OpenAI Brings Its AI Models and Codex to Google Cloud

OpenAI just made its models buyable on the one major cloud platform that didn't already sell them — and the company that used to have exclusive rights to all of this isn't even in the sentence anymore.

On October 21, OpenAI and Google announced that OpenAI's frontier models, along with its coding agent Codex, are coming to Google Cloud. Customers will be able to buy access through Google Cloud's Model Garden and Marketplace in the coming weeks, and — notably — they'll be able to pay for it using cloud spending commitments they've already made to Google, rather than opening a fresh billing relationship with OpenAI directly.

What's actually in the deal

The arrangement isn't just "OpenAI's chatbot, but on Google's servers." According to the companies, it covers:

• OpenAI's current frontier models — industry shorthand for a lab's most advanced, newest releases — available for enterprises to build on through Google Cloud's existing tooling
• Codex, OpenAI's AI coding agent, which can now be deployed inside a Google Cloud environment
• Access controls and data governance tools aimed at regulated industries — banking, healthcare, government — that need stricter oversight before they'll let any AI model near their systems
• Integration paths into Google's own agent tools, including its Antigravity development environment and Gemini Agent

That last point is the strange part. Google is, by any normal definition, OpenAI's biggest rival. Gemini and ChatGPT compete directly for the same users and the same enterprise contracts. And yet Google Cloud is now a storefront for OpenAI's products, collecting a cut every time a customer buys one.

The Microsoft chapter is officially over

For years, Microsoft's Azure was the only place you could legally run OpenAI's models at scale — a deal born out of Microsoft's multi-billion-dollar investment in OpenAI. That exclusivity ended in April, and since then OpenAI has been showing up everywhere: Amazon added OpenAI's models to Bedrock earlier this year, and now Google Cloud completes the set of all three major U.S. hyperscalers — the handful of companies big enough to rent out planet-scale computing power. It's the same multi-cloud logic that's pushed rival labs like Moonshot AI to court placing its Kimi K3 model on all three clouds too — nobody wants their model locked behind one vendor's walls anymore.

Enterprises no longer have to pick a cloud provider based on which AI lab they want to use. That choice and that provider decision have come apart — and for the companies buying this stuff, that's the whole point.

It's a sign of where the money actually is. OpenAI makes more by being available everywhere customers already have infrastructure and budget sitting, rather than forcing everyone through one door. Google Cloud, meanwhile, gets to tell enterprise customers they don't have to choose between "the cloud we trust" and "the AI model we want" — they can have both from the same vendor relationship and the same invoice.

Why this matters for Indian businesses

This isn't just a US boardroom story. Google Cloud has been investing heavily in India — more data centre capacity, in-country processing for its own Gemini models, and deep relationships with the Global Capability Centres (GCCs) that Indian arms of multinational banks, retailers and tech firms run by the thousand. Google Cloud's own leadership has called India one of its fastest-growing enterprise markets in Asia.

For an Indian bank's GCC — the in-house global offices multinationals set up here to run core technology and operations — an insurance company, or a fast-growing SaaS startup that already runs its infrastructure on Google Cloud for cost, compliance, or in-country data residency reasons, this removes a real obstacle: until now, using OpenAI's actual frontier models inside that same environment meant a separate vendor contract, separate data-handling review, and separate procurement cycle — a genuine headache for anyone also trying to stay inside India's data protection rules under the DPDP Act (the law governing how companies must handle Indians' personal data), which puts real weight on knowing exactly where customer data goes and who touches it. Buying OpenAI's models through infrastructure you've already vetted and already pay for is a meaningfully smaller lift than starting that review from scratch with a new vendor.

It also matters for cost. A lot of Indian startups and GCCs run lean on cloud spend commitments they've negotiated over years — being able to apply that existing spend to OpenAI's models instead of negotiating fresh terms is a real, practical saving, not just a convenience.

What it signals going forward

The deeper story here isn't really about OpenAI and Google patching things up. It's that the "AI model" and "the cloud it runs on" are becoming separate purchasing decisions, the same way nobody thinks twice about running Windows workloads on AWS or Linux on Azure. Once that split is normal, the competition shifts — away from who can lock in customers, toward who actually builds the best model and who runs the most reliable, cheapest infrastructure. Those are two different fights, and OpenAI just made it clear it would rather win the first one everywhere than defend a monopoly on either.