Enterprise customers who opened the Google Cloud console on Oct. 21 found a name in the model picker that had no business being there a week earlier. Alongside Gemini sat GPT-6 Astra, with GPT-6.1 Sol and Codex in the same catalogue. Google Cloud announced the addition that day, and OpenAI confirmed the arrangement.
The three models, two frontier chat systems and a coding agent, now run inside Model Garden, the company’s catalogue of first-party, third-party, and open models. Customers reach them through the console, the billing account, and the data-governance controls they already use. Access opened as a private preview for a group of enterprise accounts, according to the announcement.
Google Cloud said the move makes it the sole hyperscaler selling all three leading closed model families on one platform: its own Gemini, Anthropic’s Claude, and now OpenAI’s range. The catalogue holds more than 200 models. Charges run through Google Cloud Marketplace and count against existing cloud commitments.
That last detail may matter more to buyers than the model list itself. Large enterprises sign multiyear committed-spend contracts, and unused commitments are a chronic source of friction between procurement teams and the business units that buy AI. Letting a customer draw GPT-6 against money already promised to Google removes an internal argument that has stalled plenty of pilots.
For OpenAI the appeal is distribution. The company has spent two years pushing its models into corporate accounts where the buyer is a platform team, not an AI enthusiast. Selling through a rival’s console is unusual, but it puts OpenAI in front of Google’s enterprise base without asking those customers to open a second vendor relationship, a second security review, or a second invoice.
The arrangement would have been hard to imagine in 2023. Microsoft’s partnership with OpenAI, struck in 2019, made Azure the default home for its models, and for a period the two companies behaved as though the boundary were exclusive. OpenAI has since spread its compute across other providers, including Oracle, and has grown used to running on infrastructure that Microsoft does not own. The exclusivity softened before it broke.
Google Cloud’s own position has shifted too. It spent years as the third-place cloud, credible on data analytics and weak on the workloads that mattered most to AI startups. Its model strategy leaned on Gemini and a rotating cast of open-weight models, with Anthropic’s Claude as the marquee third-party tenant. Adding OpenAI turns the catalogue into a neutral storefront, which is a different pitch: come for the governance and the billing, stay for whatever model wins that quarter.
Analysts said the competitive pressure runs in both directions. A cloud provider that stocks a rival’s frontier model admits its own model cannot win every workload. It also captures the spend regardless of which model the customer picks, which is how marketplaces have always worked. “The cloud companies are becoming model-agnostic retailers,” said one analyst who follows enterprise AI procurement. “The model is the SKU.”
For enterprise architects, the practical effects are narrower than the press release suggests. Data residency, audit logging, and the terms under which a model provider can see prompts all still need review. OpenAI models running on Google infrastructure raise a new question about which company’s data-processing agreement governs a request, and Google’s announcement did not spell out the answer. Buyers will ask.
Google did not disclose pricing for the three models, or whether rates differ from what OpenAI charges customers directly. In private previews those numbers get negotiated alongside the cloud commitment, which means the effective discount varies by account and rarely appears in a press release.
There is also the matter of commitments. Marketplace spending that counts toward an existing contract gives procurement a clean line item, but it also concentrates dependence. A company that routes chat, code generation, and document analysis through a single cloud account has fewer exit paths than one that splits the work, and finance teams have started to notice.
Google’s own incentive is straightforward. Every model sold through Model Garden strengthens the argument that its cloud is the neutral layer where enterprises assemble AI stacks, rather than a store that only stocks house brands. That argument gets stronger with each new tenant, and weaker every time a customer builds directly against a single vendor’s API and skips the platform entirely.
Private preview terms were not disclosed. Google Cloud did not say how many customers are in the first group or when general availability would follow. OpenAI’s confirmation was equally brief.
The timing is pointed. Enterprise AI budgets for next year are being set now, and every hyperscaler wants its name on the line item before the money is allocated. Google has spent the past year arguing that choice, not lock-in, is what buyers want. Oct. 21 was the first time it could make that argument with OpenAI’s models standing behind it, and the first time OpenAI had to accept that a competitor’s console might be its fastest route into the Fortune 500.
