OpenAI puts Astra, Sol, Luna and Codex inside Google Cloud, completing its sweep of three US clouds

For most of the generative AI era, OpenAI's route to corporate buyers ran through a single door. Microsoft held the exclusive license to the company's technology, and Azure was the storefront. That arrangement has been dismantled piece by piece, and the final major piece fell into place on October 21, 2026, when OpenAI and Google Cloud announced that OpenAI's frontier models and its Codex coding agent will be sold through Google Cloud's Model Garden and Marketplace. The rollout begins with a private preview for select enterprise customers, with broader availability promised in the coming weeks. OpenAI's flagship systems now sit inside the cloud businesses of all three major United States hyperscalers.

The path there was short but consequential. OpenAI's exclusive distribution arrangement with Microsoft expired in April 2026. CNBC reported on October 21 that the day after that exclusivity ended, Amazon Web Services unveiled plans to host OpenAI models, and that Google is now getting its shot. Each hyperscaler is vying for AI market share, and each now carries a version of the same catalog. OpenAI, founded in 2015 partly as a counterweight to the concentration of AI research at companies like Google, now depends on its oldest rival's infrastructure to reach enterprise developers.

For Google, the announcement is an argument about neutrality. Google Cloud has spent the year positioning the unit as an open platform offering customers a broader choice of models, after it struck a cloud deal with OpenAI to offer the company more computing capacity last year. Thomas Kurian, chief executive of Google Cloud, has told Reuters in a previous interview that the platform is open and welcomes any model popular with developers. The commercial stakes are considerable too: OpenAI was valued at $1.4 trillion and is targeting an IPO, while Google Cloud's annualized revenue growth accelerated to about 82% in the second quarter from 63% in the first quarter, helped partly by demand for AI models and custom chips.

Key Facts

Reuters reported on October 21 that OpenAI will make its frontier AI models and Codex coding agent available through Google Cloud in the coming weeks, deepening ties between two rivals. The partnership brings OpenAI's latest models, beginning with GPT-6 Astra, into Google Cloud's Model Garden. Google Cloud said in a blog post on October 21 that the private preview covers three systems: GPT-6 Astra, built for demanding work across coding, computer use, professional tasks and scientific research; GPT-6.1 Sol, offering near Astra intelligence at lower cost across agentic coding, computer use and professional work; and GPT-6 Luna, a fast, cost efficient model for everyday work at scale.

The commercial mechanics matter as much as the model list. Customers can buy the models through Google Cloud Marketplace and use existing cloud spending commitments for the purchases. Because OpenAI models are transacted directly through the Marketplace, all usage appears on a single Google Cloud invoice, draws down existing commitments and counts toward Flexible Savings Plans across model providers. Google Cloud said customers can dynamically select and swap the optimal model based on performance, latency and cost efficiency without architectural lock in or separate vendor contracts.

Google Cloud also framed the deal as a governance story. Customer prompts, context and outputs remain entirely within the designated data perimeter and are never used to train foundation models, the company said, with Zero Data Retention options coming soon. OpenAI models run on dedicated Google Cloud infrastructure with Confidential Compute coming soon, and all queries route through hardened, isolated gateways. Every agent and model interaction operates under an authenticated identity with VPC Service Controls and unified audit logging. Reuters reported on October 21 that the arrangement will let customers in regulated industries reach OpenAI models through Google Cloud's access controls, data governance tools and infrastructure to build so called agentic AI workflows.

Scale is part of the pitch. Google Cloud said its platform carries more than 200 first party, third party and open models alongside the OpenAI lineup, and that it is now the only hyperscale cloud provider offering access to all three leading closed source LLM families on one enterprise AI platform. CNBC reported on October 21 that Google will add GPT products to Model Garden and to Antigravity, its coding tool, in the coming weeks, while Reuters described the coding component as the Codex coding agent. Box vice president of AI products Yashodha Bhavnani was quoted in Google Cloud's announcement about using OpenAI models through Google Cloud alongside its security and compliance capabilities. Most business spending on AI goes to OpenAI and Anthropic, according to data from corporate card issuer Ramp, though some corporate IT departments have integrated Google's Gemini models into their systems.

