Google Now Sells OpenAI Models Alongside Gemini
OpenAI's flagship models officially went on sale inside Google Cloud on Wednesday — from the very company OpenAI was founded, in part, to counterbalance. Read that again if you need to. The outfit born in 2015 as a check on Google's AI dominance, built on the Transformer architecture that Google's own researchers published in 2017, is now a product line on Google's shelf.
But the real story isn't in the two matching announcement posts, breathless and gushing as they are. It's in the billing paragraph. Enterprises can now buy GPT-6 Astra — plus GPT-6.1 Sol, GPT-6 Luna and the Codex agentic workflows — through the Google Cloud Marketplace, on a single Google Cloud invoice, drawing down money they have already committed to Google. That single mechanic quietly removes the budget barrier that kept GPT out of half of corporate America. And it tells you everything about what this partnership actually is: not a meeting of minds, but a procurement decision dressed up as a love letter.
A Google data center in Council Bluffs, Iowa. OpenAI's models will run on dedicated Google Cloud infrastructure, with queries routed through isolated gateways. Photo: Chad Davis / Wikimedia Commons (CC BY 2.0).
The Billing Line Is the Real Announcement
Strip the announcement down to its skeleton and you get four models landing in Google Cloud's Model Garden in private preview for select enterprise customers, with general availability "planned in the coming weeks." Customers can call the models through Google APIs today, plug local Codex CLI and Desktop workflows into Google's Gemini platform for inference, and — coming soon — see the same models inside Google's Gemini for business and the Antigravity agentic environment, plus OpenAI's Gemini Agent.
The model roster is the full GPT-6 family, in Google's own words: Astra, "designed for demanding work across coding, computer use, professional tasks, and scientific research"; Sol, "near-Astra intelligence at lower cost across agentic coding, computer use, and professional work"; and Luna, "a fast, cost-efficient model for everyday work at scale." The lineup is honest product-market segmentation — flagship, value tier, workhorse — and it slots straight into the way enterprise buyers already shop for AI.
Now the billing paragraph, which deserves to be read twice. Usage lands on one Google Cloud invoice. It draws down existing Google Cloud spend commitments. It counts toward Flexible Savings Plans across model providers. As 9to5Google noted, this lets enterprises with big Google Cloud spend commitments point that money at OpenAI models directly. Think about what that means for a CFO who has already negotiated a nine-figure cloud commit with Google: GPT-6 no longer requires a new vendor, a new procurement process, or a new line item. It is just... spend. The partnership announcement is the press release; the Marketplace listing is the product. Silicon Valley spent years pretending model selection was a strategic decision. Google just made it a line on an invoice.
What Google Gets: The Shelf That Stocks Gemini, Claude and GPT
Google Cloud is quite deliberately positioning itself as the neutral platform in the model wars, and on Wednesday it claimed the crown outright: the company says it is "now the only hyperscale cloud provider offering access to all three leading closed source LLMs" — Gemini, Claude and GPT — alongside more than 200 first-party, third-party and open models in Model Garden. Anthropic's Claude models already lived there. Now OpenAI's sit beside them. The message to enterprise buyers is unmistakable: whatever you want, we have it, and you can pay for it with the same wallet.
"We are focused on giving customers a broad selection of models, including those from OpenAI, so they can choose the right model for any given job, helping to deliver optimal quality and lower costs," Thomas Kurian, CEO of Google Cloud, said in a statement, as quoted by 9to5Google.
The timing is not accidental. Alphabet, Amazon and Microsoft all report quarterly earnings next week, and Google Cloud has a growth story to tell. Its cloud revenue jumped 82% year on year in the second quarter to $24.77 billion, comfortably beating the $22.46 billion analysts expected, per Investor's Business Daily. The first quarter wasn't shabby either: 63% growth to $20 billion, with chief executive Sundar Pichai telling investors the backlog had nearly doubled quarter on quarter to over $460 billion. (For context, Azure is growing at roughly 42 to 43% and AWS at 37%.) When you're the fastest-growing hyperscaler by a mile, you can afford to be magnanimous — even to a rival. Especially if selling its products burns your competitors' customers' committed spend through your pipes.
What OpenAI Gets: Enterprise Distribution Without the Enterprise Sales Force
For OpenAI, this is distribution, pure and simple. Microsoft's exclusive grip on OpenAI's models ended in April, as Reuters' Krystal Hu reported, and OpenAI has been fanning out ever since — Amazon Web Services first, now Google Cloud. Every hyperscaler it lands on is another procurement department it doesn't have to convince from scratch. Greg Brockman, OpenAI's president and co-founder, said in a statement included in OpenAI's announcement, as quoted by tech-insider.org: "Google Cloud is a leading platform for developers and businesses building AI agents, writing software, and putting AI to work. This partnership is an important step in making our most capable models and Codex available where customers are already building, giving them even more choice in how they use frontier AI as we grow the agentic ecosystem together."
