Story
September 12, 2026
Astra Demand Is So Fierce OpenAI Has Stopped Selling ChatGPT Pro
OpenAI is framing the Pro freeze as the narrowest way to protect access for current customers, while Astra’s rapid uptake exposes how even an aggressively expanding AI infrastructure can be overwhelmed. The company is trying to ration scarce capacity without shutting off the model’s most consequential uses.
When OpenAI first released Astra last week, it kept the new model in the hands of a small group of enterprise customers while it scaled computing capacity. That caution quickly gave way to a broader paid rollout—and to demand the company now says it cannot comfortably absorb.1
The pressure point is GPT-6 Astra’s $200-a-month ChatGPT Pro tier, which offers far more usage than the cheaper Plus plan. On Thursday, OpenAI paused new Pro sign-ups and upgrades, while leaving existing subscribers untouched. Product chief Thibault Sottiaux said Pro accounts put “the most strain on our systems,” and called the freeze “the smallest step” that could preserve the broadest access as more capacity comes online.2
The decision follows a warning from Sottiaux that Astra demand was “unprecedented,” despite OpenAI having already weathered steep growth. The strain is not merely a matter of curiosity around a new chatbot: Astra’s flagship “computer use” feature can navigate websites and complete desktop tasks, making it more compute-intensive than earlier models.1
Sam Altman acknowledged the trade-off bluntly. “This would suck,” he wrote on X, but said OpenAI would prioritize “great service for customers until we can get back on top of things.”
3 It is a familiar reflex for the company: OpenAI also suspended new ChatGPT Plus subscriptions after its 2023 developer conference triggered a usage surge.1
Even amid the clampdown, OpenAI is selectively widening access. Greg Brockman highlighted free Astra Pro access for verified U.S. clinicians through ChatGPT for Clinicians.
4 The contrast captures the company’s balancing act: protect paying users, expand infrastructure at speed, and reserve room for high-stakes professional use while a breakout model tests the limits of its systems.