OpenAI Is Now on All Three Big Clouds, and the Real Change Is on the Invoice: Google Cloud Will Resell GPT-6 Against Its Own Gemini

OpenAI's models are now sold through Google Cloud, the company that builds the Gemini family OpenAI competes with. With that, OpenAI's frontier models reach customers on all three of the largest US cloud providers. The headline is a rivalry story. The part builders should read closely is the billing paragraph in Google's announcement, because it changes how a company can buy these models.

The announcement is partly in preview and partly "coming weeks," so this post separates what is live from what is promised.

What Actually Happened

On Wednesday, October 21, OpenAI and Google said OpenAI will make its frontier models and its Codex coding agent available through Google Cloud, per Reuters. CNBC reports the GPT products are coming to the cloud's model garden and to Google's Antigravity coding tool in the coming weeks, and that the arrangement puts OpenAI's models on all three of the top cloud infrastructure providers.

Google Cloud's own post names the models. GPT-6 Astra is described as built for demanding work across coding, computer use, professional tasks and scientific research. GPT-6.1 Sol is described as near-Astra intelligence at lower cost. GPT-6 Luna is described as a fast, cost-efficient model for everyday work at scale. Google says all three are rolling out now in private preview to select enterprise customers, with more coming soon.

Two things work today, according to Google. Customers can build with the models through Google's APIs, and they can connect local Codex CLI and Desktop workflows to Google's Gemini platform for inference. Availability inside Gemini for business and inside Antigravity is listed as coming soon.

Google also made a positioning claim in the same post: that it is now the only hyperscale cloud provider offering access to all three leading closed-source large language models, alongside more than 200 first-party, third-party and open models. That is Google's characterization, not an independent finding. The three it refers to are Gemini, Claude and GPT.

The history matters for context. CNBC reports that less than six months ago OpenAI said Microsoft would no longer have the exclusive license on its intellectual property, and that Amazon unveiled plans to host OpenAI models the day after that exclusivity ended. Google is the third. The Information reports the Google deal lessens OpenAI's reliance on Microsoft, where the two companies once shared exclusivity.

Why Google Wants This: The Cloud Growth Numbers

Google does not need OpenAI to prove its cloud is growing, which makes the deal more interesting. CNBC reports that annualized revenue growth in Google's cloud segment accelerated to about 82% in the second quarter from 63% in the first, driven partly by demand for AI models and custom chips. For comparison, CNBC reports Amazon and Microsoft posted second-quarter cloud growth of 37% and 42%, respectively. All three report quarterly results next week.

Those percentages are on different revenue bases, so they are not a head-to-head ranking. Still, the direction explains the motive. A cloud provider growing this quickly on AI demand has reason to stock every model its customers might ask for, including a rival's. Google sells compute, storage, security tooling and billing on top of any model, and each workload placed on its platform is revenue regardless of whose weights run it.

There is also a demand-side gap that CNBC points to. Citing data from corporate card issuer Ramp, it reports that most business spending on AI goes to OpenAI and Anthropic, though some corporate IT departments have integrated Google's Gemini models into their systems. If the money is flowing to OpenAI and Anthropic, a cloud that only offers its own model is leaving enterprise spend with other vendors' billing relationships. Reselling GPT brings that spend onto Google's invoice.

CNBC adds that Google announced its next-generation Gemini 4 Argon model in September. Google is hosting OpenAI's models while pushing its own. It is not conceding the model race, and its framing is explicitly about matching each workload to the right model.

The Procurement Mechanics Are the Real Story

The most consequential sentence in Google's post is about money. Because OpenAI models are transacted directly through Google Cloud Marketplace, Google says all usage appears on a single Google Cloud invoice, draws down existing cloud commitments, and counts toward Flexible Savings Plans across model providers. IJR's coverage of the announcement stresses the same point: customers can use existing cloud spending commitments and Google's governance tools to access the models.

For a large company, that changes a familiar chain of approvals. A team that wants GPT-class models today often needs a separate vendor agreement, security review, procurement ticket and budget line. If the same models are billed against a cloud commitment the company already negotiated and already reviewed, the path is shorter. Committed cloud spend that would otherwise go unused can now be spent on a model it could not buy through that channel before.

Google's governance claims are part of the pitch, and some of them are still future tense. The post says customer prompts, context and outputs remain within the designated data perimeter and are never used to train foundation models. It says Zero Data Retention options are coming soon, and that OpenAI models run on dedicated Google Cloud infrastructure with Confidential Compute also coming soon. It describes unified audit logging and authenticated identity for every agent and model interaction. Treat the "coming soon" items as roadmap, not capability, until they ship and a security team verifies them.

The same pattern already exists on the other clouds in a different form. Seeking Alpha frames the news as OpenAI products now being available through "the trinity of cloud giants," with Google joining Microsoft and Amazon. Once all three offer the same model family, the model stops being the differentiator in the cloud decision. Pricing, quotas, regional availability, data controls and existing commitments take over.

What This Does and Does Not Change

It does not end Microsoft's role. Coingape's coverage describes Microsoft as remaining OpenAI's primary cloud partner, and nothing in the sources in hand suggests workloads are moving. It also does not change OpenAI's own business pressure. CNBC reports OpenAI is courting investors at a $1.4 trillion valuation and targeting an initial public offering, and describes the cloud push as part of building the business beyond its consumer roots. Distribution through three clouds is an enterprise revenue play that reaches customers who buy through their cloud account instead of signing with OpenAI directly.

It does not guarantee performance or price parity across clouds. Google's post gives no pricing for the three models, and the sources in hand do not say whether per-token rates match OpenAI's direct API or the other clouds. The preview is limited to select enterprise customers, and the "coming weeks" language from Reuters and CNBC means broad availability has no confirmed date.

It does change the negotiating position. A buyer that can reach Gemini, Claude and GPT inside one console and one commitment has a stronger hand on price than one locked into a single model vendor. Google's pitch of swapping models "without architectural lock-in or separate vendor contracts" is aimed at exactly that buyer.

What Builders Should Take From It

• Check which cloud holds your spend commitment before you pick a model vendor. If GPT-class models can now draw down a commitment you already have, the effective cost and approval time may differ from a direct contract. Ask your cloud rep for per-token pricing in writing, because none was published.
• Build a model-routing layer, not a model dependency. With the same families reachable on multiple clouds, the durable asset is a thin abstraction that lets you swap models by task, latency and cost. Hard-coding one vendor's SDK is the expensive choice.
• Treat "private preview" and "coming soon" as unshipped. Zero Data Retention, Confidential Compute, Antigravity and Gemini for business availability are all future items in Google's post. Do not promise customers features that depend on them.
• Re-run your data-residency and audit review per cloud. Google describes data perimeters, VPC Service Controls and unified audit logging. Verify how those apply to a third-party model before moving regulated workloads, and confirm whether your existing reviews carry over.
• Watch next week's earnings. Amazon, Microsoft and Google all report quarterly results next week, per CNBC. Listen for how each describes third-party model revenue and AI commitments, which will show whether hosting a rival's model is a meaningful line item.
• If you sell into enterprises, expect buyers to ask for multi-cloud model support. Customers who can buy any frontier model through their cloud account will ask why your product only works with one.

Developer312 covers the AI business signals builders actually need to act on. Get the weekday briefing at developer312.com.