Pay attention to this story because it's a perfect barometer of what's happening under the hood of the AI revolution.
Chamath Palihapitiya — the guy who became a billionaire from Facebook, crowned himself the king of SPACs, got his ass kicked by the market when the bubble popped, and now runs software companies — dropped a bomb that barely anyone is talking about: his company is ditching Cursor because the spending on AI tokens was becoming unsustainable.
Read that again: unsustainable.
Cursor Is the Darling — But Darlings Don't Come Cheap
For those not in the loop, Cursor is a code editor supercharged with artificial intelligence. It's basically a guy who codes alongside you, suggests code, completes functions, debugs errors. It's the dev's copilot. The thing is good. It's really good.
But "good" and "economically viable" are two completely different conversations.
Every time Cursor makes a suggestion, completes a line of code, or analyzes an entire file, it consumes tokens — which are the processing units of language models like GPT-4, Claude, etc. And those tokens cost money. Real money. Money that shows up on the invoice at the end of the month.
And when you scale that across an entire company of software engineers working all day long, the bill becomes a monster.
The Problem Nobody Wants to Talk About
This is where things get interesting — and where the AI circus would rather change the subject.
The dominant narrative in Silicon Valley is: "AI will replace programmers, slash costs, multiply productivity by 10x!" VCs post this on Twitter/X between sips of artisanal coffee and private jet flights. LinkedIn gurus parrot it on repeat.
But the reality Chamath just exposed is much rawer: generative AI, in its current form, has a marginal cost that is far from negligible. Every interaction costs. Every prompt costs. Every token costs.
It's like hiring a genius intern who charges by the word. At first it seems cheap. When he won't stop talking, you look at the invoice and your stomach drops.
This reminds me of a Nassim Taleb quote: "If something looks too good to be true, calculate the hidden cost." The hidden cost of the AI-in-code-editors revolution is the token bill.
What This Means for the Market
First: this is not a death certificate for AI in software development. Far from it. It's a course correction. And course corrections are healthy — they're what separates companies that survive from those that become failure case studies on podcasts about dead startups.
Second: this raises a question that should be on the mind of every investor and every tech manager: what's the real ROI of generative AI tools when you put all the costs on the table?
Because it's one thing for an individual dev to pay $20 a month for Cursor Pro and think it's the greatest thing ever. It's an entirely different thing for a company with 50, 100, 200 engineers running it all day, consuming tokens like it's water.
Third — and maybe most importantly: the companies selling "pickaxes" in this AI gold rush (OpenAI, Anthropic, Google with Gemini) are in an absurdly powerful position. They charge per token. The more people use it, the more they rake in. Their business model is basically a faucet everyone needs to turn on, and the water bill goes to you.
The Lesson Chamath Is Giving Away for Free
The billionaire didn't go crying on a podcast. He simply did the math, saw it didn't pencil out at the level they were using it, and switched tools or approaches. That's skin in the game. That's a guy with real money on the table making a real decision.
Compare that to the big-bank analyst who's never touched a line of code in his life telling you that "AI will revolutionize everything" while recommending stock in a company that's burning cash.
The question that remains is: if even Chamath — a guy with practically unlimited resources — thought the token bill was too damn high, what's going to happen when mid-sized companies try to scale these tools?
Think about that before you buy into the pretty narrative that AI is nothing but sunshine and infinite productivity. Every lunch comes with a check. And at this table, the check is arriving in tokens.
