Encrypted City 2026/9/11 Update: Earlier today, Thibault Sottiaux (Tibo), head of OpenAI’s core products and platform, announced that subscriptions to the $200 Pro plan have been paused. Because these subscription tiers place the heaviest burden on the system, the company hopes to take the smallest step possible to continue providing the broadest access.
The following is the original content:
After OpenAI’s latest flagship model, GPT-6 Astra, went live, the hype has continued to rise—potentially even starting to affect new user subscriptions to ChatGPT Pro. In a post on X, Tibo—the head of OpenAI’s core products and platform and also involved in Codex development—said that the demand brought by Astra has been truly unprecedented. Even though OpenAI has gone through multiple rapid growth phases in the past, he has never seen usage demand of this scale.
Tibo said: Astra’s demand is truly unprecedented. We are using every available method to support the demand, but even though we’ve experienced very steep growth before, I’ve never seen anything like this.
He further pointed out that OpenAI’s top priority at this stage is ensuring that existing users can maintain good service quality. However, if Astra’s demand continues at its current level, the company may have to temporarily stop new Pro subscriptions.
One week after Astra went live, OpenAI has repeatedly faced pressure on compute capacity.
Astra’s supply-and-demand problem has actually been evident since the first day of release. After OpenAI launched GPT-6 Astra on September 3, it didn’t immediately roll it out to all paying subscribers. Instead, it adopted a phased rollout. The first wave prioritized making it available to certain enterprises and security-related customers, and only later gradually expanded it to Pro, Business, Plus, and products like Codex.
Because many paying users still couldn’t use Astra after the model was released, OpenAI CEO Sam Altman once apologized for the confusion during the rollout, saying the team is expanding model access capabilities as quickly as possible.
At the time, Tibo also announced that for paid plan users, for each day they were delayed in getting access to Astra, OpenAI would issue one “banked reset,” so that the chance to reset usage credits could be stored and used later.
OpenAI’s official explanation shows that on September 3 and 4, qualifying Plus, Pro, and Business users received banked reset, reflecting that during Astra’s early launch it indeed faced more obvious capacity constraints.
Turning from “can’t use Astra” to “might not be able to buy Pro”
The main issue during Astra’s initial launch was “when existing paying users can get access to the model.” But Tibo’s latest remarks show that OpenAI’s contemplated measures have now extended to limiting new demand. If Astra usage continues to grow rapidly, OpenAI may choose to pause accepting new ChatGPT Pro subscriptions, reserving limited inference compute power for existing paid users.
This also highlights another bottleneck in the competition among frontier AI models: once model capabilities improve, what truly limits large-scale product adoption may not be the model itself, but whether the GPU, data centers, networking, and inference infrastructure can keep up with demand. From Astra’s initial rollout, which required opening access in batches and compensating with banked reset, to now when OpenAI has started discussing pausing new Pro subscriptions, the problems facing GPT-6 Astra don’t seem to be about whether anyone wants to use it. Instead, the question is whether OpenAI can provide enough compute power for everyone to use it.
Even more worth noting is that in recent years, OpenAI’s compute procurement has already reached the scale of “national-level infrastructure.” The Stargate program launched in 2025 aims to invest $50 billion over four years and build 10GW of AI data center capacity. In the same year, OpenAI and Oracle also advanced up to 4.5GW of new capacity. Overall Stargate compute under development at one point exceeded 5GW, with more than 2 million chips. By September 2025, OpenAI also said it has planned infrastructure investments of nearly 7GW and over $40 billion.
But even after Astra went live, capacity still remained tight—suggesting that the inference demand for frontier models may be growing faster than the pace of data center expansion. Recently, OpenAI has even continued looking overseas for additional compute capacity, including signing a multi-year contract with Firmus, which is backed by Nvidia investment, and adopting capacity from two data centers in Malaysia.
If this situation—where any added compute capacity is quickly consumed by model demand—continues, Astra will not only prove that the AI infrastructure boom is not over. It could further strengthen the long-term capital expenditure cycles for GPUs, HBM, data centers, power, and cooling equipment. What the market truly needs to watch won’t be just whether tech giants are willing to build data centers, but whether the revenues and demand generated by AI services can continue to absorb such massive infrastructure investment.
At present, it appears that at least the signals OpenAI has given for Astra are still relatively positive: the compute bottleneck remains a supply shortage, not a demand shortage.
This article is republished with authorization from: (Chain News)
Original title: (GPT-6 Astra demand surges! OpenAI: If it continues, it may pause new Pro subscriptions)
Original author: Neo
“GPT-6 Astra demand explodes! OpenAI pauses new Pro subscriptions: how this plan burdens the system” was first published on “Crypto City.”