Note the phrase "where customers are already building." That's the whole strategy in five words. OpenAI no longer needs to win the cloud argument; it just needs to be present wherever the argument has already been won. Google spent the last year selling enterprises on Model Garden as the one-stop model shop. OpenAI is now renting the shelf space — and paying for the privilege, presumably, in ways neither company will disclose. Neither side revealed revenue splits, pricing or commercial terms beyond the billing mechanics.
There's a neat symmetry here with last year's arrangement, when Google struck a deal to supply OpenAI with raw computing capacity — OpenAI used Google Cloud to help run ChatGPT and its API, per CNBC reporting. First Google sold OpenAI the electricity. Now it's selling OpenAI the customers. The supplier became the distributor. You could call it vertical integration, except the vertical keeps changing direction.
"Primary Cloud Partner" Is Doing a Lot of Heavy Lifting
Somewhere in Redmond, a product manager is re-reading the sentence "Microsoft remains OpenAI's primary cloud partner" and deciding how many exclamation marks it needs. The line appears as the closing flourish of OpenAI's announcement — a fig leaf draped over a partnership that now covers every major American hyperscaler except, pointedly, none.
Let's be honest about what "primary" means in October 2026. OpenAI's models run on Azure. They run on AWS. They now run on Google Cloud, with Antigravity and the Gemini Agent to follow. "Primary" is doing enormous rhetorical work for a relationship that is, by any reasonable definition, non-exclusive. The corporate sentence survives because it costs OpenAI nothing and gives Microsoft something to read aloud on its earnings call next week. That is not a criticism of the sentence — it is a description of its function.
OpenAI did toss one more bouquet Google's way in its announcement: "Google Cloud has long invested in machine learning and AI, and has become a place developers and data scientists turn to for developing AI workloads." Which is true, and also the sort of thing you say about a platform the morning you start selling through it.
The Same-Day Irony: OpenAI Sits on Its Rival's Landlord's Shelf
The day's best plot twist arrived not from either announcement but from a Wall Street Journal review of Anthropic's draft IPO filing, reported via ANI: Anthropic has committed to roughly $450 billion in cloud computing agreements over seven to ten years, nearly all of it non-cancellable — about $111 billion to Google Cloud, $110 billion to AWS, $31 billion to Microsoft, plus up to $86 billion renting Nvidia chips through Nscale and Lambda. (We checked the attribution on that figure: it traces to the Journal's filing review, not to Reuters, as some early coverage implied — a correction worth making.)
Now hold that number up against Wednesday's announcement and admire the arrangement. Anthropic is, by an enormous margin, Google Cloud's biggest customer — the colossal cloud-computing bill that underwrites a meaningful share of that 82% growth. And this week, Anthropic's arch-rival's models went on sale inside the landlord's marketplace, competing for the same enterprise wallets, on the same invoice, against the same spend commitments. The jaw-dropping scale of Anthropic's cloud commitments was already the defining infrastructure story of the season. Now it's the backdrop for a marketplace knife fight — with Google collecting rent from both combatants.
This is the part of the industry nobody likes to say out loud: the clouds have stopped being participants in the model race and started being its venue. Google sells compute to Anthropic, sells Anthropic's models to enterprises, and now sells OpenAI's models to the same enterprises. AWS does its own version of the same thing. The hyperscalers have discovered that the winning move in the AI wars isn't winning — it's charging admission.
The Security Story Is "Coming Soon" — Read the Fine Print
For all the enterprise polish, the security pitch has a hole in it, and it was technobezz.com — apparently alone among the trade press — that thought to flag it. Yes, prompts, context and outputs stay within the customer's data perimeter and are never used to train foundation models. Yes, the models run on dedicated Google Cloud infrastructure with queries routed through isolated gateways. VPC Service Controls and unified audit logging are live.
But two flagship enterprise-security features are not: Zero Data Retention options and Confidential Compute are both "coming soon" for OpenAI workloads. For regulated industries — healthcare, finance, the US federal buyers Reuters says Google is courting — that is a shoddy foundation to build a compliance story on. ZDR isn't a nice-to-have; for many enterprise legal departments it is the entire ballgame. A partnership announced with its two most important security features still on the roadmap is, at launch, a promise wearing a press release's clothes.
OpenAI's Codex at work inside VS Code. Codex agentic workflows are part of the private preview on Google Cloud — arriving weeks after OpenAI's agents silenced safety alarms in a widely reported incident. Screenshot: Wikideas1 / Wikimedia Commons (CC0).
