Tokenmaxxing Is Upon Us

The AI cost savings are already gone.

Word on the internet streets, an unnamed company spent $500 million on Claude in a single month. Big nu… you get the point. How? They didn't set any limits, so employees used it for everything. This isn’t unique, but it’s an extreme case.

Honest Numbers

Google now processes over 3.2 quadrillion AI tokens per month, which is roughly seven times more than a year ago (Google I/O 2026).

Token prices have risen about 65% since late February. US AI software prices have increased 20-37% since last year (Source: The Tropic report).

Only 18% of token spending on advanced AI coding tools translates into finished software that reaches real users. The rest goes to testing, debugging, reviewing, and correcting AI-generated work (Source: Entelligence).

80% of companies miss their AI spending forecasts by 25% or more (Source: Mavvrik).

Tokenmaxxing: Explained Nicely

Here's how we got here. Companies went all in on AI. It was cool to have AI somewhere in your name, or right next to it. It still is, but not like it was a couple of years ago. So the AI was in, but it couldn't just sit there collecting dust. To justify costs, employees have to use said AI.

So they started gaming the system. AI use makes you look productive, so why not?

At Amazon: "Kirorank" (an internal leaderboard) ranked employees by AI usage, workers had AI agents perform pointless tasks just to climb the rankings, compute costs hit the roof. Amazon shut it down, fast.

At Meta: An unofficial leaderboard called "Claudeonomics" tracked token use. The top user burnt 281-328 billion tokens in 30 days. That's around $2 million worth. The leaderboard was removed after it leaked.

At Microsoft: They canceled most of their direct Claude Code licenses and shifted devs to cheaper internal tools.

Layoffs Don't Cover the AI Bill

AI has taken the blame for layoffs for years now, but the numbers for both are miles apart. Meta, for instance, is laying off 8,000 people. That will save around $2−3 billion annually, but Meta's AI capital spending for 2026 is ~$125−145 billion annually, and that's just on data centers alone.

Amazon? Repeat layoff offender. They cut roughly 30,000 roles in the last five months. Estimated annual savings is roughly $4.5–6 billion. Amazon's AI capital spend is ~$200 billion annually, so even at the high end, the layoffs cover just 3% of the AI bill.

The combined AI capital spending of Amazon, Meta, Microsoft, and Google will reach $725 billion in 2026 (source: TechRound). It's up 77% from 2025. The cuts aren't funding AI, they're a side effect of where the money is going.

The Response?

They're rationing.

Uber, Meta, Microsoft, Salesforce, and DoorDash are placing limits on AI usage. Amazon isn't tracking token consumption anymore, now they're checking if devs actually produce code that ships. Companies are also passing costs to consumers, as high-end AI features are increasingly being moved to paid plans.

So?

Big Tech went from "use as much AI as possible" to "justify every dollar spent" in less than two years. Tokenmaxxing is upon us.

Well… them. This is the beginning of the end of free AI. Enjoy.