GPT lands in Google Cloud, paid from budgets companies already promised to Google

On October 21, OpenAI and Google Cloud announced that GPT models will run in Model Garden, the model catalog on Google's Gemini platform. Companies that already run on Google Cloud get GPT without a new OpenAI contract and without another vendor in their books.

Three GPT models reach Google Cloud, for select customers only

According to the Google Cloud blog, three models are involved. GPT-6 Astra targets demanding coding, computer use and scientific work. GPT-6.1 Sol promises near-Astra performance at a lower price, and GPT-6 Luna is meant for cheap everyday work at scale. The models are rolling out in private preview to select enterprise customers. Everyone else can fill in an interest form.

Developers can call the models through the API and connect local Codex CLI and Codex Desktop directly to inference on Google Cloud. OpenAI models are also slated to arrive later in Gemini for business and in Google's Antigravity development environment. OpenAI's own announcement page could not be loaded during verification, so this piece relies on the Google Cloud blog and the text OpenAI posted on its community forum.

The accountant decides this, not the benchmark

The most interesting line in the announcement has nothing to do with performance. OpenAI models are bought through Google Cloud Marketplace. Usage lands on a single Google invoice, draws down existing cloud commitments and counts toward Flexible Savings Plans. A company that promised Google a certain level of spend can now burn part of that promise on GPT.

For procurement teams, that is a big relief. A new AI vendor usually means a security questionnaire, a legal review and a separate budget line. Here all of that disappears because the Google contract already exists. Google also claims to be the only hyperscaler offering Gemini, Claude and GPT side by side, along with more than 200 other models.

For OpenAI, it is another step away from a single cloud partner. Microsoft invested 1 billion dollars in the company in 2019, and for years Azure was the main way enterprises brought OpenAI models into their own infrastructure. The line "if you want GPT, go to Azure" is losing its force.

The most sensitive guarantees are still labeled "coming soon"

Regulated companies will look at data protection first. Google promises that prompts and outputs stay within the customer's data perimeter and are never used to train models. Zero Data Retention and Confidential Compute, however, are only marked "coming soon". A bank or a hospital that cannot send sensitive data to a model without them will have to wait.

The announcement gives no prices, no regions and no word on whether the models will run in European data centers. For a European company that must know where its data is processed, that is a serious gap. Private preview also reveals nothing about how fast the service will open to other customers.

Pricing and European regions will show how serious both sides are

The first signal will be general availability and the list of regions. If GPT reaches European Google Cloud regions with Zero Data Retention, it becomes a real alternative to Azure and to OpenAI's direct API. If it stays in private preview for a handful of customers for a long time, it is mostly a marketing move.

The second signal is price. If GPT on Google Cloud costs more than buying directly from OpenAI, companies will pay a premium for convenient procurement. The third is Codex: how many teams actually point their Codex CLI at Google Cloud inference instead of keeping their OpenAI account.

Lilith's verdict

A buyer whose budget is locked into a Google commitment can now order GPT without signing anything new with OpenAI. Microsoft just lost a comfortable argument it used for years to pull companies into Azure.