Cheaper tokens haven't solved the problem—they've made it worse
Inside the world's largest technology companies, a quiet paradox is unfolding: the price of artificial intelligence keeps falling, yet the bills keep rising. As meter-based pricing made AI costs newly visible, organizations discovered that cheaper tokens had not reduced spending—they had simply unlocked more uses, spreading AI across every corner of the enterprise until the monthly totals climbed into the billions. What began as a productivity revolution is now a budget crisis, and the companies that led the AI charge are quietly telling their employees to slow down.
- Tech giants are receiving billion-dollar AI bills even as the per-token cost of computation continues to drop—cheaper prices have multiplied usage, not reduced it.
- Meta has begun actively discouraging employees from using AI tools, and other major firms are following quietly, reversing the frantic adoption that defined 2024 and early 2025.
- Engineers are building internal systems to flag expensive queries before they run, rationing a technology that was supposed to make them more productive.
- The meter pricing model meant to create efficiency has instead revealed that AI economics at scale are fundamentally broken—proliferation outpaces every unit-cost saving.
- Companies are now exploring flat fees, usage caps, and in-house model development in a scramble to restore budget predictability before the next invoice arrives.
Something unexpected is happening inside the world's largest technology companies. The price of artificial intelligence keeps dropping—a task that cost ten dollars six months ago might cost a dollar today—yet companies are spending more on AI than ever before, and their employees are starting to push back.
When Meta, Google, OpenAI, and others adopted meter-based pricing that tracks every token in every query, they gave organizations unprecedented visibility into their AI spending. What those organizations found was sobering: bills climbing into the billions. Engineers began rationing API calls. Some stopped experimenting altogether. Others built tools to catch expensive queries before they ran. The technology meant to multiply productivity had become a budget line item demanding constant vigilance.
The deeper irony is structural. Cheaper tokens unlocked new use cases rather than reducing costs. What began as a specialized tool for a handful of teams spread across entire organizations—customer service bots, documentation systems, code generation, content creation—each individually cheaper, but collectively staggering when the monthly bill arrived. Meta has been explicit, implementing policies to curb employee consumption even as per-unit costs fall. The New York Times reported that tech workers have essentially maxed out their AI usage and are now actively minimizing it—a sharp reversal from the adoption frenzy of recent years.
The road ahead is uncertain. Some companies are exploring flat fees, usage caps, or tiered access to restore budget predictability. Others are investing in proprietary models to escape the meter entirely. But economists warn the dynamic will only intensify: the cheaper AI becomes, the more it proliferates, and the higher total spending climbs. Either companies find ways to genuinely reduce consumption, or they accept AI as a permanent, massive operating cost. The meter keeps running either way.
Something unexpected is happening inside the world's largest technology companies. The price of artificial intelligence—measured in tokens, the tiny units that make up every prompt and response—keeps dropping. By almost any economic logic, cheaper inputs should mean lower bills. Instead, companies are spending more money on AI than ever before, and their employees are starting to fight back.
The paradox is straightforward enough to state but harder to live with. When Meta, Google, OpenAI, and others began charging by the token—a meter-based pricing model that tracks exactly how much computational work each query demands—it created unprecedented visibility into AI spending. What companies discovered was sobering. The bills were climbing into the billions. Engineers and product teams, suddenly aware of the cost attached to every API call, began rationing their use. Some stopped experimenting with AI altogether. Others built internal tools to flag expensive queries before they ran. The technology that was supposed to be a productivity multiplier had become a budget line item that demanded constant vigilance.
The irony deepens when you look at the numbers. Token prices have fallen steadily as competition intensified and models became more efficient. A task that cost ten dollars six months ago might cost a dollar today. But companies aren't celebrating. Instead, they're spending ten times as much overall because the cheaper price has unlocked new use cases. What started as a specialized tool for a handful of teams has spread across entire organizations. Customer service bots, internal documentation systems, code generation, content creation—each one individually cheaper than before, but collectively adding up to something that catches executives off guard when the monthly bill arrives.
Meta has been explicit about the problem. The company has begun actively discouraging employees from using AI tools, implementing policies designed to curb consumption even as the per-unit cost continues to fall. Other major tech firms are following similar paths, though often quietly. The New York Times reported that tech workers have essentially maxed out their AI usage and are now trying to minimize it—a reversal of the frantic adoption that characterized 2024 and early 2025. The shift reflects a hard truth: abundance creates its own discipline. When something is free or nearly free, you use it without thinking. When you can see the meter running, you start to ask whether you really need it.
What's happening inside these companies hints at a larger reckoning ahead. The meter pricing model was supposed to create efficiency by making costs transparent. Instead, it's revealed that the economics of AI at scale don't work the way anyone expected. Cheaper tokens haven't solved the problem of unsustainable spending—they've made it worse by encouraging proliferation. An economist quoted in recent reporting warned that this dynamic will only intensify as prices fall further. The cheaper AI becomes, the more companies will use it, and the higher their total bills will climb.
The question now is whether this pattern can hold. Companies are clearly uncomfortable with the trajectory. Some are exploring alternative pricing models—flat fees, usage caps, tiered access—anything to restore predictability to their budgets. Others are doubling down on internal AI development, hoping to escape the meter entirely by building their own models and running them on their own hardware. What seems certain is that the current model, where token prices fall while overall spending rises, cannot continue indefinitely. Something will have to give: either companies will find ways to genuinely reduce their AI consumption, or they'll accept that AI has become a permanent, massive line item in their operating costs. The meter keeps running either way.
Citações Notáveis
Tokens are getting cheaper, but companies are spending even more on AI as a result— Top economist quoted in reporting
A Conversa do Hearth Outra perspectiva sobre a história
Why would companies keep spending more if the price per token is falling? Shouldn't that be good news?
It would be, except that cheaper prices make it rational to use AI in more places. You start using it for things you wouldn't have paid for before. The savings get swallowed by expansion.
So it's like how cheaper electricity led to more electricity use, not less?
Exactly. Except electricity companies could plan for that. AI companies didn't anticipate how fast the usage would spread once the price dropped.
And now the employees are pushing back?
Yes. Once people could see the meter running, they realized they were being wasteful. It's a form of awareness that actually constrains behavior.
Does that mean meter pricing was a mistake?
Not a mistake—just revealing. It showed that the real problem isn't the unit cost. It's that companies don't know how to budget for a technology that's simultaneously getting cheaper and more essential.
What happens next?
Either companies find ways to genuinely limit usage, or they accept that AI is now a permanent, massive expense. There's no middle ground where prices keep falling and spending stays flat.