Why Big Companies Are Rationing AI, and What Small Businesses Can Learn

The end of "use as much AI as you can"

Last month, Apple quietly cut how much some of its engineers can spend on AI tools each day, from $300 to $200. It is planning further cuts. Uber, Tesla and Microsoft have all done something similar this year. After a period when big companies encouraged staff to use as much AI as possible, they now want to know what they are getting for the money. In our view, this is a healthy change, and smaller businesses can learn a lot from the mistakes that led to it.

We've taken a closer look at what happened and what it means for businesses using AI day to day.

How AI is charged for

It helps to know how the bill works. Many AI tools charge by the "token". A token is a small piece of text, roughly part of a word. Every question you ask and every answer you get uses tokens, and the more powerful the model, the more each token costs.

For one person asking a chatbot a few questions, the cost is tiny. But the newer AI "agents", which work through a task on their own, can use huge numbers of tokens without anyone noticing. That is where the trouble started.

What went wrong

Earlier this year, "tokenmaxxing" became a trend in tech companies. The idea was that staff who used the most AI were the most innovative, so some firms ranked teams on leaderboards by how many tokens they used. Unsurprisingly, people used more.

Uber is the best-known example. It rolled out an AI coding tool to around 5,000 engineers in December 2025. By April, it had used its entire AI coding budget for 2026. Its chief technology officer said he had spent $1,200 in a single two-hour demonstration. In June, Uber capped each engineer at $1,500 a month per tool.

It wasn't alone. Tesla set a limit of $200 a week per employee. Microsoft moved many of its engineers off an outside AI coding tool and onto its own, GitHub Copilot. Several companies quietly took down their usage leaderboards.

Apple's cuts, reported by Bloomberg this month, are the latest. Some engineers are down from $300 a day to $200, with some teams likely to drop to $100 and some non-engineering teams to as little as $25. Staff can ask a senior manager for more if they need it. Apple has also asked some teams to use cheaper AI models where they can.

Interestingly, some Apple staff told Bloomberg that few people ever used the full $300. So the savings may be small. The message matters more: AI spending now has to show a return.

The real lesson: measure results, not usage

The mistake wasn't using AI. It was measuring the wrong thing. Counting tokens tells you how much AI someone used, not whether it did anything useful. It is a bit like judging a salesperson by how many phone calls they make, rather than how many sales they close.

In our experience, the businesses getting the most from AI start with a specific job, such as answering routine customer questions, drafting quotes or summarising long documents. They then measure whether that job is now quicker, cheaper or better. The cost of the AI only makes sense next to that result.

What this means for smaller businesses

Most small businesses won't run up a bill like Uber's. A few monthly subscriptions to tools like ChatGPT, Claude or Copilot are a fixed cost. But as more businesses connect AI to their own systems, or try agents that work on their own, the risk of an unexpected bill grows. A few simple habits help:

Set limits and alerts. Most AI services let you set a monthly spending cap or a warning when you reach a certain amount. Do it on day one, not after the first surprise.

Use the right tool for the job. The most powerful model isn't always needed. Simple jobs, like tidying up an email, can often be done just as well by a cheaper model.

Watch for pricing changes. Some providers are moving from a flat monthly fee to charging for what you use. Check what you are signed up to and what happens if usage grows.

Cut the low-value uses. Turning a PDF into slides with AI might feel productive, but if no one needed the slides, it is cost without value.

Review it regularly. Once a month, look at what you are spending on AI and what it is doing for you. Keep what works and stop what doesn't.

We've written in more detail about why AI costs behave so differently from normal software in FinOps for Agents, if you'd like the fuller picture.

If you'd like help working out where AI is genuinely paying its way in your business, and where it isn't, get in touch. We're always happy to talk it through.

Emma at Cow-shed