Here is the finding that should reshape how you budget for AI. Between the first quarter of 2025 and the first quarter of 2026, the blended cost of AI fell about 67% per unit, according to figures presented at FinOps X 2026. Yet enterprise AI bills went up, not down. Cheaper tokens did not produce cheaper AI.
The reason is simple arithmetic that most 2026 budgets got wrong. Total spend equals price per unit multiplied by volume consumed. The price fell sharply. The volume grew faster. And almost nobody was tracking the second number.
What is AI FinOps?
AI FinOps is the practice of giving AI consumption the same financial discipline you already apply to cloud, headcount, and any other operating cost. It makes AI spending visible, attributes it to teams and use cases, and ties it to the value produced. In short, it is cost accountability for a resource that scales with every prompt.
The discipline is not new, it is borrowed. FinOps grew up managing unpredictable cloud bills, and AI is the next unpredictable bill.
The shift is real. FinOps X 2026 reported that the share of FinOps teams managing AI spend jumped from roughly 31% two years ago to about 98% today. Managing AI cost has become a core finance function, not an afterthought.
Why do AI bills keep rising when prices are falling?
AI bills keep rising because falling unit prices quietly invite higher consumption. When each token or query costs less, teams use far more of them, and modern agentic systems can consume tokens at a rate no human typist ever could. The saving per unit is swallowed by the growth in units.
The 2026 examples are stark. Ride-hailing firm Uber reportedly exhausted its annual AI budget in four months and capped employee AI spending at US$1,500 per month.
Tesla capped employee AI spending at US$200 per week from early July, and Walmart, Amazon, and Cisco introduced their own controls.
These are not small companies with weak finance teams. They are sophisticated operators who discovered that without consumption controls, cheaper AI simply means you buy much more of it.
What was tokenmaxxing, and why did it end?
Tokenmaxxing was the 2025 mindset of maximising AI usage on the assumption that more tokens meant more productivity. It ended in mid-2026 when enterprises saw the bills and realised that raw consumption is a cost, not an achievement. The new success metric is value produced per dollar of AI spend, not tokens consumed.
The cultural change is the point. In the tokenmaxxing era, the heaviest AI user looked like the most advanced team.
In the efficiency era, the team that ships the most business value per dollar wins. This reframing, from consumption as a badge to consumption as a budget line, is the real shift enterprise leaders need to internalise for 2026.
Why does AI cost matter more for mid-market Hong Kong firms?
AI cost matters more for mid-market firms because they lack the balance sheet to absorb a runaway bill the way a global giant can. A Hong Kong company of 50 to 500 employees feels an unbudgeted six-figure AI overrun immediately, and it lands on the department head who championed the project. The margin for error is thinner.
Large firms can cap spending and move on. A mid-market operator has to justify every dollar to a CFO who is already sceptical of AI returns.
This is where an uncontrolled AI programme becomes a career risk, not just a cost risk. The leader who cannot explain the AI line on the budget loses the credibility to ask for the next investment.
How do you build an AI FinOps framework?
You build an AI FinOps framework by making AI spend visible, attributable, and matched to value, in that order. You cannot control what you cannot see, so visibility comes first. Four practical steps turn the principle into an operating routine your CFO will recognise.
1. Meter everything. Track AI consumption by team, use case, and model. A single blended invoice hides the runaway use case that is driving the overrun.
2. Attribute cost to an owner. Every AI use case has a budget owner who sees its spend and its output side by side.
3. Match model to task. Route simple tasks to smaller, cheaper models and reserve premium models for work that genuinely needs them. Paying flagship prices for trivial queries is the most common source of waste.
4. Measure value, not volume. Report outcomes per dollar, hours saved, tickets resolved, revenue influenced, not tokens consumed.
Together these four turn AI from an open tab into a managed investment.
Does AI FinOps mean spending less on AI?
No. AI FinOps means spending deliberately, not spending less for its own sake. The goal is to redirect budget from low-value consumption toward high-value use cases, so the same dollar produces more. A well-run AI programme often spends more in the right places and far less in the wrong ones.
Cutting AI spend blindly is as damaging as letting it run wild, because it starves the use cases that actually pay back.
The discipline is about signal, not restraint. When you can see which use cases return value and which merely consume, you can fund the winners with confidence and retire the rest. That is the difference between a defensible AI budget and a nervous one. With UD, AI works for you, not the other way around.
The takeaway for enterprise leaders
Falling AI prices are not a reason to relax. They are the exact condition under which unmanaged consumption explodes. The enterprises that stay in control in 2026 are the ones that treat every prompt as a purchase and every use case as an investment with an owner and a return.
Cheaper AI rewards the disciplined and punishes the careless. AI FinOps is how you make sure you are the former.
Ready to Take Control of Your AI Spend?
Understanding AI FinOps is the first step. Applying it to your own stack is where the value is. UD's AI Ready Check helps you see where your AI spend goes and where it pays back, and we'll walk you through every step, from cost visibility to model selection and value tracking, backed by 28 years of Hong Kong enterprise experience.
