AI for HumanIT
All posts
30 Sept 2026 · 5 min read

OpenAI's Dots and the awkward question for Azure and Google Cloud

OpenAI launched Dots at DevDay 2026, bundling compute for computer-using agents into its subscriptions. What that might mean for the cloud providers underneath.

AgentsOpenAICloudProducts
A large split violet and coral sphere hovers above standard cloud server racks, absorbing small data dots to represent a unified compute layer.

OpenAI announced Dots yesterday at DevDay 2026. You can read the launch post and the wider DevDay recap for the full detail. I want to focus on one part of it, because I think it matters more than it first appears.

In short, a Dot takes care of the compute, and to some extent the apps, that a computer-using agent needs to do its job. That is the part most of us have been bodging together ourselves.

What a Dot changes for people building agents

If you've tried to get an agent to use a computer, clicking around, filling in forms, moving between applications, you'll know the model is only half the problem. The other half is where it runs. It needs a machine, an environment, the right software installed, and something that keeps it all going while the agent works.

Until now that has mostly been your problem. You spin up a virtual machine with a cloud provider, set it up, secure it, pay for it by the hour, and hope the agent doesn't leave it in a mess.

Dots moves a lot of that inside OpenAI. The environment and some of the apps come with it. For anyone building agents, that removes a real layer of friction. Fewer moving parts means fewer things to break, and fewer bills from different places to reconcile at the end of the month.

The pricing is the clever bit

The first Dot comes free with a Pro or Business Premium subscription. You pay for token usage, not for the machine.

That is a smart move. It makes the compute feel like it costs nothing. Anyone already on one of those plans can try a computer-using agent without first working out a separate hosting bill. The cost you do see, tokens, is the one people are already used to.

It also changes who you're buying from. If the environment comes bundled with the subscription, the obvious next step is not "which cloud should I host this on?" but "do I need a second Dot?" The buying decision stays with OpenAI.

I'd be surprised if that was an accident.

Where this leaves Azure and Google Cloud

This is the question I keep coming back to. What does this mean for the big cloud providers?

My assumption is that Dots are running on someone else's infrastructure today. OpenAI's long relationship with Microsoft and Azure is well known, and it has spread its compute across other providers too. I haven't seen anything that says where Dots actually run, so treat that as my guess rather than a fact.

Even so, the direction is worth thinking about. Before Dots, a business building computer-using agents might have used OpenAI for the model and Azure or Google Cloud for the machines. The cloud provider had a direct relationship with that customer. It saw the usage, sent the invoice and could sell other services alongside.

With Dots, that relationship moves up a level. The customer deals with OpenAI. The cloud provider, if it's involved at all, sits behind OpenAI as a wholesale supplier. That is still business, but it's a thinner and less comfortable kind. You're selling capacity to one very large customer instead of services to thousands of smaller ones.

Then there's the second step. It wouldn't take much for OpenAI to host this itself. It has been open about wanting its own data centre capacity, and a product like Dots gives it a clear reason to use it. Once the customer relationship, the billing and the environment all sit with OpenAI, where the servers are is an internal decision. Customers probably wouldn't notice if the hardware moved.

That's the uncomfortable part for Azure and Google Cloud. The more of the stack OpenAI owns, the less the cloud providers underneath look like partners and the more they look like suppliers. Suppliers can be replaced.

In fairness, both Microsoft and Google have their own models and agent products, so they aren't standing still. They can bundle in the same way, and they already own the enterprise relationships many businesses rely on. But it's notable that a model company is now packaging the compute layer, not just renting it.

What I'll be watching

I haven't built anything on Dots yet, so this is about the announcement, not about how it holds up in practice. A few things I want to find out:

  • How far the apps go. "To some extent" is doing a lot of work. Whether a Dot can run the software a real business uses, or only a narrow set, will decide how useful it is.
  • What token usage looks like in practice. Computer-using agents can be chatty. A free environment isn't much of a saving if the tokens add up quickly.
  • What happens after the first Dot. The free one gets you started. The price of the second one tells you what the product really costs.
  • Where the data sits. For anyone with governance or compliance duties, knowing where the environment runs and who can see inside it isn't optional.

I'll try it properly and report back on what held up. For now, my view is this: Dots make computer-using agents easier to start with, and they also quietly change who owns the customer. Keep an eye on both.

The more of the stack it owns, the less the cloud providers underneath look like partners and the more they look like suppliers.
Bruce Cullen
Keep reading

More from the blog