What happened
On 21 October OpenAI and Google Cloud announced that OpenAI's models are coming to Model Garden, Google Cloud's model catalogue. GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna are in private preview for selected enterprise customers from that day. General availability was promised "in the coming weeks", with no date. Antigravity and Gemini Agent are to follow, and developers can already route Codex CLI and Codex desktop requests to Google's infrastructure.
For procurement, the billing route matters most. The models are bought through Google Cloud Marketplace, usage lands on a single Google Cloud invoice, and it draws down existing spend commitments with Google. According to the announcement, prompts, context and outputs stay inside the customer's designated data perimeter and are not used to train models. Zero data retention and Confidential Compute, however, are listed as "coming soon".
The background is short. Microsoft's exclusivity on OpenAI models ended this year, AWS has hosted them since April, and Microsoft remains OpenAI's primary cloud partner. Google says it is now the only hyperscaler offering all three leading closed model families: its own Gemini, Claude and GPT. That is Google's claim, not an independent measurement.
What is not known yet
The launch-day materials contain no pricing for OpenAI models bought through Google and no list of regions where they will run. Quotas and model identifiers, which normally sit on the catalogue card, are missing too. For a company in Poland that has to keep data in the EU, the missing region is the biggest gap, and for now it cannot be closed.
Our read
Flagged as our interpretation: for the last few years the choice of model often decided the cloud, or added a second contract. If you wanted GPT but ran on Google Cloud, you opened a separate contract with OpenAI or a separate Azure subscription. After this announcement that reason disappears at all three major cloud providers. Model and cloud stop being one decision.
That does not mean the Google route pays off by itself. We expect, and this is speculation, that in many companies the spend commitment, not the price per task, will decide the purchasing route. A commitment you cannot use is lost, so anything that draws it down looks free on the spreadsheet. Yet today the Google route has an unpublished price, unknown regions and two data-protection features marked "coming soon". The convenience of one invoice is real; it just has to be weighed against what you give up for it.
Where the lock-in sits is changing too. When Stripe was buying OpenRouter, we wrote that the hedge against a single model provider had itself become a risk. Here it runs the other way: the model gets easier to swap, while the cloud contract through which you pay for every model at once carries more weight.
Why it matters
Private Equity
In a portfolio where companies pay OpenAI and their cloud provider separately, a consolidation opportunity appears. When assessing AI across portfolio companies, it is worth calculating how much model spend could move through existing cloud commitments, and what the company would lose on data-processing terms. Consolidating before general availability and before the price list is published is a bet, not a saving.
Enterprise
If your organisation runs on Google Cloud and buys GPT directly, do not move traffic during the preview. First get three things in writing: an EU region, the price per million tokens compared with OpenAI's list, and the date from which zero data retention applies. Only then does a comparison with your current contract make sense. While you are at it, check that procurement does not count commitment drawdown as savings on models, because those are two different numbers.
SMB / mid-market
If you have no spend commitment with Google Cloud, nothing changes for you today. Direct API access remains the simplest route. Do not sign a cloud commitment just to buy a model through a marketplace.
One move for this week
List the routes through which you pay for models today: directly with the provider, through a cloud, through a routing gateway. For each route note the price per million tokens, the processing region and the data-retention terms. When Google publishes its pricing and regions, you add one row and the comparison is ready. Describe your case: mailto:[email protected]?subject=Rozmowa%20z%20Aurora%20AI