We're All Drunk on Tokens (And Someone Else Is Buying)
- 28 May, 2026
I came across a tool called toktrack that charts my Claude Code token usage over time. I’m on the $200-a-month Max plan, so I’d never thought hard about it.
Then I saw the chart.
Twenty-seven days of May: $2,160.73 in token consumption. I paid $200. Somebody paid the other $1,960.
That’s not a discount. That’s a subsidy.
And companies are making permanent decisions based on the $200.
I’m more productive than I’ve ever been. That’s exactly what scares me. I’m burning tokens as fast as I can to get as much done before last call.
Someone Else Is Buying the Round
The cheap token party is going to end.
Anthropic, OpenAI, and Google aren’t running charities. The all-you-can-drink plans are land-grab pricing, built to get us hooked while the market sorts out who wins. Big Tech announced $740 billion in AI capital spending for 2026 alone, up 69% from the year before. Wall Street looks at that number and sees a financing opportunity. I look at it and see a bar tab that’s due soon.
I’m using their money to boost my productivity. And so is everyone else who feels superhuman right now. The euphoria is real, propped up by promotional pricing.
Even the people selling the compute admit it dwarfs payroll. Nvidia’s Bryan Catanzaro put it plainly: “For my team, the cost of compute is far beyond the costs of the employees.”
Ask Uber
What does the bill look like without the subsidy? Ask Uber.
Uber rolled out Claude Code (the same tool charting my $2,160) across its engineering team. Then it burned through its entire 2026 AI coding-tools budget in four months. Their CTO disclosed it to The Information in April. By then, 95% of Uber engineers were using AI tools monthly, 70% of committed code was AI-generated, and per-engineer spend ran anywhere from $500 to $2,000 a month — a 4x spread inside the same company.
Uber’s COO, Andrew Macdonald, went on the Rapid Response podcast and coined a word for it: “tokenmaxxing.” Rising token consumption with no matching rise in useful output. His exact words: “That link is not there yet, right?”
A year’s budget, gone in four months. Running the same tool I run.
Cheaper Tokens Won’t Save You
“Sure,” you’re saying, “but tokens get cheaper every month. This sorts itself out.”
That’s the first thing everyone says. It’s also wrong. Make tokens cheaper and I don’t spend less. I just run more agents. Economists call it Jevons’ paradox. To me it’s just the bill.
The numbers bear it out. Token prices have fallen roughly 280x in two years. Enterprise AI spending over the same stretch more than tripled. Cheaper per unit, more units consumed, higher total cost. Every time.
And the consumption is wildly unpredictable. I had two Claude Code sessions yesterday. One cost $45.47 and produced 2,515 lines of code. The other cost $83.33 for 3,042 lines. Same person, same tool, same day — and the second session cost 50% more per line. I have no idea why.
If I can’t predict my own bill across two sessions, no CFO is predicting it across fifty engineers.
You Can’t Un-Fire a Team
Here’s where the math actually breaks.
When you fire someone and hand their work to AI, you’re not swapping a salary for an API bill. You’re swapping a known cost for an unknown one. The salary had a ceiling. The tokens don’t.
And the salary was reversible in a way the layoff is not. You can’t un-fire a team. Token prices, on the other hand, move the moment the model companies decide the land grab is over.
GitHub already started. On June 1, Copilot shifts from flat-rate subscriptions to usage-based billing tied to actual token consumption. One developer modeled their bill jumping from about €67 a month to €966. Same usage. New price.
Microsoft saw it coming. After handing Claude Code to thousands of engineers, it started pulling the licenses back in May and pointing people at a cheaper tool. It could do that because it hadn’t fired anyone to pay for it.
The companies trading headcount for tokens don’t have that exit. They’re underwriting a permanent decision with promotional pricing. They’re calling it efficiency. It’s risk arbitrage on a price that won’t hold — and they’re on the wrong side of it.
I’m not the anti-AI guy. I run an AI-first company and I’m not slowing down.
I’m the AI-first guy who looked at the meter.
$2,160 of work for $200. Enjoy the open bar while it lasts — just don’t fire the people you’ll want around when the tab arrives.
Stay Ahead in Product Management!
Ready to elevate your product management game? Join our community of passionate professionals and be the first to receive exclusive insights, tips, and strategies directly in your inbox. Subscribe now to our newsletter and ensure you're always in the loop with the latest product management trends and wisdom bombs!