Companies rushed to adopt AI tools last year. At first, expenses seemed small. A few pilot projects, some ChatGPT seats, and trial runs. Now, the bills have exploded.

Uber spent its entire 2026 AI budget by April, barely four months in. Five thousand engineers used AI coding assistants heavily. Some paid $500 to $2,000 each per month just for tokens—the units that measure AI usage. That added up to billions in costs. The technology worked well, generating most of the code in some teams. But the price tag shocked finance departments.

This isn’t unique to Uber. Other tech firms faced similar issues. Meta, Amazon, Salesforce, and Coinbase found their AI token bills climbing fast. Some companies even introduced leaderboards to reward the biggest token users. That created a culture of “tokenmaxxing,” where employees tried to rack up usage just to show they were on board.

That enthusiasm backfired. Spending soared beyond expectations. At Amazon, internal token leaderboards were scrapped after criticism. Meta revised how it tracked usage. Uber capped monthly AI spending per engineer to control costs. Executives admit it’s hard to link soaring AI expenses directly to productivity gains.

Why Token Pricing Breaks Budgets

AI providers have switched from flat subscription fees to token-based billing. Tokens track how much text or code the AI processes. This makes costs unpredictable. One engineer’s bill can be ten times higher than another’s. Costs vary widely depending on how they use the tool.

Traditional software budgets expect fixed costs per user. AI’s variable pricing breaks that model. Finance teams struggle to forecast expenses. Some companies lack systems to monitor or cap token consumption. Without controls, usage can spiral out of control.

On the tech side, the AI itself hasn’t failed. Engineers use tools exactly as intended. But the pricing math creates huge surprises. Token costs per unit have dropped nearly 98% since 2022. Yet total spending has tripled. This is because agents and workflows consume many times more tokens than simple chats.

Balancing AI Adoption with Cost Control

Some companies, like 8×8, see AI as a cost saver. They canceled other software subscriptions because AI tools replaced them. Their AI bills remain below the savings, for now. They track usage with dashboards and encourage employees to use AI wisely. Still, they may set future caps as usage grows.

Others take a tougher approach. Coinbase set spending limits by role and seniority. Walmart restricted internal AI programming tools. Many firms move routine tasks to cheaper AI models. They assign complex work to premium systems only when needed. This multi-tier model routing reduces costs.

Boards and finance teams demand value, not just usage. They want clear proof that AI spend delivers faster cycles, better products, or higher revenue. If projects miss targets, leaders want the power to pause or stop funding. This shift marks a move from AI hype to disciplined management.

Finance leaders now treat AI expenses like cloud computing: variable costs needing constant monitoring. AI budgeting requires real-time dashboards, alerts before overspending, and team-level accountability. Without these, runaway token bills will keep surprising CFOs and CEOs.

The tokenmaxxing phase showed that “everyone should use AI” without guardrails can lead to waste. The next phase focuses on “smart AI use.” This means matching the right model to the right task, tracking consumption closely, and linking spending to measurable business results.

AI is a powerful tool, but it’s no longer free to experiment with at scale. Companies must balance enthusiasm with financial discipline. Otherwise, AI costs can quickly look like payroll—big, recurring, and hard to control.