Apple has begun rationing the AI its employees can buy by the day. According to Bloomberg, citing people familiar with the change, the company quietly lowered some engineers’ daily spending caps on AI models from $300 to $200 in October. Further cuts under discussion could take some engineering teams to $100 a day, while certain non-engineering groups already on a $100 cap may face limits as low as $25. Apple has not publicly confirmed the figures; a spokesperson declined to comment when asked by outlets covering the report.
The numbers are less interesting than what they signal. After a year in which large tech employers urged staff to “use AI everywhere,” the bill for tokens—the units of text and other data models process—has become a line item worth managing. Apple is still pushing AI deeper into engineering and internal operations. It is also putting a meter on that usage. Those two facts sit together awkwardly, and that tension is the real story.
What is confirmed, attributed, and still open
Confirmed from Apple’s own fiscal 2026 results (year ended Sept. 26, 2026): research and development expense rose 32% to $45.765 billion, and total operating expenses rose 22% to $75.539 billion. The company has linked higher R&D in part to infrastructure and AI investment. On a Monday earnings call, CFO Kevan Parekh said Apple continues to increase overall AI investment while funding the product roadmap.
Attributed to anonymous sources via Bloomberg (Nov. 4, 2026), and repeated by secondary outlets including 9to5Mac, CNA/Reuters, and The Next Web:
- Some engineers’ daily AI allowances cut from $300 to $200 in October.
- Further engineering cuts under consideration, potentially to $100 a day.
- Non-engineering teams already on $100 daily caps; some may fall to $25.
- Exemptions available via senior leadership for people who need more.
- Budgets cover tokens across services staff can access, including Anthropic’s Claude, OpenAI’s ChatGPT, and Google’s Gemini.
- Some teams are being steered toward cheaper models or on-device options.
Still open: how many employees are affected, which tools sit inside which caps, how much money the October cut actually saved, and whether the $100/$25 figures become policy or remain options under review. Internal reactions, also secondhand, are mixed—some staff say few people hit the old $300 ceiling; others read the cuts as part of wider cost pressure under CEO John Ternus, who took the role on Sept. 1 and has pushed for a leaner organization.
Why token caps appear when AI spend is rising
Apple’s employee allowances and Apple’s corporate AI budget are not the same ledger. R&D and infrastructure spending fund models, Private Cloud Compute capacity, product features, and outside cloud. Daily token caps govern how much individual workers can burn on third-party APIs and coding assistants while doing their jobs. Cutting the second does not reverse the first. It can even coexist with rising AI investment: build product AI centrally, police discretionary inference at the desk.
That split matters for readers who treat “Apple cuts AI spending” as a single sentence. Capex and R&D can climb while personal usage gets rationed. CFO Dive’s reading of the same earnings cycle noted Apple’s property, plant and equipment spend fell even as R&D rose—another reminder that Apple’s AI posture is not a simple “spend more everywhere” story.
The industry shift: from tokenmaxxing to metered AI
Across Silicon Valley, early AI adoption often rewarded volume. Internal culture sometimes treated heavy model use as a proxy for productivity—“tokenmaxxing,” in the phrasing that circulated among developers. As coding agents and assistants became common, the marginal cost of another prompt was easy to ignore until finance saw the invoice.
Apple is not alone in reintroducing friction. Coverage of the Bloomberg report placed it next to other employer moves, including Goldman Sachs restricting some AI coding tools after productivity and risk reviews, and reports that Accenture tightened centrally funded Claude allowances so extra spend had to sit on project budgets. The pattern is consistent: after open-ended experimentation, organizations invent governance—daily caps, cheaper default models, on-device routing, and exception processes.
Who benefits from that shift? Providers of efficient or local models, spend-control and observability tools, and teams that can show output per dollar rather than tokens per day. Who loses? Workflows that equated unlimited high-end model calls with velocity, and vendors whose pricing assumed unconstrained internal use.
What a $200 day actually buys—and what it doesn’t prove
Two hundred dollars a day is still a large allowance for most knowledge work. Heavy coding and debugging sessions can consume that quickly on frontier models; many employees reportedly never approached $300. So the October cut may be mostly symbolic for average users and binding mainly for intensive agentic coding loops.
What the cap does not prove: that Apple’s AI tools fail a productivity test. Bloomberg’s sources do not publish measured output gains or losses. Rationing can mean weak ROI, runaway usage by a minority, risk management, or simply a new CEO imposing cost discipline. Treating the leak as proof that “AI doesn’t work at Apple” overclaims the evidence.
What it does suggest: AI inference inside large companies is graduating from free utility to managed resource—closer to cloud spend than to email. Once there is a meter, product managers and engineers will optimize for cost as well as quality, which is usually how technologies mature.
What to watch next
Three follow-ups would turn this from a leak into a durable industry story. First, whether Apple publishes or leaks clearer rules: which models, which teams, what exemption rate. Second, whether further cuts to $100 or $25 land as policy. Third, whether peers disclose similar internal metering, and whether any company pairs caps with hard productivity metrics instead of anecdotes.
Until then, the careful reading is narrow but important. Apple still says it is investing heavily in AI. It is also, according to reporting it has not confirmed, telling some employees that unlimited token spend is over. The meter is the message.