Tibo, OpenAI's Chief Product Officer, stated on X that new subscriptions for the $200 Pro 20X tier will be suspended to ensure a high-quality experience and seamless access to GPT-6 Astra for existing users. Tibo acknowledged having underestimated the scale of demand for Astra.
OpenAI's product Astra has triggered a rare demand shock, forcing the company to implement traffic throttling measures.
On September 11, Tibo, OpenAI's Chief Product Officer, posted on X stating that to ensure a good experience and smooth access to GPT-6 Astra for existing users, new subscriptions for the $200 Pro 20X tier would be suspended.
He stated that existing subscribers would not be affected, and all other plans and API services would remain operational as normal.
This sudden restriction also sparked mockery within the community. Some users shared screenshots of Tibo's previous tweets where he confidently asserted, "Do not worry, we have compute," to which Tibo himself responded with resignation:
Indeed, Astra should not have been underestimated.
In fact, this move came just two days after Tibo's first public warning. Wall Street News noted that on September 9, he had posted that demand for Astra was "unprecedented" and hinted that if the pressure continued, new Pro subscriptions might have to be paused. Now that the warning has materialized, the entry point for users attempting to subscribe to OpenAI's highest-tier paid plan is temporarily closed.
The leap in Astra's capabilities has shifted the "AI narrative" from a "debate over demand" back to "physical bottlenecks."
The release of GPT-6 Astra marks a substantial expansion of capability boundaries.
OpenAI explicitly describes its capabilities as: operating computers, browsing the web, using software, writing and running code, completing complex workflows, and executing long-horizon tasks in scientific, cybersecurity, and professional domains.
The leap in Astra's capabilities will fundamentally alter the logic for estimating AI demand. In the era of Answer Engines, AI demand could be approximated by "User Count × ARPU"; however, this formula is no longer valid in the era of AI Agents.
Total token demand can be roughly expressed as: Number of Agents × Tasks per Agent × Token Consumption per Task. All three variables face simultaneous upward pressure following the release of Astra.
First is the increase in the number of agents. An individual may simultaneously run multiple agents for personal assistance, programming, research, and finance, while enterprise usage could scale from dozens to thousands.
Secondly, the complexity of tasks handled by individual agents is increasing. As models become more capable of executing complex tasks, users are increasingly inclined to assign heavier and longer-duration workflows to agents.
Thirdly, agents are transitioning from "occasional invocation" to "continuous operation." Greater task complexity leads to longer reasoning chains, larger context windows, more frequent tool calls, and consequently, higher token consumption.
This implies a scenario where a 20% increase in the number of agents could drive a 300% surge in total token demand. OpenAI’s suspension of new subscriptions for its $200 Pro 20X plan aligns closely with this logic: enhanced capabilities directly amplify consumption, indicating a strong positive correlation between the two.
Wall Street News noted that a previous report by Morgan Stanley highlighted the significance of GPT-6 Astra, stating that it shifts the narrative on AI bottlenecks from "how much infrastructure is needed to meet known demand" to "how many new workloads will become economically viable as model intelligence improves."
The former represents demand-side skepticism, while the latter reflects supply-side pressure. Both have a direct impact on the pricing logic across the computing power, electricity, and semiconductor supply chains.
Editor/lambor