Could This Mark the Beginning of a Crisis for Token Economies?

Microsoft recently introduced significant price adjustments to GitHub Copilot, changes so substantial that a Reddit user referred to the situation as the Tokenpocalypse within their company.

In the most recent episode of TechCrunch’s Equity podcast, Kirsten Korosec, Sean O’Kane, and I explored how these price shifts might impact the broader AI industry. As Anthropic and other major AI companies prepare to go public, questions about profitability are emerging, which could lead to similar price hikes and additional usage restrictions as businesses work to control expenses.

Sean raised a key question: “Can these AI development teams manage to reduce costs while still advancing the technology enough to meet customers’ willingness to spend?”

Kirsten pointed out that this development also highlights how rapidly trends are evolving. Within just a few months, companies became obsessed with maximizing token usage before turning against that approach due to the high associated costs. As AI firms prepare their IPO filings, she wondered, “How can they accurately document these risks when things keep changing right before our eyes?”

Continue reading for an excerpt from our discussion, edited for brevity and clarity.

Anthony Ha: When we were planning this episode, Sean, you described the current situation as the Tokenpocalypse. I’d like to hear your thoughts further on that term, especially considering Microsoft’s decision to charge more per token for GitHub Copilot instead of using a flat rate.

The entire AI ecosystem relies heavily on investment funding, meaning many services that appear cost-free are actually quite expensive. Now, as more of these costs begin to be passed on to end consumers, we’ll need to see how this affects consumer behavior. While the exact consequences remain unclear, it’s likely to cause significant discomfort for users.

Sean O’Kane: It’s interesting to consider how many token-related risks will appear in Anthropic’s S-1 filing. I’ve discussed this issue frequently on this show, as we keep encountering similar patterns — for example, Uber went from saying it had quickly exceeded its budget for AI spending within a month and a half to later deciding that costs were too high and needing to impose limits on usage across the company.

That rapid turnaround is concerning. It raises the question of whether these AI development teams can reduce costs while still advancing technology enough to align with customers’ spending capacity.

It’s also worth reflecting on ChatGPT Plus, which was initially priced at $20 per month without a clear strategy behind that amount. People have continued to pay more for higher-tier models, yet even those prices fall short of covering the true cost. That remains one of the biggest challenges in this space.

Kirsten: All of this shows just how quickly things are changing. The trend of maximizing token usage peaked within six months and is now viewed negatively. As you mentioned, the pricing structure was established before business models for AI companies had fully taken shape.

At the same time, governments are trying to keep up with these developments. This week, President Trump signed an executive order that allows the government to review powerful AI models, though it covers only a limited scope. The pace of change I’m witnessing is unprecedented.

That’s why I’m eager to see the S-1 IPO registration documents, as they will reveal important risk factors. It’s difficult to document risks when they are constantly evolving on a day-to-day basis.

Anthony: Uber serves as an interesting example, Sean. While it has faced criticism regarding its high AI spending, it also demonstrates that companies can overcome financial difficulties over time by scaling up and expanding their operations. Different business areas had to be developed, and both customers and drivers faced pressure, but these efforts ultimately helped Uber become profitable.

I believe similar transformations will be necessary for many AI companies if they want to survive in the long term.

Sean: Could these AI development teams find ways to cut costs just as dramatically as Uber did by squeezing expenses from its drivers over the years? I’m not sure. In many cases, these are more straightforward and harder-to-reduce costs, so it will be interesting to see how they handle this challenge.