Pranit GargBuilder · GTM · Nomad
Anthropic Pulls the Oldest Trick in SaaS Pricing

Anthropic Pulls the Oldest Trick in SaaS Pricing

March 17, 2026·5 min read

I pay $200/month for Claude Max. Last week, my limits got noticeably worse. Then Anthropic announced a 2x off-peak usage promo.

Sounds generous. Here's what actually happened.

The Playbook

The mechanics are straightforward once you see them.

Step 1: Quietly reduce the baseline limits. No announcement. No changelog. Usage just feels worse one week.

Step 2: Announce a "generous" 2x off-peak usage promo. The framing is pure goodwill.

Step 3: The 2x absorbs the reduction. The new lower limit with a 2x multiplier feels roughly like what you had before. You don't notice the downgrade because the promo masks it.

Step 4: The promo ends. You're left at a baseline lower than where you started. But there's no reference point to compare against, so it just becomes the new normal.

The Pricing Page Tells the Story

This is the smoking gun.

Go look at Anthropic's pricing page right now. The API side is crystal clear: $5 per million input tokens, $25 per million output tokens. You know exactly what you're paying for and exactly what you get.

Now look at the consumer plans. "More usage." "5/20x more usage." More than what? They never say.

It would be trivially easy to put a token count on that page. They do it for the API. They choose not to for subscriptions.

That's not an oversight. That's a strategy.

Undefined limits mean unlimited flexibility to quietly adjust the ceiling downward. And we'd never know, because there was never a number to compare against.

You can't accuse someone of moving the goalposts when they never told you where the goalposts were. That's the whole point.

The 135x Subsidy

Here's the math that makes all of this make sense.

One user documented $27,000 in API-equivalent compute consumed in 23 days on a $200/month plan. That's a 135x subsidy. Anthropic loses money on every heavy Max user.

They burn 70 cents of every dollar they bring in. Inference costs came in 23% higher than their own projections. Breakeven isn't expected until 2027 or 2028.

We've seen this exact pattern before. Uber subsidized both sides of the market for years. Riders got cheap rides. Drivers got bonus payouts. Everyone thought "this is just how it works now." Then the fee switch turned on.

We are living through the most subsidized era in the history of software and most people are sleeping on it.

The weekly rate limits that started in August 2025 were the first signal. Every "generous" promo since then has been the subsidy quietly correcting.

The Play

This isn't a reason to panic. It's a reason to move faster.

Max out your Claude subscription while it's subsidized. Build with every tool available. Ship fast.

$200/month for what costs $27,000 in raw compute won't be a thing in 2027.

Enjoy the party while it lasts.