OpenAI Halts $200 Pro Plan Sign-Ups as Compute Strains Under Astra Surge

Skyrocketing appetite for its flagship model Astra has forced OpenAI to halt new sign-ups for its top-tier $200-per-month Pro subscription, as reported by TechCrunch. The freeze highlights how rapidly next-generation artificial intelligence workloads are testing the physical limits of cloud infrastructure.

Thibault Sottiaux, product leader at OpenAI overseeing core products including ChatGPT and Codex, announced the decision on X, noting that the Pro tier creates the heaviest burden on backend compute. By cutting off access at the top end, the company hopes to maintain reliable performance for current paying subscribers while leaving cheaper and developer-facing tiers open.

According to Sottiaux, the intake freeze is a targeted measure designed to avoid broader platform disruptions. Lower-priced tiers like Go and Plus, as well as developer access through the OpenAI API, continue to accept new users for now.

Sottiaux had signaled that restrictions might be imminent just a day earlier, describing the influx of users following Astra's arrival as unprecedented even for an organization accustomed to hypergrowth. He stated that the primary focus would remain on protecting the user experience for existing members before reopening registrations, though the company has not provided a timeline for lifting the pause or revealed daily sign-up volumes.

The surge in traffic comes just a week after OpenAI rolled out Astra on September 3, 2026. Pitched by the lab as a generational breakthrough that marks the start of the artificial general intelligence era, the system features enhanced reasoning, automated coding and autonomous computer-use capabilities across consumer and business tiers.

The Compute Bottleneck Strikes Again

For frontier AI developers, capacity bottlenecks are becoming a recurring tax on success. While expanding usage limits on platforms like Codex as recently as last month indicated growing confidence in capacity, Astra's intense reasoning workloads consume significantly more compute per query than traditional conversational models.

Until OpenAI secures additional capacity or optimizes model efficiency, capping access to heavy spenders represents an awkward compromise: turning away high-margin revenue simply to keep the servers from buckling under their own momentum.