Apple trims engineers' AI budgets to $200 a day

The open AI bar at work is closing. Apple just set last call.

Apple reduced some engineers' daily AI spending caps from $300 to $200 last month, Bloomberg reported Wednesday, according to 9to5Mac's account of the report.

The caps, by the number

Per Bloomberg's report, as relayed by 9to5Mac:

• Some engineers' daily caps fell from $300 to $200 in October.
• Further cuts are planned for some engineering teams, possibly to as little as $100 a day.
• Many non-engineering groups already work under a $100 daily cap, and Apple has considered going as low as $25 for some teams.
• Employees who need more can ask senior leadership for an exemption.
• Some teams are being encouraged to use cheaper models, or models that run on their devices.

$25 a day, at the low end.

Bloomberg ties the cuts to cost-cutting under CEO John Ternus. 9to5Mac notes Apple's fiscal fourth-quarter R&D spending rose 32% to $45.8 billion, partly on infrastructure and AI costs.

There is a catch in Apple's own numbers. Bloomberg's sources said "few workers were using their full $300 daily allowances," and Bloomberg says it is unclear how much Apple will save.

Budgets are the story, Apple is the example

Apple employees have access to Anthropic's Claude, OpenAI's ChatGPT and Google's Gemini, 9to5Mac reports. It does not say which tools the new limits cover.

Apple isn't alone. 9to5Mac cites a report that Accenture capped its centrally funded Claude allowance at $29 per employee per month.

So the cap is real, even if Apple's first cut barely bites. A $300 ceiling few people hit was a statement that AI spend didn't need managing. A $100 ceiling, or $25, is a statement that it does.

Yes, $200 a day is still a lot of tokens. For most people it's plenty. But an engineer running long agent sessions on a frontier model can spend real money fast, and that's exactly who these caps are aimed at.

What it means for you

If your employer pays for your AI, assume a budget is coming, and assume it will be set by someone looking at a spreadsheet, not at your work.

You'll do better if you show up with the split already done. Which tasks need the best model you can get? Which are formatting, extraction or quick rewrites a small model handles fine? That habit has a name, model routing, and Apple is pushing its teams toward it. Some of Apple's teams are also being pointed to on-device AI, which costs nothing per token.

If you run a team, set the budget before finance does, and set it from usage data. Bloomberg's own reporting suggests the first cap was set above what most people used.

What to watch

Whether Apple follows through on $100 for engineering teams, and whether exemptions become routine. A cap with easy exemptions is a reporting tool. A cap without them changes how people work.

Then watch your own company's next tooling memo.