The AI Bill Is Due: Enterprises Are Still Trying to Figure Out What They’re Actually Getting for Their Money

Earlier this year, token-maxxing was the hottest trend in Silicon Valley. CEOs were practically begging employees to push AI usage as far as it would go. Use it for everything. Write emails with it. Generate slide decks. Chat with it about your cat. The message was clear: more tokens, more better.

Then the bill came due.

Uber reportedly blew through its annual AI budget in a few months. Some companies quietly cut Claude licenses for parts of their org. Meta killed its internal leaderboard. The party didn’t just slow down—it got awkward fast.

NEA’s Tiffany Luck recently spoke about this tension, and honestly, it’s refreshing to hear someone say what a lot of us have been thinking: enterprises are still figuring out their AI ROI. Not “optimizing” it. Not “scaling” it. Figuring out if it even exists.

I’ve been watching this space long enough to remember the last few hype cycles—cloud computing, big data, blockchain. The pattern is always the same. First, everyone piles in because FOMO is a hell of a drug. Then, six to eighteen months later, someone asks: “Wait, what did we actually get from this?”

AI is no different. The difference this time is the sheer scale of spend. Companies aren’t just buying a few SaaS licenses. They’re dropping millions on API calls, fine-tuning, model hosting, and the infrastructure to support it all. When Uber burns through its annual AI budget in months, that’s not a rounding error. That’s a structural problem.

What’s interesting is that the backlash isn’t coming from Luddites. It’s coming from the same people who were pushing token-maxxing six months ago. CFOs are now asking hard questions. Procurement teams are looking at usage dashboards and wondering why a single department burned through $200K in API credits generating marketing copy that no one read.

The truth is, most enterprises don’t have a clear framework for measuring AI ROI. They have vibes. They have anecdotes. “We saved time writing emails.” Great. How much time? How much did that time cost? Could you have spent that money on something else with a clearer return?

Some companies are already course-correcting. Instead of blanket licenses for everyone, they’re doing targeted rollouts. Instead of letting every team run wild with API calls, they’re setting budgets and requiring business justification. It’s less fun, but it’s more sustainable.

I don’t think this means AI is a bust. Far from it. The technology is real and it’s powerful. But the hype-to-reality gap is wider than most executives want to admit. The companies that figure out how to measure and maximize ROI—rather than just maximizing token count—will be the ones that actually benefit.

Everyone else will just get a very expensive lesson in what happens when you let engineers run the budget.