ChatGPT: OpenAI is falling behind and suspends Pro subscriptions

You may have followed the launch of GPT-6 Astra on September 3 with the same excitement as OpenAI executives, who saw it as “the beginning of the AGI era.” Seven days later, the excitement changed sides. On September 10, Thibault Sottiaux, head of product for ChatGPT and Codex, announced on X the suspension of new subscriptions to the Pro plan, priced at $200 per month (€229 including VAT in France).The justification comes down to one sentence from the product lead: « Demand for Astra is really unprecedented ».

Demand for Astra is really unprecedented. We’re pulling all the levers possible to sustain the demand, but I’ve not seen anything like it until now and we went through very steep growth before. Priority will always be to keep excellent service for existing users, but we might…

— Tibo (@thsottiaux) September 9, 2026

Who is affected by the freeze on Pro subscriptions?

Current Pro subscribers retain access to GPT-6 Astra, with no additional restrictions, and the Go, Plus, Business and Enterprise plans remain open, as does the API for developers. Only newcomers are turned away, with no reopening date (OpenAI says it is adding capacity, which does not really constitute a timetable). Sottiaux presents the measure as « the smallest possible step » to continue serving the greatest number of people: the door is being closed so as not to degrade the room.

The model in question was trained on more than 100,000 GPUs at the Stargate data center in Texas, and can operate a computer (spreadsheets, forms, browsing) at a speed described as superhuman. It reached Plus and Pro subscribers in the days following its announcement, and the rush did the rest. As early as September 9, Sottiaux warned that interest exceeded that of OpenAI’s previous launches, which had already been explosive; the next day, the window closed.

In November 2023, following its first developer conference, OpenAI had already paused sign-ups for ChatGPT Plus, overwhelmed by the influx, and it took a month to reopen them (a déjà vu, in other words). Last July, Chinese company Moonshot AI had the same misadventure with Kimi K3: subscriptions were frozen 48 hours after the launch, then reopened in batches as servers were delivered. Scarcity does not take sides.

Why OpenAI prefers shutting the door to slowing down

Closing access to a service that loses money on every customer may seem like a sensible decision. Sam Altman had acknowledged this as early as January 2025: Pro subscribers used the service so heavily that OpenAI was losing money on each of them, and nothing suggests that the equation has reversed with a far more resource-intensive model. In return, OpenAI pays for itself in other ways. With consumer plans, including Free, Plus and Pro, conversations are used by default to train future models unless the subscriber disables the option in their settings (Business, Enterprise and API accounts are excluded). From this perspective, every session by a heavy Astra user is worth at least as much in data as in dollars, and the below-cost pricing looks less like a calculation error than an investment.

Last spring, Anthropic made the opposite choice: letting everyone in while tightening Claude’s usage limits, at the cost of a month of backlash from Claude Code users in March and April.In late April, the company acknowledged three engineering missteps, including a reduction in default reasoning effort, introduced to cut response times, while pledging never to deliberately degrade its models (cautiously worded). On May 6, it signed an agreement with SpaceX, which had meanwhile become the home of xAI, for more than 300 megawatts and 220,000 Nvidia GPUs, and immediately doubled its limits. In short, two strategies for the same shortage: rationing usage or rationing access. OpenAI chose the latter, as Moonshot had before it, and as OpenAI itself had done three years ago.

👉🏻 Follow tech news in real time: add 01net to your sources on Google, and subscribe to our WhatsApp channel.