Analysis

The bigger picture here is that frontier model distribution has become a cloud business, and the labs that build the models increasingly need the pipes more than the pipes need any single lab. A few years ago an exclusive cloud partnership was a moat. Today it is a channel. OpenAI's decision to sell through Google Cloud and Amazon Web Services, after the Microsoft exclusivity lapsed, reflects a straightforward calculation: model quality wins benchmarks, but distribution wins enterprise contracts, and enterprise contracts are what a company valued at $1.4 trillion and aiming at an IPO needs to show.

For Google the calculus is subtler and arguably more revealing. Google competes with OpenAI in core AI products and in search, yet it is now a reseller of OpenAI's models. The economics of building and deploying large models push the two toward cooperation even as their consumer products collide. Google Cloud's growth gives it the confidence to play the neutral platform of AI infrastructure, and neutrality is a good business when you own the data center, the chips and the billing relationship. The company is betting that being the marketplace matters more than being the only tenant.

The risk is commoditization, and it cuts toward OpenAI. Once a model is available on all three hyperscalers and can be swapped at will without architectural lock in, price and latency become the deciding variables rather than brand loyalty. That is excellent for enterprise buyers and brutal for margins. Google's own framing, which emphasizes dynamic model selection and cost efficiency, describes a marketplace in which no single model family holds permanent leverage. What this really means is that the cloud layer has won the argument about who controls enterprise AI adoption. The labs supply intelligence, and the hyperscalers supply the invoice, the compliance perimeter, the identity system and the audit trail.

Why It Matters

For enterprise buyers the practical effect is consolidation of procurement. A company that has already committed to spending with Google Cloud can now draw those commitments down on OpenAI models, count them toward Flexible Savings Plans and keep everything on one invoice. For regulated industries, the combination of Google Cloud's access controls, governance tools and audit logging with OpenAI's models may remove the compliance objection that stalled earlier deployments. Box's endorsement is an early signal that software vendors are already building on the combination.

For the cloud market the numbers tell the story. Google Cloud's annualized revenue growth of about 82% in the second quarter dwarfed Amazon's 37% and Microsoft's 42%. Volume beats exclusivity when you are growing that fast, and hosting a competitor's flagship models costs little in prestige if it wins workload share. The deal also puts pressure on Microsoft, which spent years positioning Azure as the home of OpenAI and now watches its former exclusive partner appear in rival storefronts.

For OpenAI the deal is a hedge against dependence. Its models now reach customers through three separate commercial channels, none of which it controls outright. That is safer than a single partner, but it also means OpenAI's pricing power is increasingly set by marketplace dynamics rather than by the mystique of a frontier lab. The $1.4 trillion valuation and the IPO ambition depend on converting that reach into durable revenue.

Next Up

The near term milestones are the ones Google Cloud listed in its announcement. Private preview customers are being onboarded now, general availability is promised in the coming weeks, Zero Data Retention options are coming soon, and Confidential Compute for OpenAI models on dedicated Google Cloud infrastructure is also on the roadmap. CNBC reported on October 21 that the GPT products will land in Model Garden and Antigravity in the coming weeks, so the developer tooling side of the rollout bears watching as closely as the model side.

Beyond that the questions are strategic. OpenAI is targeting an IPO and needs enterprise revenue to support its valuation. Google needs to prove that openness is a durable cloud strategy rather than a temporary tactic while it promotes its own Gemini 4 Argon model, announced in September, as a first party alternative. Microsoft, Amazon and Google now all sell the same frontier models, which means the next phase of competition will be fought over price, governance and developer experience rather than over who holds the exclusive.