One company spent $500 million on AI in a single month. No breach. No fraudulent invoices. Their employees had no limits on how many licenses they could use, and some were running the model to check the weather. That report, covered by Fast Company in May 2026, became an instant case study in a problem most executives already suspected but hadn't yet priced: AI spend scales faster than anyone watches it.
What Enterprise AI Spend Actually Looks Like
The $500 million bill is the extreme. But the underlying behavior appeared at every large company that deployed AI broadly in 2025 and 2026.
Uber burned through its entire 2026 AI budget by April. Amazon noticed its employees were competing to be the heaviest AI token consumers and shut down the internal leaderboard that had been tracking it, because the competition was driving consumption rather than productivity. Microsoft pulled its Claude Code licenses from internal users after costs climbed beyond what the rollout plan anticipated.
The structural problem in each case was identical. AI access gets granted at the point of interest. Teams add tools on top of tools. No one with budget authority is tracking what's running, what it costs per seat, or whether the output justifies the bill. By the time the invoice arrives, the behavior is embedded and the spend is already a fixed habit.
The Same Pressure at a Smaller Scale
Most small businesses aren't at risk of a $500 million bill. But tighter margins mean the same dynamic hits harder, sooner.
A 12-person team lead adds an AI writing tool in January. By June:
- Three other team members have independently signed up for competing tools
- The original subscription has been upgraded to an enterprise tier no one approved
- One tool is billing to a card the company didn't know was on file
Total monthly cost: around $800. Nobody knows. Nobody is looking. And because the tools are technically useful, nobody asks. The ROI conversation never happens because the spend conversation never happens first.
This is the exact gap SaaS Squash was built to close. We map what you're actually running: AI tools, overlapping SaaS licenses, forgotten subscriptions. We cut what isn't earning its cost and build the visibility layer that keeps it from recurring. Not a one-time audit. A standing function on a fixed retainer that typically offsets itself in cancelled licenses within the first quarter.
What Watching the Meter Actually Looks Like
Control isn't a freeze on AI adoption. It's a policy that travels faster than the next "just try it" Slack message, with someone who has the mandate to enforce it.
The companies that avoided the Uber outcome didn't get lucky. They had a spending review process before the tools went live, not after the CFO noticed the bill. That process doesn't need to be bureaucratic. It needs to exist.
AI should be reducing your overhead. The moment it becomes the overhead, you've lost the plot. The only way back is visibility. That's the work.