OpenAI has stopped accepting new subscribers to its $200-per-month ChatGPT Pro plan, blaming infrastructure strain from demand for Astra, the model it launched September 3. Product leader Thibault Sottiaux announced the pause on X, saying the Pro tier puts the heaviest load on the company's systems and that disabling signups was the least disruptive way to protect existing customers. The Plus, Go, Enterprise, Business, and API tiers remain open.
The move is unusual for OpenAI, which has historically absorbed demand spikes by adding capacity rather than closing the door. Sottiaux, who runs Codex and ChatGPT, framed the freeze as temporary but did not commit to a reopening date or disclose how many users were signing up daily.
The warning had been telegraphed. On Wednesday, before the freeze took effect, Sottiaux said the company was pulling every available lever to keep up, and floated a possible Pro pause if the surge continued.
Key facts
- 01OpenAI paused new signups to its $200-per-month Pro plan on Wednesday, citing infrastructure strain from Astra demand.
- 02Astra launched September 3 and rolled out across Pro, Plus, Enterprise, Business, and API tiers.
- 03Plus, Go, Enterprise, Business, and API plans remain open to new subscribers.
- 04OpenAI raised Codex usage limits last month, suggesting the capacity crunch is a recent phenomenon.
- 05OpenAI marketed Astra as the beginning of the 'AGI era,' language that further stoked signup demand.
Astra itself is the reason. OpenAI pitched the model as a generational step in reasoning, coding, and computer use, and marketed the release as the start of what it called the 'AGI era' — language calibrated to draw exactly the kind of signup wave that is now overwhelming the Pro tier. Astra rolled out across Pro, Plus, Enterprise, Business, and the API simultaneously, but Pro users get the largest allocation of compute-heavy features, which is why that tier is the pressure point.
The Pro plan sits at the top of OpenAI's consumer stack. At $200 per month it is priced roughly 10x the Plus tier and is aimed at developers, researchers, and heavy ChatGPT users who need the highest usage limits and earliest access to new capabilities. Those are precisely the users most likely to run Astra on long agentic tasks that consume the most GPU time per query.
“We wanted to take the smallest step that allows us to continue giving the broadest access possible”— Thibault Sottiaux, OpenAI product leader
Capacity constraints are not new for the company. OpenAI raised Codex usage limits as recently as last month, a signal that inference headroom existed at the time and that the current crunch is Astra-specific rather than a chronic shortage. Sottiaux's phrasing — that he had 'not seen anything like it' despite prior periods of steep growth — suggests the Astra curve is steeper than anything ChatGPT itself produced in 2023 or 2024.
For rivals, the freeze is a mixed signal. Anthropic, Google, and xAI have all been positioning their top-tier subscriptions against ChatGPT Pro, and a closed door at OpenAI is an obvious opening. But the underlying story — a model launch so oversubscribed that the company has to turn away $200-per-month customers — is also a demand demonstration competitors will struggle to counter with anything short of a benchmark win of their own.
The obvious skeptical read is that OpenAI is manufacturing scarcity to build hype around Astra and the 'AGI era' framing. That reading is hard to sustain: turning away subscribers at $200 a head is expensive, and the more plausible explanation is that Astra's per-query compute cost is materially higher than prior models and that OpenAI underprovisioned. The company has not disclosed either the compute profile of Astra or how long the pause will last.
The freeze reframes the AI-infrastructure conversation for the next quarter. OpenAI's Stargate buildout, Microsoft's Azure capacity, and the entire GPU supply chain have been sized against projected model-launch demand curves that Astra is apparently blowing past. If a flagship consumer tier can be forced closed by a single model release, the ceiling on how fast frontier labs can monetize their best work is set by data-center construction schedules, not by product-market fit. That is a very different constraint than the one the market has been pricing in.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
