A company that made $132 billion in profit last fiscal year is now counting how much its engineers spend on AI per day. On Wednesday, November 4, 2026, Bloomberg reported that Apple lowered the daily AI spending cap for some engineers from $300 to $200 in October, with further cuts possible. Some teams could fall to $100 a day, and some groups outside engineering could go from $100 to as little as $25. (Sources: TechCrunch, Reuters via The Standard)
The headline number sounds strict until you do the math. Over about 21 working days, $200 a day is roughly $4,200 a month, the same as 210 ChatGPT Plus subscriptions at $20 each. Meanwhile Accenture funds $29 a month per person. Apple's cut is real news, but the more useful story is the spread between those two numbers, and what it says about the end of unlimited AI at work.
Quick summary: Per Bloomberg-based reports on November 4, 2026, Apple cut some engineers' daily AI allowance from $300 to $200 and may go to $100; some non-engineering teams could drop from $100 to $25. Staff can ask senior leaders for exemptions. Few workers reportedly used the full $300, so the savings are unclear. Apple's limit is still well above those reported at Tesla, Uber and Accenture, and far above the $13 per active day Anthropic gives as average Claude Code spend in enterprise deployments.
What was reported, and what wasn't
We could not open Bloomberg's original article, so this relies on outlets summarizing it, and they agree on the core figures. The allowance covers tokens, the metered units AI models process, and staff have access to Anthropic's Claude, OpenAI's ChatGPT and Google's Gemini. Some teams are being pushed toward cheaper models or models that run on the device. Employees who need more can ask senior leadership for an exemption. (Sources: 9to5Mac, The Next Web)
Three things are missing. The reports do not say how many engineers or teams are affected, and they do not say how much Apple will save. Apple has not confirmed the numbers: The Next Web and Implicator say it declined to comment, and Reuters said it did not immediately respond. 9to5Mac is also careful to note that the specific tools covered are unclear. Treat the $100 and $25 figures as options reportedly under consideration, not as policy. (Sources: The Next Web, Implicator)
Apple's cap against everyone else's
Companies report these limits in different units, which hides how far apart they are. Here they are converted to a rough monthly figure, assuming 21 working days or about 4.33 weeks per month. The conversions are our math, and a cap is a ceiling rather than what people actually spend.
| Company | Reported limit | Rough monthly ceiling | How it works |
|---|---|---|---|
| Apple (some engineers) | $200 a day, down from $300 | About $4,200, down from about $6,300 | Exemptions via senior leaders; $100 a day possible |
| Apple (some non-engineers) | $100 a day, $25 considered | About $2,100, or about $525 at $25 | The lower figure is under consideration |
| Uber | $1,500 a month per tool | $1,500 per agentic coding tool | Dashboard tracking; can be exceeded with permission |
| Tesla | $200 a week, from July 6 | About $870 | Sign-off above the cap; xAI beta products excluded |
| Accenture | $29 a month for Claude, from October 1 | $29 centrally funded | Overage is charged to a project or team budget |
The Uber, Tesla and Accenture figures also trace to leaks and memos rather than public policy. Uber's cap was reported by Bloomberg, Tesla's by The Information and Accenture's by 404 Media. Accenture's model is the most interesting because it is not really a hard cap: an internal notice reportedly moved spending above $29 onto project or team budget codes, so the cost becomes something a project has to justify. Some staff had been spending as much as $1,000 a month. (Sources: TechCrunch on Uber, Electrek on Tesla, The Next Web on Accenture)
The real number: what do heavy users spend?
The best public benchmark comes from Anthropic's own documentation, which raised its estimate from $6 to $13 in April. Across enterprise deployments, it says Claude Code averages around $13 per developer per active day and $150 to $250 per developer per month, with costs below $30 per active day for 90% of users. Set against that, Apple's new $200 cap is about 15 times the average active day and more than 6 times the 90th percentile. Even a $100 cap would sit above 3 times that 90th percentile. (Sources: Anthropic docs, Business Insider via Yahoo Finance)
That lines up with Bloomberg's reported detail that few Apple workers used the full $300. It also suggests a cautious reading: on these numbers, a cut from $300 to $200 mostly trims the extreme tail, and the savings may be modest. There are limits to the comparison, though. Anthropic's figure covers one tool across many companies, while Apple's budget reportedly spans several vendors and may be used far harder than the average enterprise seat.
