Remember when ChatGPT first launched and the thing you had to worry about was the little “at capacity” cat illustration? You’d hit refresh a few times, wait five minutes, and get back in. Annoying, but free. Nobody was out any money. The failure mode was patience.
Now the failure mode is a closed door on a $200-per-month subscription. OpenAI has paused new sign-ups for ChatGPT Pro because demand for its new Astra model is straining the system. Existing subscribers keep their access. Everyone else waits. That is a genuinely unusual situation for a software product, and as someone who spends most of his working hours evaluating AI tools for a living, I think it deserves more attention than the standard “wow, AI is popular” coverage it’s getting.
What actually happened
The short version: Astra arrived, people wanted it badly, and the compute did not stretch far enough. Thibault Sottiaux, who leads product for Codex and ChatGPT, described the pause as the least disruptive option available, saying the goal was to take the smallest step that still lets OpenAI serve the broadest set of users. Someone on the OpenAI side also noted that demand for Astra is unprecedented and that they’re pulling every lever they can to keep up, having already been through steep growth periods before.
Worth remembering that OpenAI has not exactly been passive about capacity. The company moved beyond its Microsoft Azure partnership, brought in CoreWeave, and launched Stargate, a $500 billion four-year infrastructure effort. And it still ran out of room. That’s the part I keep turning over.
Why this matters if you actually pay for these tools
Here is my honest reviewer’s take, and it’s not the flattering one. When a vendor can’t sell you a product because it doesn’t have the capacity to deliver it, you are not buying software in the traditional sense. You are buying a share of a constrained physical resource, and the vendor gets to decide how thin to slice it.
That has practical consequences for anyone building a workflow on top of these tools:
- Access is not a given. If you were planning to add Pro seats for your team next quarter, that plan is now contingent on someone else’s rack space. Budget approval doesn’t help if the checkout button is off.
- Quality can be a hidden variable. A sign-up pause is the visible, honest form of rationing. The invisible forms — shorter context handling under load, slower responses, quiet routing to lighter models — are the ones that make tool evaluations frustrating, because you can’t tell whether you’re testing the model or the traffic.
- Existing subscribers just got a moat. If you’re already paying, your subscription is temporarily more valuable than it was last week. That’s a strange thing to say about a SaaS product, and it creates a real incentive to hold a seat you’re not fully using.
The credit where it’s due
I’ll say the nice thing too, because I think it’s earned. Pausing new sign-ups is the responsible move compared to the alternatives. Taking $200 from customers you can’t serve well is worse. Silently degrading the product for everyone, including people who’ve been paying for months, is worse. Sottiaux’s framing — the smallest step that preserves access for the most people — is the correct framing, and it’s the one that protects existing customers rather than chasing new revenue.
I’ve reviewed plenty of tools that chose the other path: sell first, apologize later, blame “scaling challenges” in a blog post three weeks after the churn started. This is not that. It’s a company saying out loud that it has a ceiling.
What I’d actually do right now
If you’re on the outside looking in, my advice is to not treat this as urgency. A pause is a bad reason to panic-subscribe to something else, and it’s an especially bad reason to conclude that Astra is the only tool that could do your job. You have not evaluated it. Neither have most of the people telling you it’s essential.
If you’re already paying, this is a good moment to be honest about whether you’re using what you have. Rationed access has a way of making people feel attached to a seat rather than productive in it.
And if you’re making tooling decisions for a team, the lesson generalizes past OpenAI. Build your workflows so that a single model being unavailable is an inconvenience, not an outage. Keep a second option you’ve actually tested. Know which of your tasks genuinely need the expensive tier and which ones you’re routing there out of habit.
The demand story here is real. But the more useful story is that even a company spending half a trillion dollars on infrastructure can hit a wall. Plan your stack like that’s a normal event, because for the next while, it probably is.
🕒 Published: