OpenAI Partners with Google Cloud, GPT-6 Series and Codex Launch

Wow, GPT is moving into Gemini's home!

Just today, OpenAI and Google Cloud officially announced a partnership:

GPT-6 Astra, GPT-6.1 Sol, GPT-6 Luna, and Codex will officially launch on Google Cloud.

The first batch of models is already being opened to some enterprise customers, and next they will also be put into Google Antigravity and Gemini Agent.

△

How is it that in the blink of an eye, sworn rivals have suddenly become channel partners and started doing business with each other?

More importantly, after Azure and AWS, OpenAI has now officially gathered all three major U.S. cloud providers.

OpenAI, once firmly locked inside Azure, is now completely leaving the era of exclusive distribution behind.

Sworn rivals have started selling each other's goods

Currently, GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna have entered Google Cloud's model garden and opened private previews to some enterprise customers.

Enterprises can directly call these models through Google Cloud's API, or connect local Codex CLI and desktop workflows to Google's Gemini platform, letting Codex complete inference via Google Cloud.

△

OpenAI says it will roll out a general availability plan in the coming weeks.

What may really strike a chord with enterprise customers in this partnership are probably these two points:

1. Enterprises can directly buy OpenAI's models in Google Cloud Marketplace using the funds in their Google Cloud accounts.

2. The invocation costs of different models can be consolidated into the same Google Cloud bill, and can even be included in cross-model Flexible Savings Plans.

This means that in the future, enterprises can freely switch between the two companies' models without re-signing contracts or opening separate bills.

When they want to use them, they can simply check the box in the familiar Google Cloud work environment.

At the same time, enterprises can continue to use Google Cloud's existing data governance, security, and permission control systems; prompts, context, and outputs all remain within the specified data boundaries, while model invocations must pass through an isolated gateway and unified identity authentication.

Once OpenAI's models launch on Google Antigravity and Gemini Agent in the future, everyone will be able to flexibly switch models to get work done according to actual needs.

Google Cloud is also quite proud of this, and couldn't resist putting its hands on its hips and boasting in a blog post:

We have become the only hyperscale cloud provider on the market that simultaneously offers the three leading closed-source large models, plus more than 200 self-developed, third-party, and open-source models!

Wait, wasn't OpenAI previously bound exclusively to Azure?

That's right.

In the past, Azure held the exclusive cloud distribution rights for OpenAI's stateless API.

Even if OpenAI cooperated with other vendors, the relevant API calls would in principle still have to run on Azure.

But on April 27 of this year, this exclusive binding finally loosened.

Microsoft and OpenAI revised their cooperation agreement and crossed out this exclusivity restriction.

From then on, OpenAI can provide products to customers through any cloud platform, and the OpenAI model and product licenses held by Microsoft have also become non-exclusive.

The lock had just been opened, and the next day OpenAI slipped into AWS.

On April 28, GPT and Codex swiftly moved into Amazon Bedrock.

AWS customers can directly call the models using their existing identity and enterprise control systems, and the related costs can also offset existing cloud spending commitments.

Now, OpenAI has stepped into Google Cloud as well.

And from Azure exclusivity to AWS to Google Cloud, OpenAI took only half a year to spread its models across the three major U.S. cloud providers.

Wherever enterprises' business is rooted in the cloud, it follows to set up shop there, with the main theme being: sell wherever you go.

At the same time, Microsoft has not been left behind by OpenAI.

Under the new April agreement, Microsoft remains OpenAI's primary cloud partner, and OpenAI products will also land on Azure first.

As expected of a first love, its position is rock solid.

It's just that Azure is no longer exclusive, and neighboring AWS and Google Cloud can finally sell the same products.

Models fight fiercely, cloud providers quietly collect the money

Everyone knows that GPT and Gemini are old rivals, fighting on the arena every day.

So why have OpenAI and Google Cloud suddenly joined hands?

In fact, both of them are clicking away at their own abacuses.

What OpenAI is after is enterprise customers.

Where data is stored, who existing contracts are signed with, and which cloud the security and audit systems run on also directly influence the final choice.

So OpenAI has started drilling in along the customer relationships that cloud providers have already built.

If an enterprise puts its business on Azure, it sells on Azure; if the business runs on AWS, it drills into Bedrock;

The data sits in Google Cloud, and it has also brought GPT and Codex into the model garden.

Customers don't need to go out of their way to OpenAI's own API platform; GPT can enter the places where customers work every day on its own.

When customers choose GPT, even though the model belongs to a competitor, Google still holds on to Marketplace, cloud infrastructure, data governance, and security services.

In fact, the more fiercely the models fight, the more enterprises want to switch back and forth based on different tasks, and the higher the value of Google Cloud as a model marketplace becomes.

As for enterprises, what they're really after comes down to two words: convenience.

No need to move data, no need to switch cloud platforms, no need to sign a separate set of vendor contracts; they can assign different tasks to Gemini, GPT, or Claude based on capability, latency, and cost.

In the past, choosing a model more or less carried a sense of "bet on one and the whole family follows";

But now, within the same cloud platform, a whole row of shelves has already been set out: use whichever works best, switch to whichever is cheaper. It really is much more convenient.

Reuters also summarized it as a broader industry trend:

Hyperscale cloud providers like Microsoft, AWS, and Google Cloud are becoming the most important commercial distribution channels for frontier model labs.

Among them, model companies are responsible for making the goods and competing on capability;

Cloud providers are responsible for holding enterprise customers, procurement entry points, and billing systems, and then taking a cut from every transaction.

△Image source: Reuters

So on the model arena, GPT and Gemini will still have to fight in the future;

But off the arena, Google Cloud has already quietly set up a shelf for GPT.

Competition is competition, business is business. These Silicon Valley folks really are better than one another at making a living (doge).

One More Thing

Amusingly, the hands that should be held have been held, but the fights that should be fought—Altman and Pichai, these two rascals, haven't held back at all.

Not long ago, Altman was still taking a jab:

Google is just bolting AI onto old products like Search; the AI-native products OpenAI built from the ground up are the real way.

In the blink of an eye, Pichai fired back hard on the same Alex Kantrowitz podcast:

The Gemini App was an AI product built from scratch to begin with.

By that logic, everyone is now stuffing Agents into products—isn't OpenAI also bolting an Agent onto ChatGPT?

The boomerang had just flown one lap, and the next day, GPT was served up on Google Cloud.

Hmm... you two just keep loving and fighting like this.

Reference link:

This article comes from the WeChat public account "Quantum Bit", author: Focus on Frontier Technology, published by 36Kr with authorization.