The tail is where the money goes. Gartner Peer Insights data cited in June found 23% of tech leaders spending $200 to $500 per developer per month on tokens, and 6% of organizations paying more than $2,000. The Next Web reported in June that Microsoft had found some engineers spending $500 to $2,000 a month on Claude Code alone, and Electrek reported that some Tesla engineers were burning thousands of dollars of tokens each week. Gartner's headline forecast is that AI coding costs will top the average developer's salary by 2028, though that is a prediction, not a measurement, and its analyst told The Register it does not describe every developer worldwide. (Sources: Gartner release, Computer Weekly, The Register, Implicator, The Next Web)
Why token budgets exist now
Per-token prices have been falling, so why are companies rationing? Because agents use far more tokens than chat. The Next Web says per-token prices fell about 98% from late 2022 while enterprise AI bills tripled as agentic tools consumed more per task. Anthropic's docs explain the mechanics: Claude Code resends conversation history with each request, and agent teams use roughly 7 times more tokens than standard sessions when teammates run in plan mode. (Sources: The Next Web, Anthropic docs)
Incentives made it worse. Earlier this year some companies ranked staff on token consumption, a habit nicknamed tokenmaxxing. Uber had leaderboards, according to The Information, and its CTO said in April that it had burned through its annual AI budget in about four months. Meta and some Tesla teams also had usage rankings. The reversal has been quick: Microsoft's Jay Parikh told employees that "tokenmaxxing is not what we are optimizing for," and every Microsoft division has had a token cap since July, per reports. Databricks is a holdout, with an engineering leader saying in June that its budget was still unlimited. (Sources: TechCrunch, The Next Web on Microsoft, The Next Web on tokenminimizing, 9to5Mac)
Why Apple, and why now?
Apple is not short of cash. Its 2026 annual report shows net income of $132.2 billion on the year ended September 26. But research and development rose 32% to $45.8 billion, and total operating expenses rose 22% to $75.5 billion. The filing says the R&D increase was driven primarily by higher infrastructure costs, including AI investments, and headcount. (Sources: Apple FY2026 Form 10-K, Implicator)
Coverage ties the cap to John Ternus, who became CEO on September 1 and has pushed the company to run "faster and leaner." Even so, keep the scale in view. If a single engineer really did hit a $300 limit every working day, the $100 cut would save about $25,200 a year. That is meaningful per person, but Apple's R&D bill grew by more than $11 billion in one year. Token caps look more like a discipline signal than a line item that moves Apple's results, which is our reading rather than anything Apple has said. Some staff reportedly see it as a sign of wider cost cutting and worry about layoffs; that is employee sentiment, not a company plan. (Sources: The Next Web, 9to5Mac)
What $200 of tokens actually buys
Apple's contract rates are not public, so use list prices as a guide. Claude Opus 5.5 costs $4 per million input tokens and $20 per million output tokens, which means $200 buys 50 million input tokens or 10 million output tokens before caching. Sonnet 5.5 is half that price at $2 and $10. Haiku 5.5, on prompts up to 100,000 tokens, is $0.10 and $0.50, or one-fortieth of Opus. That is why telling teams to use cheaper models is such a strong lever. Prompt caching also matters: an Opus 5.5 cache read is $0.20 per million, 95% below the base input price. (Sources: Anthropic pricing docs, Claude pricing page)
If your employer sets a cap, or you pay your own way
Whether you are a manager or an individual, the practical moves are the same, and most come straight from Anthropic's and Gartner's guidance. Gartner's blunt line is that token discipline will not emerge through developer choice alone, because developers optimize for speed and convenience.
- Start with a baseline. Anthropic recommends a small pilot group before wider rollout; Gartner suggests classifying work as developer-led, developer with an agent, or fully agent-led.
- Match the model to the task. Use a smaller model for simple subagent work and reserve the top model for hard architecture problems.
- Clear context between unrelated tasks. In Claude Code, a stale session gets resent with every request, so a one-line question late in the day can carry a whole day's history.
- Be specific. Vague requests like "improve this codebase" trigger broad scanning, while naming the file and function limits what the model reads.
- Watch background work. Scheduled tasks, subagents and agent teams keep consuming tokens even when you have stepped away.
- Prefer soft limits for real work. Uber's permission route, Apple's exemptions and Accenture's project codes all leave a path for high-value use, so a heavy user is not simply cut off.
If you pay for your own AI, compare before you upgrade. Flat plans such as Claude Pro at $20 or ChatGPT Plus at $20 cap your risk at the subscription price, while API billing has no natural ceiling unless you set one. Check your plan's usage page for a week before moving to a $100 or $200 tier.
01How much can Apple employees spend on AI per day?
Per Bloomberg-based reports, some engineers went from $300 to $200 a day in October, and some teams may fall to $100. Some non-engineering groups have a $100 daily cap, which Apple has reportedly considered lowering to as little as $25.
02Has Apple confirmed this?
No. The reports rely on unnamed people, and Apple declined to comment according to outlets including The Next Web.
03Is $200 a day a lot for AI?
Compared with public benchmarks, yes. Anthropic says enterprise Claude Code use averages around $13 per active day, and stays below $30 for 90% of users. Apple's budget can span several vendors, so it is not a perfect comparison.
04What is tokenmaxxing?
It is the practice of consuming as many AI tokens as possible, often encouraged by internal leaderboards. Several companies that ranked staff this way have since dropped the rankings or added spending caps.
05Which other companies cap employee AI spending?
Reported examples include Uber at $1,500 a month per tool, Tesla at $200 a week, Accenture at $29 a month of central funding for Claude, and division-level budgets at Microsoft. Walmart, Meta and AT&T have also been reported to limit use.
06Will this lead to layoffs at Apple?
Nothing in the reports says so. Some employees reportedly worry about it, but that is sentiment rather than an announced plan.
If you want to test whether a cheaper model is good enough for your work before paying for a top tier, LumiChats gives you 40+ AI models in a single subscription, so you can run the same task through several and keep your own budget predictable.
