OpenAI Puts GPT-6 Astra and Codex on Google Cloud, Completing Its Break From Microsoft Exclusivity

OpenAI is bringing its most capable artificial intelligence models — including the flagship GPT-6 Astra and its Codex coding agent — to Google Cloud, the companies said on Wednesday, in a deal that completes the ChatGPT maker’s break from its long-standing exclusive relationship with Microsoft and places its technology on all three of the world’s biggest cloud platforms.

According to a company blog post cited by Reuters, OpenAI’s frontier models will arrive in Google Cloud’s Model Garden in the coming weeks, where customers will be able to buy them through the Google Cloud Marketplace and pay for them using their existing cloud spending commitments. The models will also be made available inside Google’s own developer tools, including Antigravity and the Gemini Agent, an unusually direct integration between two companies whose flagship products — ChatGPT and Gemini — compete head-on for users and enterprise contracts.

Announcing the move on Wednesday, October 21, OpenAI president Greg Brockman said: “This partnership is an important step in making our most capable models and Codex available where customers are already building.” Google Cloud chief executive Thomas Kurian said the company was “focused on giving customers a broad selection of models, including those from OpenAI.”

Reuters reports that the arrangement will also let customers in regulated industries access OpenAI models through Google Cloud’s access controls, data governance tools and infrastructure, aimed at building what the industry calls agentic AI workflows — systems in which models carry out multi-step tasks on behalf of users rather than simply answering questions.

The partnership makes Google the third major hyperscaler to distribute OpenAI’s models, after the company’s exclusive distribution arrangement with Microsoft expired in April. OpenAI has since expanded to Amazon Web Services and now to Google Cloud. CNBC reports that Amazon began hosting OpenAI models in April, shortly after the Microsoft exclusivity ended.

Despite the deepening ties, OpenAI was careful to preserve the hierarchy of its cloud relationships. “Microsoft remains OpenAI’s primary cloud partner,” the company wrote in its announcement, according to the Tesla News Wire’s account of the post. The two rivals are now bound together in multiple ways: Reuters notes that Google last year struck a separate cloud deal to provide OpenAI with more computing capacity.

The end of the exclusive era

For years, OpenAI’s relationship with Microsoft was one of the most scrutinised alliances in technology. Microsoft was not only the company’s exclusive cloud provider but also its largest backer and, after a carefully negotiated restructuring, holds the exclusive licence that lapsed this April. That exclusivity gave Azure a unique moat: enterprises that wanted OpenAI’s models at scale essentially had to route through Microsoft’s cloud.

The expiry of that arrangement has been followed by a rapid, deliberate diversification. OpenAI moved first to Amazon Web Services in April — a deal that CNBC noted came the day after exclusivity ended — and has now completed the set with Google. As a result, corporate developers will soon be able to access frontier OpenAI models from whichever of the three clouds already hosts their data and applications.

The shift reflects a structural reality of the AI business in 2026. The exclusive tie-up made sense when OpenAI was a fast-growing startup that needed a patron to fund its enormous computing bills. With the company now courting investors at a reported $1.4 trillion valuation and racing to raise a further $30 billion, according to Reuters reporting this week, it can no longer afford to leave enterprise money on the table. OpenAI has said the expansion is critical for building its enterprise business beyond its consumer roots with ChatGPT.

A truce between rivals

The deal is also a striking example of the coopetition that now defines the AI industry. Google’s Gemini models compete directly with GPT across consumer assistants, search, enterprise contracts and developer APIs — Google announced its next-generation Gemini 4 Argon model only last month. Yet the economics of building and deploying large models keep pushing the two companies toward cooperation even as they fight for the same customers.

For Google Cloud, the logic is straightforward. Kurian has previously told Reuters that the Google cloud platform is an open platform and welcomes any model popular with developers. Adding OpenAI’s models strengthens a strategy of offering customers a wider range of third-party models rather than relying only on its own — a neutral-marketplace posture that differentiates Google from Microsoft, whose cloud remains most closely associated with a single AI provider. CNBC notes that Google’s cloud revenue growth accelerated to about 82% in the second quarter, versus roughly 37% at Amazon and 42% at Microsoft, with all three reporting quarterly results next week.

Analysis: Why It Matters

Distribution is now the battlefield, not just the models

Strip away the announcements and this week’s AI news reads like a single story told in several chapters. Anthropic’s draft S-1 filing put hard numbers on the cost of the frontier race: $80.7 billion in net losses in six months and at least $518 billion committed to compute over the next decade, even as the company set aside 5% of its shares for a public benefit fund ahead of its planned IPO. Meta launched its first frontier API model, Muse Wonder, at aggressive pricing. DeepSeek open-sourced the inference engine behind its production services.

The pattern is unmistakable: the models are increasingly abundant and increasingly expensive to serve. What is scarce is distribution — the channels through which models actually reach paying enterprise customers. By placing GPT-6 Astra and Codex wherever customers already build, OpenAI is treating distribution as the primary product, not an afterthought. The $80 billion Anthropic loss figure, much of it accounting noise, matters less than the $518 billion promise of future compute spending: someone has to pay for the next decade of model serving, and the answer, for now, is enterprise subscriptions routed through cloud marketplaces.

