OpenAI Hits Capacity Wall as Astra Demand Forces Pause on Premium Plan
OpenAI has temporarily stopped accepting new subscriptions and upgrades to its premium $200-a-month ChatGPT Pro plan, citing a surge in demand for its newest and most advanced model, Astra, that the company says has strained its computing systems beyond expectations.
The decision, confirmed through the company’s official help documentation and a social media post from product leader Thibault Sottiaux, marks a rare instance of a major technology company voluntarily turning away high-value subscription revenue in order to protect service quality for existing customers.
“Unprecedented” Demand for a New Flagship Model
Astra, which OpenAI launched on September 3, was initially made available only to a limited set of enterprise customers before its broader rollout to paying subscribers. According to Sottiaux, the response since general availability has exceeded anything the company has previously experienced, even accounting for past periods of rapid user growth following major product launches.
In social media posts explaining the decision, Sottiaux said the ChatGPT Pro tier places the greatest strain on OpenAI’s underlying systems, making it the logical target for the smallest possible step that would still preserve broad access to the company’s technology. Existing Pro subscribers retain full access to their plans and to Astra, while the pause applies only to new sign-ups and upgrades from lower tiers.
What Remains Available
Despite the pause, OpenAI’s other subscription tiers continue to operate normally. The company’s lower-cost Plus and Go plans remain open for new sign-ups, with Astra available on those tiers at reduced usage limits, while enterprise accounts and API access for developers are similarly unaffected. The restriction is specific to what OpenAI internally refers to as the Pro 20X tier, the company’s most compute-intensive consumer offering.
The company has not provided a timeline for when new Pro sign-ups might resume, saying only that it is working to add capacity as quickly as possible. Notably, the move echoes a similar episode nearly three years earlier, when OpenAI briefly paused sign-ups for its original ChatGPT Plus tier after a surge in usage following one of its developer conferences — though this time the scale of infrastructure involved, and the sums of money being turned away, are considerably larger.
A Second, Different Kind of Pause
This is not the first time Astra’s rollout has been interrupted. The model was previously paused for several weeks in August after OpenAI determined it could not rule out the system posing a critical cybersecurity capability risk, a decision made on safety grounds rather than infrastructure limits. European regulators, including the EU’s cybersecurity agency, have reportedly been given early access to review the model alongside comparable systems from competing AI labs as that safety review continues. This week’s pause, by contrast, is explicitly framed by the company as a supply and capacity issue rather than a safety measure.
For European customers specifically, the capacity pause layers on top of an existing limitation: Astra has not yet been deployed within a dedicated European data zone, meaning organizations in the region seeking to keep sensitive data processing within EU boundaries currently have no paid option to do so through OpenAI’s own infrastructure.
Scaling Up for the Months Ahead
OpenAI has signaled that significant new computing capacity is on the way. The company has previously outlined plans that include more than 10 gigawatts of Nvidia systems, a 6-gigawatt agreement with Advanced Micro Devices for graphics processing units, and a partnership with Broadcom to deploy 10 gigawatts of custom AI accelerators, all expected to come online through the second half of 2026 and beyond.
Industry observers say the episode illustrates a broader dynamic reshaping the AI industry this year: even the best-funded companies in the sector are struggling to keep pace with surging demand for their most capable models, turning computing capacity itself into one of the tightest bottlenecks in the entire AI economy. For OpenAI, the challenge now is converting that demand into durable, well-served revenue once its infrastructure catches up rather than losing prospective high-value customers to rivals in the meantime.
