GitHub Copilot has moved away from a flat monthly subscription and toward billing based on token usage, a change that quickly drew criticism from developers. On Reddit, users began calling it “Tokenpocalypse,” a label that captures a wider concern: this is not just a pricing tweak for one product, but a sign that AI companies are changing how they charge for access.
In TechCrunch’s Equity podcast, Anthony Ha, Sean O’Kane, and Kirsten Korosec framed the change as part of a broader shift across the AI industry. Their argument was blunt. For a long time, many AI services were kept cheap by investor subsidies, while the actual cost of training and inference remained high. Now those costs are showing up more clearly in what users and enterprise customers are asked to pay.
The $20 monthly benchmark is under pressure
Sean O’Kane pointed to an early pricing decision that shaped the market: when ChatGPT launched in late 2022, OpenAI’s $20 monthly fee became a benchmark for the sector. He said that number did not reflect a deep pricing strategy at the time, yet it anchored expectations across the industry. As model training and inference costs keep rising, that gap between what AI actually costs and what users expect to pay has become harder to ignore.
GitHub Copilot’s token-based model is being read as one of the clearest signs that the old pricing structure is breaking down. For developers, the appeal of a predictable “all-you-can-use” plan is being replaced by metered consumption. That is why the reaction has been so strong.
Uber’s budget squeeze puts enterprise usage in focus
One of the sharpest examples discussed on the podcast came from Uber. According to Kirsten Korosec, the company burned through its annual AI budget in less than four months, forcing management to set usage caps for employees. She also described a rapid cycle that unfolded in roughly one and a half months: AI tools were introduced, use surged, spending overshot, and limits had to be imposed.
That experience also changed the tone around “tokenmaxxing,” a term used for pushing token consumption as high as possible to extract maximum output from models. What had been celebrated by parts of the developer community only six months earlier is now being treated as wasteful. Uber’s co-founder and COO said internally that token usage does not scale in direct proportion to useful output, a point that cuts to the center of enterprise spending discipline.
Anthropic’s IPO filing may face token-cost questions
The pricing debate is also reaching public markets. The report says Anthropic has confidentially submitted an S-1 filing and is nearing a $1 trillion valuation. Sean O’Kane raised a pointed question: how many token-related risk factors will appear in that filing?
Kirsten Korosec said those risks are changing in real time, which makes disclosure harder for AI companies whose business models, pricing plans, and cost structures are still moving quickly. The report also noted that President Trump signed a narrower artificial intelligence executive order this week, requiring the government to have opportunities to review powerful AI models. Cost pressure, regulation, and IPO disclosures are now colliding at once.
From subsidy-fueled growth to measured returns
Anthony Ha compared the current moment in AI to Uber’s earlier years, when the ride-hailing company was widely seen as unprofitable until it adjusted its operations, entered new markets, and changed pricing on both sides of its platform. The harder question for AI companies is whether they can make a similar transition when computing costs are far less flexible.
That is why the “Tokenpocalypse” label has spread beyond a Reddit joke. It points to a larger reset in how AI products are priced, how usage is evaluated, and how long companies can keep absorbing costs before passing them on.