Microsoft just rewrote the pricing on GitHub Copilot, and the change was steep enough that one company started calling it the “Tokenpocalypse.” According to TechCrunch AI, the team behind the Equity podcast (Anthony Ha, Kirsten Korosec, and Sean O’Kane) dug into what that shift signals for the rest of the AI ecosystem. Their read: the era of cheap, investor-subsidized AI is ending, and the bill is about to land on customers.
This is significant because it marks the moment the AI business model stops being theoretical and starts being painful.
What’s Actually Changing
For two years, AI tools have felt almost free. They weren’t. As Anthony Ha put it on the podcast, the whole ecosystem is “heavily, heavily subsidized by investor money,” so things that seem to have no cost are in fact incredibly expensive. Microsoft moving Copilot toward per-token charging instead of a flat rate is the first big crack in that wall.
The backdrop matters too. Anthropic and other labs are preparing to go public, which forces an awkward conversation about profitability. Once you file an S-1, you have to explain how you make money, not just how fast you grow.
The Uber Comparison, and Why It’s Not a Clean One
The optimists have a ready answer: Uber looked wildly unprofitable for years, then hit scale and closed the gap. TechCrunch AI reports the panel pushed back on that comfort.
Here’s the catch. Uber only got profitable by transforming itself. It expanded into new business lines and, as Ha noted, squeezed both drivers and riders along the way. O’Kane’s question is the sharp one: can AI labs find anything “squishy enough” to squeeze, or are these just hard, straightforward compute costs that don’t bend? Right now it looks like the latter.
Korosec’s framing is the one to sit with. The whole “tokenmaxxxing” craze, where companies raced to throw more tokens at every problem, became a trend, peaked, and fell out of favor inside roughly six months. Even Uber, a heavy AI user, blew through its budget, then slapped caps on internal usage. That’s a fast round trip from enthusiasm to rationing.
Why This Matters Now
Three forces are colliding at once:
- Pricing was never designed. The original $20 ChatGPT Plus tag wasn’t strategy. O’Kane described it as “let’s spit out a number.” The whole industry has been reckoning with that guess ever since, and even premium tiers don’t close the gap to true cost.
- IPO pressure is real. Public filings demand honest risk factors. Korosec’s open question: how do you even write those risks when they evolve day by day?
- Regulators are circling. The same week, President Trump signed a narrow executive order giving the government a path to review powerful AI models. Pricing chaos and policy scrutiny are arriving together.
The Next 1 to 3 Years
Expect the subsidy to keep thinning. More products will shift from flat fees to metered, per-token billing. More companies will impose internal usage caps, the way Uber did. And the gap between what AI costs to run and what customers will happily pay becomes the defining business question of the sector.
The labs are betting they can collapse costs through better models and cheaper inference faster than customers lose patience. That race, cost curve dropping versus price tolerance shrinking, decides who survives.
What to Do About It
If you build on AI or buy it for your business, start preparing now:
- Audit your token spend. Know exactly what your AI usage costs today, before a pricing change makes it a surprise.
- Model the per-token future. Re-run your budgets assuming flat-rate deals disappear. If the math breaks, you have a problem to solve early.
- Build switching optionality. Avoid deep lock-in to one provider’s pricing. Keep your prompts and workflows portable.
- Tie usage to value. Reserve heavy token spend for tasks with clear ROI. The “throw tokens at everything” phase is over.
The cheap-AI honeymoon is closing. The companies that treat tokens like a real budget line, starting today, will be the ones still standing when the subsidies dry up. You can find the full conversation at the original source.