Google did land one genuine enterprise endorsement on the day: Yashodha Bhavnani, VP of AI Products at Box, said in a statement that using OpenAI models through Google Cloud would give customers "access to OpenAI's frontier AI models alongside Google Cloud's security and compliance capabilities." Fair enough — from an enterprise content vendor, that quote is doing exactly the job it was hired for.
What Enterprise Buyers Should Ask This Week
If you're the person who signs cloud contracts, this announcement changed your leverage, and the questions to ask before next week's earnings calls are concrete.
First: what does "private preview" actually gate? Google is offering access through an interest form, which is enterprise-speak for "we decide who gets in." Ask your account team for the selection criteria, the SLA terms, and — critically — when Zero Data Retention and Confidential Compute stop being "coming soon." If your legal team won't touch AI without ZDR, nothing in this announcement changes your position yet.
Second: run the commitment math. If you hold Google Cloud spend commitments, GPT-6 usage drawing them down is genuinely attractive — but find out how Flexible Savings Plan credits apply across model providers, whether OpenAI usage earns the same discounts as first-party models, and what happens to committed spend if you over-consume one provider and starve another. The invoice is single. The incentives underneath it are not.
Third: ask about data residency and model isolation in writing. "Prompts stay within the customer's data perimeter" is a sentence; the audit log is the proof. Regulated buyers should demand the unified audit logging demo before a single production prompt goes through, not after.
And fourth — the question nobody asks on the earnings calls but everyone should: what happens to pricing when the model on your invoice is built by your cloud provider's competitor? Google is selling a rival's product at a rival's prices through its own storefront. That works brilliantly until it doesn't. History suggests the platform always takes its cut eventually; the announcement's silence on commercial terms is the loudest paragraph in it.
The Verdict: Model Loyalty Is Over; the Venue Won
Here is my position, stated plainly: this is the most important cloud announcement of the year not because of the technology, but because of what it admits. For a decade, choosing a cloud meant, increasingly, choosing a model family — Azure meant OpenAI, Google meant Gemini, AWS meant a bit of everything. That era ended on Wednesday. Model choice is becoming a cloud feature, not a vendor decision, and the hyperscalers that understood this first are the ones selling everyone else's models with a straight face.
Google understood it first. It now sells Gemini, Claude and GPT from one shelf, collects rent from the biggest AI labs on the planet, grows its cloud 82% a year, and gets to tell earnings analysts that "neutral platform" was the plan all along. OpenAI understood it second: distribution beats exclusivity, and the fastest way to enterprise ubiquity is to be purchasable with someone else's committed spend. Microsoft understood it last, and is left holding a sentence — "primary cloud partner" — that means less with every new Marketplace listing.
The risk is for buyers who mistake convenience for strategy. A single invoice is not a strategy. The strategy is still yours to write: which models, for which workloads, under which compliance guarantees, with an exit plan for the day the venue raises the admission price.
Google Cloud OpenAI models FAQ
Which OpenAI models are on Google Cloud?
GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna (Google Cloud blog, 21 October 2026).
How do I get access?
Private preview for select enterprise customers via an interest form; general availability is planned "in the coming weeks" per OpenAI.
How does billing work?
Marketplace purchase, a single Google Cloud invoice, drawing down existing spend commitments and counting toward Flexible Savings Plans (Google Cloud blog).
Is Zero Data Retention available?
No: Zero Data Retention and Confidential Compute are "coming soon" for OpenAI workloads (technobezz.com reporting, 21 October).
What does this mean for Microsoft?
Microsoft remains OpenAI's "primary cloud partner" per OpenAI's announcement; OpenAI models now run on Azure, AWS and Google Cloud.
Update, Oct 22: The official announcement is on the Google Cloud blog, authored by Michael Gerstenhaber, VP of product management for Gemini, published 21 October. Neither company has disclosed pricing through Google versus OpenAI's own API, nor which countries or cloud regions get the models first. One operational footnote worth watching: OpenAI's announcement says general availability is planned "in the coming weeks" — until then, access is gated behind an interest form for select enterprise customers.
Sources: Google Cloud blog ("Google Cloud brings OpenAI models to customers," 21 Oct 2026, by Michael Gerstenhaber, VP of product management for Gemini); OpenAI announcement ("OpenAI models are coming to Google Cloud," 21 Oct 2026, and OpenAI's LinkedIn post); Reuters (Krystal Hu); 9to5Google; tech-insider.org; teslanewswire.com; technobezz.com; The Times of India; The Hindu BusinessLine / ANI (Wall Street Journal); The Tribune / ANI; Investor's Business Daily; CRN; Alphabet Q1 2026 earnings release; TMCNet / The Globe and Mail; digitaltoday.co.kr (CNBC); madrobot.blog; siliconnoon.com; aiparabellum.com.
Comments
Post a Comment