AI Spending Grew 5x in 16 Months. Most Businesses Have No Cap.

Uber gave 5,000 engineers access to AI coding tools in December 2025. By April, the company had burned through its entire annual AI budget. R&D spending hit US$951 million in Q1 alone — a 17 per cent year-on-year increase, as reported by Fortune. The cause was not reckless spending. It was usage-based pricing meeting uncapped demand.

The details are instructive. Adoption of Claude Code went from 32 per cent of Uber’s engineers in February to 84 per cent by March. Monthly API costs per engineer ranged from US$500 to US$2,000. At Priceline, a Cursor contract renewal came back four to five times more expensive, and one engineer spent US$40,000 on tokens in a single month, according to Faros AI CEO Vitaly Gordon speaking to TechCrunch. A healthcare organisation consumed a trillion tokens over six months, generating more than US$6 million in unplanned costs.

These are not stories about AI failing. They are stories about AI working — and nobody tracking the meter.

Across the businesses tracked by Ramp’s AI Spend Index — published June 2026 — total AI spending grew 497 per cent between January 2025 and April 2026. Token usage grew 1,001 per cent over the same period. Per-token prices dropped, but consumption overwhelmed the savings. Premium models now account for 55.9 per cent of costs, up from 5.7 per cent in June 2025, as businesses shifted from cheap models to the most capable ones.

The per-employee numbers tell the story. The median business spent US$46 per employee per month on AI in April 2026. But the range ran from US$3 for light users to US$7,500 at the most AI-intensive firms — a 680-fold gap. Month-over-month spending swings of 40 per cent or more were common. Only 15 per cent of enterprises could predict their AI costs within 10 per cent accuracy, according to Elvex’s analysis of enterprise AI budgets.

For an Australian trades business or professional services firm, the risk is not a US$500 million token bill. It is the gap between what you budgeted and what you actually spend.

Most Australian SMEs today pay flat subscriptions for AI: $20 to $30 a month per person for ChatGPT, Claude, or Copilot. Predictable. Manageable. We wrote recently about the opposite problem — businesses paying for AI licences nobody uses. The emerging risk is the reverse: costs spiralling when everyone uses AI and nobody tracks how much.

The shift is already in motion. As businesses move from general-purpose chat to API integrations, agentic workflows, and embedded AI features in their industry software, costs become usage-based. Every document summarised, every report drafted, every scheduling decision optimised consumes compute tokens — and tokens cost money. The platforms Australian businesses rely on — Xero, MYOB, ServiceTitan — are embedding AI features that will migrate from flat pricing to usage-based tiers as capability grows.

This is a cost intelligence problem. Uber’s COO Andrew Macdonald put it plainly, as reported by Fortune: if you cannot draw a direct line from AI spending to useful output, the trade becomes harder to justify. The same logic applies at any scale. A 10-person firm spending $300 a month on AI subscriptions can absorb the cost without scrutiny. That same firm running agentic document processing at variable token rates needs to know what it is spending — and what it is getting back.

First, audit your AI tools. Know which charge flat fees and which charge per use. Usage-based billing often hides inside platform features you already pay for — check your renewal terms and look for per-token or per-query pricing in the fine print.

Second, set spending caps. Most AI platforms and API providers offer budget limits. Set them before your team discovers how useful the tools are, not after. Uber imposed a US$1,500-per-month-per-tool cap on engineers retroactively. You can set yours proactively.

Third, tie spend to outcomes. If a tool saves your team five hours a week, that has a dollar value — at a loaded labour rate of $85 to $95 an hour for trades, five hours is roughly $450 a week. Make sure the AI bill sits well below it. If the maths works, scale confidently. If it does not, you have found the problem early enough to fix it.

Key takeaways

Scaling your AI without the surprises?

A short conversation can help you map what AI costs should look like for your business — and where the spending caps should sit. Book a call.