You pay $200 a month for a tool you rely on. And every two weeks, the company you pay changes the rules. Welcome to the modern AI subscription, where you are not a customer—you are a hostage.
This week, Anthropic quietly slashed Claude Code’s weekly usage limits by 17%. But they didn’t announce it as a cut. They wrapped it in the kind of marketing speak that makes you want to throw your monitor out the window, claiming the change will “make it feel like you’re getting more from Claude.”
When a company tells you they are cutting your limit to make you “feel” like you’re getting more, they are telling you they think you’re stupid.
You’ve probably noticed the pattern by now. One week it’s a token limit reduction. The next week it’s a throttling update. The week after that, they’re threatening to change the entire structure of your plan. The fatigue is real, and it’s by design. You hate being jerked around by Anthropic, but where exactly are you going to go? OpenAI? Their reset policies and constant throttling are just a different shade of the same abuse.
We are trapped in an oligopoly of broken promises, paying premium prices to beta-test products that change their terms every time the wind blows in Silicon Valley.
The official excuse from every AI lab right now is “compute shortage.” They want you to believe that GPUs are just so incredibly scarce, they simply have no choice but to throttle your usage. It’s a convenient cover story for a catastrophic business model failure.
Here is the truth tech companies don’t want you to realize: they never priced AI at what it actually costs to run. They priced it at what it took to acquire you. This strategy of entering the market at a massive loss to hook users and then cranking up the prices is getting old. The “unlimited” era was never a reality; it was a venture-capital-funded illusion. Now, the VC money faucet is slowing to a trickle, and the bill for all that subsidized compute is finally coming due.
The scarce resource isn’t GPUs—it’s your trust. And these companies are burning it to the ground to preserve a growth story their investors demanded.
If you pay for any AI service right now, this is your canary in the coal mine. Pricing is heading up. Limits are heading down. The product you fell in love with will slowly degrade into a stripped-down shell of its former self, all while the marketing team insists you’re getting a better deal.
Stop falling for the spin. Stop being loyal to companies that view you as a metric in a spreadsheet. Diversify your tools, use APIs where you pay for exactly what you burn, and stop trusting the “unlimited” promises of an industry that has proven, time and time again, that it doesn’t know its own costs.
The bill is due. Don’t let them pay it with your sanity.
FAQ
Q: Isn't the compute shortage real?
A: The demand for compute is absolutely real, but the shortage is a symptom of a broken pricing model, not an unexpected disaster. They sold $100 worth of compute for $20 to buy market share. The 'shortage' is just their losses catching up to their user base.
Q: Should I cancel my AI subscriptions?
A: Not necessarily, but you need to stop treating them like reliable utilities. If a tool is critical to your workflow, look into pay-as-you-go API access where you control the spend, rather than relying on flat-rate tiers that can be devalued overnight.
Q: Isn't this just normal tech startup scaling?
A: No. Normal software scaling doesn't punish the user when it succeeds. In SaaS, more users means more profit margins. In AI, more users means exponentially more compute costs. It's a fundamentally flawed business model masquerading as a software business.