You Can't Buy ChatGPT's $200 Plan Right Now — And It's Not Scarcity Marketing
Try to hand OpenAI $200 a month for the most expensive ChatGPT plan right now and the checkout screen gives you exactly one line:
This plan is temporarily unavailable for new purchases. Existing subscriptions are unaffected.
The site isn’t broken and your card wasn’t declined. OpenAI switched off sales of its own priciest product — and judging by the timing, that switch was built for precisely this moment.
First, get clear on what actually stopped selling
The plan that got pulled is ChatGPT Pro at $200/month, which the community and press generally call “Pro 20x” because its headline pitch is 20x the usage of Plus. Strictly speaking OpenAI’s own pages just call it ChatGPT Pro, with “20x” describing the multiplier rather than serving as the product name — but since Pro now comes in $100 and $200 flavors, “Pro 20x” and “Pro 5x” are the less confusing labels.
The blast radius is narrower than you’d assume, and also more insidious:
| Action | Still possible? |
|---|---|
| New customer buys Pro $200 | No |
| Upgrade to Pro $200 from Free / Go / Plus / Pro $100 | No |
| Existing Pro $200 subscriber keeps using it | Yes, entirely unaffected |
| Downgrade from Pro $200 to something cheaper | Yes |
| Buy Plus ($20), Go ($8), or Pro $100 | Yes, all unaffected |
| Business / Enterprise, API | Yes, all unaffected |
In other words: the exit is open, the entrance is shut. That asymmetry is the single most important thing here, and we’ll come back to it.
As for how the two Pro tiers differ — per write-ups from Japan’s mihata and China’s 36Kr, Pro $200 comes with 200 weekly Astra credits versus 50 on Pro $100. Both get GPT-6 Astra; what differs is volume, not capability. That matters, because what got switched off isn’t a feature, it’s a quota.
OpenAI’s explanation: demand is “unprecedented”
The timeline is three days long.
GPT-6 Astra shipped on September 3, and OpenAI described it at the time as the start of the “AGI era.” On September 9, OpenAI’s Head of Core Products and Platform, Tibo Sottiaux (Thibault Sottiaux), posted a warning on X:
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.
He left a hint in the same breath: if this kept up, “we might have to pause new Pro subscriptions for a bit.”
By September 10 the warning had become an action:
To make sure our current users have an incredible experience and continued access to Astra, we are going to pause subscriptions to our $200 Pro plan.
One more line from Sottiaux is worth writing down, because it explains why only this one tier got cut: “We wanted to take the smallest step that allows us to continue giving the broadest access possible.” Pro puts the heaviest load on the systems, so Pro went first — Plus, Go, and the API all carried on untouched, and Astra itself is still reachable on Plus, Go, Business, Enterprise, and the API.
Why is Astra so expensive to serve? Fortune points at its “computer use” capability: Astra can interact with a desktop the way a human would, filling out forms and navigating web pages at “superhuman speed” — and that lets it burn through users’ usage allowances far faster than the previous flagship, GPT-5.6 Sol. Sam Altman himself called the Astra launch “messy.”
Is this scarcity marketing? Altman answered that nearly two years ago
Limited availability, can’t-get-it, official statements about overwhelming demand — the combination does look an awful lot like manufactured scarcity. But three pieces of evidence point the other way.
One: this plan was already losing money, and Altman said so himself. On January 5, 2025, he wrote on X:
Insane thing: We are currently losing money on OpenAI Pro subscriptions!
His stated reason was that “people use it much more than we expected,” and he noted he had set the price himself expecting it to be profitable. Note that Pro was already $200 back then. So for the 20 months before the pause, the company had publicly conceded that the price doesn’t cover real usage. A product that loses money isn’t a loss to the company when it stops selling.
Two: unpack the quotas and the $200 tier is the cheapest compute on the menu. 36Kr’s analysis makes a counterintuitive point: relative to Pro $100, Pro $200 means “paying twice as much for four times the quota” (200 credits versus 50), which makes it the cheapest unit of compute in the entire lineup. That analysis further estimates Pro 20x goes gross-margin-negative at just 5.7% utilization, against 11.4% for Plus and Pro 5x — and to be clear, that’s 36Kr’s estimate, not a financial figure OpenAI disclosed. But it points the same direction Altman did: OpenAI assumed most people wouldn’t burn through their quota, and heavy users went ahead and burned through it.
Three: this isn’t the first time. Computerworld quotes Kanerika CRO Bhupendra Chopra saying this is “at least the third time in two years OpenAI has throttled or paused access at launch,” reflecting a structural problem — “model capability keeps outpacing data center buildout.”
Following that thread into capital spending is illuminating. Per Fortune, OpenAI’s available compute grew from 0.2 GW in 2023 to roughly 1.9 GW in 2025 — 9.5x in two years — with 10 GW of Nvidia systems, a 6 GW AMD agreement, and 10 GW of custom Broadcom accelerators lined up for the second half of 2026. Compute went up nearly tenfold and it still isn’t enough to sell. That’s a considerably worse look than pretending not to have enough.
