or data center locations involved, so the effect on chip, power and cooling supply chains cannot yet be quantified. The Codex integration also points to growing demand for agentic coding workloads, which tend to generate long, compute-intensive sessions rather than single queries.

Regional implications

Neither announcement named the Google Cloud regions where OpenAI models will be served during preview or at general availability. That detail will matter most in Europe, India and other jurisdictions with data residency rules, where enterprises often need inference to stay within national or regional boundaries. Until region lists and data-residency commitments are published, multinational buyers will likely limit early pilots to non-sensitive workloads.

In the United States, the partnership gives public-sector contractors and regulated industries another route to OpenAI models through an existing cloud relationship, subject to the compliance certifications Google Cloud extends to third-party models. For Asia-Pacific customers already standardized on Google Cloud, the single-invoice model removes a common barrier to cross-border software procurement.

What comes next

• General availability of OpenAI models in Model Garden, which OpenAI expects in the coming weeks.
• Publication of pricing, quotas and supported Google Cloud regions.
• Rollout of Zero Data Retention options and Confidential Compute for OpenAI workloads, both listed by Google as coming soon.
• Integration of OpenAI models into Google Antigravity and Google's Gemini business offerings.
• Additional enterprise customer announcements beyond Box as the preview expands.

Sources

• OpenAI: OpenAI models are coming to Google Cloud (Oct. 21, 2026)
• Google Cloud blog: Google Cloud expands model choice and flexibility, bringing OpenAI models to customers (Oct. 21, 2026)

Image: Google data center in Council Bluffs, Iowa, by Chad Davis via Wikimedia Commons, CC BY 2.0. Cropped and edited by Credence Research.