The procurement argument is the quiet killer feature

The most consequential line in the announcement may be the dullest: customers will be able to use their existing cloud spending commitments for the purchases. Large enterprises typically negotiate multi-year committed-spend contracts with a single cloud provider, often worth hundreds of millions of dollars. Until now, a company committed to Google Cloud that wanted OpenAI’s models faced procurement friction — a second vendor, a second bill, a second security review.

Removing that friction is how enterprise software is really sold. The model with the best benchmark score does not always win; the model that fits inside an existing procurement process does. Axios framed it the same way this week: winning enterprise AI is about making models available wherever companies already run their software and store their data. OpenAI’s expansion to all three clouds is, in this light, less a technology announcement than a procurement announcement.

Regulated industries are the next frontier

Reuters specifically highlighted that customers in regulated industries will be able to access OpenAI models through Google Cloud’s access controls, data governance tools and infrastructure. Banks, insurers, healthcare providers and governments have been among the slowest adopters of frontier AI precisely because of data residency and governance concerns. Google Cloud’s compliance certifications and its Gemini Agent tooling give OpenAI a pre-built route into sectors where trust and auditability matter more than raw model capability.

The agentic AI workflows both companies emphasise — systems that act on a user’s behalf across software tools — are the use case enterprise buyers are actually being pitched this year. Codex, OpenAI’s coding agent, arriving inside Google’s Antigravity development environment suggests the first battleground will be the software development lifecycle itself: the companies are not just selling model API calls, they are selling the tooling developers live inside.

The cloud war just got a new front line

With all three hyperscalers reporting earnings next week, the timing is pointed. Google’s reported 82% cloud growth figure gives Kurian’s open-platform argument commercial weight: customers are voting with their wallets for a cloud that hosts everyone’s models rather than one vendor’s. Microsoft still holds its position as OpenAI’s primary partner and a major stakeholder in its success, but the aura of exclusivity — the sense that Azure was the only serious home for OpenAI’s models — is gone.

That leaves a cleaner competitive picture: three clouds, each offering the same frontier models, competing on speed, security, price and developer experience. In such a market, differentiation shifts from who has the model to who serves it best. For customers, that is arguably the healthiest possible outcome — and the one least likely to persist if consolidation or capacity constraints return.

The safety conversation continues in parallel

The deal lands in a week heavy with AI safety and policy news. Eight US states have sued OpenAI, alleging ChatGPT gave dangerous medical advice and coached self-harm — the broadest state-level legal challenge the company has faced. The US Department of Homeland Security has ordered federal civilian agencies to remove all DeepSeek technology, including third-party tools built on it, deepening the split between American and Chinese AI stacks. And OpenAI, Anthropic and Google DeepMind have confirmed weeks of coordinated work on cyber-focused AI safety measures, a notable departure from their usual competitive posture.

Distribution and safety are two faces of the same coin. The more widely frontier models spread — across clouds, into regulated industries, inside developer tools — the more the question shifts from who builds the models to who governs their use. The Google deal makes OpenAI’s models more accessible; the lawsuits and the DeepSeek order are reminders that accessibility and accountability are now being litigated simultaneously.

What to Watch Next

1. General availability and pricing: the models are expected in Google Cloud’s Model Garden “in the coming weeks,” with access currently by request. Pricing on Google Cloud had not been published in the announcements — enterprise buyers will compare it closely with Azure and AWS terms.
2. Codex inside Antigravity and the Gemini Agent: whether OpenAI’s coding agent genuinely changes developer workflows inside Google’s tooling will be the first real test of the partnership’s depth, beyond the distribution mechanics.
3. Cloud earnings next week: Azure, AWS and Google Cloud all report. Watch for AI-related growth figures — and for commentary on how model marketplaces are affecting margins.
4. The funding race: OpenAI’s reported $30 billion raise at a $1.4 trillion valuation and Anthropic’s planned November IPO set the backdrop. Distribution deals like this one are the revenue story both companies must tell investors to justify their compute spending.
5. The regulatory front: the eight-state lawsuit against OpenAI, the DHS order on DeepSeek, and the joint safety work by OpenAI, Anthropic and Google DeepMind will determine how far — and how freely — frontier models can be distributed in regulated markets.

Sources

• Reuters: OpenAI to sell its top models on Google Cloud in latest enterprise push (21 October 2026)
• Tesla News Wire: OpenAI models and Codex coming to Google Cloud Model Garden (21 October 2026)
• CNBC (via Remotestake): OpenAI to bring AI models to Google’s cloud, in further expansion beyond Microsoft (21 October 2026)
• The Daily Post of AI, Issue 16: Anthropic’s draft S-1 — $80.7bn loss and $518bn compute commitment (21 October 2026)
• AI Briefing: DeepSeek DeepInfer open-sourced; DHS orders DeepSeek removal; eight US states sue OpenAI (21 October 2026)