The uncomfortable detail: this is a switch that can be flipped remotely
That said, it would be wrong to call the whole thing above reproach.
The X account @btibor91 (Tibor Blaho), which tracks strings in the ChatGPT web app’s code, pointed out something specific: purchasing of Pro 20x in the web app “can now be paused remotely,” and that line — “This plan is temporarily unavailable for new purchases” — was already sitting in the front end, written and ready.
The distinction matters. It means the pause wasn’t engineers yanking cables at midnight; it was a pre-designed feature flag that can be toggled at will. You can read that as mature capacity management, or as OpenAI having already concluded that this will keep happening and productized it accordingly. Both readings hold, and they aren’t mutually exclusive.
Chopra’s comment lands harder in that light: “It signals that frontier capacity is still rationed. Consumer power users are the release valve. Enterprise contracts are what the vendor protects.” Who gets cut off first tells you the priority order. And priority order is a business decision, not a law of physics.
His three questions for enterprise buyers are worth copying down too: What throughput is contractually committed? What happens to your workloads during demand spikes? How quickly can you fail over to an alternate model? His conclusion is blunt — production-critical workloads belong on an enterprise agreement with committed capacity, and “it should have a second model tested and ready.”
It happened in 2023, and that time it lasted about a month
The closest thing to a forecast for how long this lasts is 2023.
OpenAI held its first DevDay on November 6, 2023 and launched GPTs. Nine days later, on November 15, Altman announced a pause on new ChatGPT Plus subscriptions:
The surge in usage post devday has exceeded our capacity and we want to make sure everyone has a great experience.
Sales didn’t reopen until December 13, when Altman posted “thanks for your patience while we found more gpus” — 28 days end to end.
This time OpenAI has given no restoration date at all, saying only that it’s adding capacity as fast as it can. Worth noting: what got paused in 2023 was the $20 entry plan, the highest-volume tier by far. What got paused this time is the $200 top tier — the lowest volume, but the most brutal consumption per user. The revenue pressure in the two cases is completely different, so “28 days last time” is useful as an order of magnitude and worthless as a promise.
That same week, the other $200 “20x” plan ended up in court
Here’s the real coincidence: two days before OpenAI stopped selling Pro 20x, on September 8, a different $200 “20x” plan got sued.
A group of Claude users filed a class action against Anthropic alleging misleading marketing of its Max subscription. Max has been $100 (5x) and $200 (20x) since launching in April 2025, sold on the promise of 5x or 20x more usage than Pro. Per Engadget, the suit cites a viral post arguing that the $200 20x plan claims four times the capacity of the $100 5x plan, but once weekly caps are factored in it works out to only about 1.7x. Plaintiffs’ attorney Monica Vaca put it this way: “This is hard for consumers — they don’t know what’s in the black box.” Anthropic had not responded as of publication.
Put the two events side by side: one week, two companies, two $200 “20x” plans — one pulled from sale, one in litigation. The shared pathology isn’t that one company is unusually bad. It’s the practice of pricing by multiplier: what gets promised is a relative multiple, while what gets delivered is constrained by absolute caps, weekly quotas, and real-time compute supply all at once. As models get stronger and users get better at using them, that promise stops being deliverable — OpenAI chose to close the entrance, Anthropic chose to hold the caps, and both paid for it.
One Hacker News commenter summed it up cleanly: these plans offered “subsidized usage rates that were a really great value.” Great enough to be unsustainable, which is exactly where we are.
If you’re on it, or wanted to buy it, don’t do these things right now
Back to that asymmetry: the exit is open, the entrance is shut. It sets a trap that’s very easy to walk into.
The key risk, flagged in mihata’s write-up: if you’re a Pro $200 subscriber today and you cancel, then once the billing period ends and the subscription actually terminates, you cannot buy back in until OpenAI lifts the pause — and nobody knows when that is. Your safe window is the current billing cycle, and only that.
So the moves to avoid are concrete:
- Don’t cancel just because you won’t use it much this month
- Don’t schedule a downgrade (downgrades are permitted, which is precisely the problem — you can go down but you can’t come back up)
- Don’t let the card on file expire
- Don’t migrate or switch accounts right now
If you’re on the wanted-to-buy side, there are still paths. Pro $100 is completely unaffected, with 50 weekly Astra credits — same capability, less volume. Plus ($20) and Go ($8) are both still purchasable and still get you Astra, just with tighter limits. And if your needs are genuinely heavy, the API bills per token and has not been paused at any point.
One last observation worth holding onto: in this episode, OpenAI protected existing subscribers and enterprise contracts, and sacrificed incremental consumer power users. If your work genuinely runs on these models, then Chopra’s advice isn’t only for enterprises — making sure you have a second usable model in hand beats betting on when OpenAI reopens the till.
