# OpenAI Pauses New ChatGPT Pro Sign-Ups Because Astra Demand Exceeds Its Compute

The 200-dollar tier places the heaviest load on OpenAI's systems, and existing subscribers keep access while the company adds capacity.

- Published: 2026-09-11T06:22:22.078Z
- Canonical: https://polylog.news/ai/2026-09-11/openai-pauses-new-chatgpt-pro-sign-ups-because-astra-demand
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [TechCrunch](https://techcrunch.com/2026/09/10/openai-puts-pro-subscriptions-on-hold-due-to-astra-demand/), [Polylog editors](https://polylog.news), [OpenAI News](https://openai.com/index/gpt-6-astra/)

OpenAI stopped accepting new subscriptions to its 200-dollar-per-month ChatGPT Pro plan. Thibault Sottiaux, the company's vice president of product and platforms, said on September 10 that the Pro tier places the heaviest load on OpenAI's systems and that the pause protects service for everyone else, [TechCrunch reported](https://techcrunch.com/2026/09/10/openai-puts-pro-subscriptions-on-hold-due-to-astra-demand/). Existing Pro accounts keep working. Other subscription tiers and the API remain open.

The cause is GPT-6 Astra, which [OpenAI shipped on September 3](https://openai.com/index/gpt-6-astra/) with a roughly 1.05 million token context window and list pricing of 10 dollars per million input tokens and 50 dollars per million output tokens, with a surcharge above 272,000 input tokens. Sottiaux said internal demand for the model has been unprecedented and that the company is working to add capacity. Russian-language AI channels [carried the same account](https://t.me/ai_machinelearning_big_data/10901), noting that OpenAI has handled traffic spikes before but not at this rate.

When a vendor says demand is overwhelming, that claim also serves the vendor's own interest, and the underlying usage data is not public. The harder-to-fake signal is the action itself. Turning off revenue at the top of its price list is expensive, and a company with spare serving capacity would not do it. Long-context reasoning models are the most capacity-intensive product a lab can sell, because a million-token session holds an enormous key-value cache in accelerator memory for the entire run.

For engineers the practical lesson is capacity risk. If a production system depends on the newest frontier model at peak hours, availability is now a scheduling problem, not just a rate-limit setting.

## What this means

Serving capacity, not model quality, currently limits how fast OpenAI can convert demand into revenue, and that constraint passes directly to accelerator, memory and data center suppliers as sustained order pressure. It also gives competitors a specific opening: teams blocked from Pro access will test Claude, Gemini or open-weight deployments, and some will not return. Watch which resolves first, added capacity or customer substitution to rivals, because that determines whether this episode shows up in OpenAI's results as deferred revenue or as a permanent loss of market share.

## What to watch

- How long the pause lasts, since a short reopening suggests a routing problem while a long one suggests a genuine shortage of serving hardware.
- Whether OpenAI introduces usage caps or tiered latency on Astra rather than reopening Pro at the same terms, which would signal that long-context inference is not profitable at current prices.
- Whether rival vendors publicly advertise available capacity for long-context workloads, which is how share shifts during a supply squeeze.