Apple reportedly cut its engineers' daily AI budgets from $300 to $200, with further cuts to $100 under consideration — and some non-engineering teams possibly falling to $25. Employees may request exceptions, and Apple is encouraging the use of cheaper models and on-device AI. For the world's most valuable company, a company that could fund a $300-a-day budget for every employee without noticing, the move is striking. It is also a signal: the era of unlimited AI spending inside Big Tech is ending.
The Numbers Behind the Cut
The figures are small in absolute terms but large in what they reveal. A daily budget of $300 per engineer buys a great deal of frontier-model usage — thousands of queries, long-context reasoning, heavy code generation. Cutting it to $200 is a one-third reduction. A further cut to $100 would halve it again. And $25 a day for non-engineering teams is a different world entirely: enough for routine assistance, not for heavy lifting.
The exception process matters too. When employees must request permission to exceed a budget, usage changes — people think before they prompt, choose cheaper models for simple tasks, and reserve the expensive ones for work that genuinely needs them. That behavioural shift is arguably the point. Apple is not just saving money; it is teaching its workforce to treat AI compute as a scarce resource with a price.
Why the Richest Companies Are Rationing
The first reason is scale. A $300 daily budget sounds generous for one engineer. Multiply it by tens of thousands of engineers, and by the millions of employees across Big Tech who now use AI tools daily, and the total becomes enormous — hundreds of millions of dollars a year in model costs, growing as usage grows. Frontier-model inference is expensive because it is computationally intensive: every query runs on specialised hardware that costs real money to buy and to power.
The second reason is that the spending was never disciplined. In the first wave of enterprise AI adoption, companies raced to give employees access to the best models with few constraints, because the priority was learning what AI could do. That phase is over. Companies now have usage data, they know which tasks genuinely benefit from the most expensive models and which do not, and they are optimising accordingly.
The third reason is strategic. Apple is encouraging cheaper models and on-device AI — running models on the device itself rather than in the cloud. On-device inference costs Apple almost nothing at the margin once the hardware is shipped, and it aligns with the company's privacy positioning. The budget cuts and the on-device push are two sides of the same strategy: reduce dependence on expensive cloud inference.
What "Cheaper Models" Really Means
The encouragement to use cheaper models reflects a change in the model landscape itself. Two years ago, there was effectively one tier that mattered: the frontier model, and everything else was noticeably worse. Today there is a real ladder — small, fast, cheap models that handle routine tasks well; mid-tier models for most professional work; and frontier models reserved for the hardest problems. The skill of using AI well increasingly means choosing the right rung of that ladder, not defaulting to the top.
What It Means for AI Startups
For AI startups, Apple's move carries two messages. The first is about pricing. If the world's richest company is rationing model spend, then every customer is price-sensitive — and startups that sell AI products must be able to explain their unit economics clearly. A product whose margins depend on unlimited frontier-model usage at today's prices is fragile. Founders should be able to answer, in a pitch, what happens to their margins if model prices fall by half — or if their customers start rationing the way Apple has.
There is also a subtler implication for startups that build on top of frontier models. If the model providers themselves are cutting prices and improving efficiency, the cost of the underlying capability will keep falling. That is good for startups that use models as an input, and bad for startups whose entire value proposition is access to a model.
The Bigger Picture
Apple's budget cut is not a retreat from AI. It is the opposite: a sign that AI has become so central to how the company works that its costs must be managed like any other major line item. The companies that win the AI era will not be the ones that spend the most on models. They will be the ones that spend most intelligently — matching the right model to the right task, building on-device where it makes sense, and treating inference as a resource to be optimised, not a tap to be left running.
Angeled.com connects founders with investors — and the founders who understand their AI costs before they raise are the ones investors want to back.
Read Next
OpenAI Releases Codex Security as Open-Source CLI and SDK
Databricks Unveils Always-On Genie, an AI Agent That Stops Waiting to Be Asked
DeepMind Launches AlphaGenome 2, a DNA Model That Reads Genomes Cell by Cell
ElevenLabs Unveils 13 New Products and Upgrades at Its New York Summit
Google Launches Gemini Decision 1, a Model That Answers Instead of Writing
Runway Open-Sources Praxis-1, Its Robot Model That Learned by Watching Video



