Almost no organisation has a dedicated AI budget. But everyone is spending on AI. Which raises a question worth asking: where is the money coming from?

The answer says more about how organisations are actually thinking about AI than the strategy decks do.

The four AI conversations every leader is having

Most conversations with senior leaders right now fall into one of four buckets.

The first is teams still trying to work out which use cases to pursue. The second is teams that have picked a use case and suddenly discovered their data isn’t ready. The third is pilot purgatory: it worked in a controlled environment, but the path to production isn’t clear. The fourth, which is rapidly becoming the loudest, is teams that got something into production and watched the token costs blow out.

That last group is where CFOs enter the conversation, and it’s shifting a lot of teams towards hybrid setups. Some tokens generated in the cloud, some running on local infrastructure, to keep unit costs manageable.

The AI budget nobody actually has

Most organisations don’t have a dedicated line item for AI spend. What they have is quiet reallocation.

Two patterns show up over and over. The first is deferred infrastructure refreshes: sweating existing assets a little longer to free up budget for AI initiatives. The second, and this one is more interesting, is cost-cutting announcements that use AI as the justification.

When you dig into those cost cuts, AI is usually the permission slip. Since, a lot of organisations over-hired during the pandemic and would have had to restructure anyway. AI just makes the story easier to tell to the market.

The wrong question, and a better one

The framing “how do I reduce cost with an AI agent” is the wrong starting point. It leads to defensive thinking and vanity projects.

The better question is: if we designed this work today, with the tools that now exist, what should people own, what should software assist with, and what should be fully automated? That reframes AI from a cost lever into a workflow redesign question, and the answers tend to be much more useful.

There is a related structural shift worth naming. AI has decoupled growth from headcount. For most of business history, growing revenue meant growing the team roughly in proportion. That is no longer true. The interesting conversation isn’t “how do I cut headcount.” It’s “how do I grow without adding it.”

How to measure AI maturity properly

Most organisations measure AI maturity by how many tools they’ve bought or how many pilots they’ve launched. That is counting activity, not value.

A better measure: how many of those tools have actually made it into production and are generating real ROI. If you follow the money, most spending is still going into projects where the ROI hasn’t been defined at the outset. That is where the vanity projects hide.

A quiet point on Australian AI infrastructure

Australia is taking on a leadership role in AI data centres. There is land, and there are good energy sources, which makes it an attractive location for building out capacity. Most of that infrastructure is being built by Australian-owned and operated businesses.

What is less discussed: an increasing share of the funding for those projects is coming from outside Australia. That is a structural shift worth watching. As Australian data centres generate more of the AI tokens businesses rely on, there is a case for leaders to be more deliberate about supporting Australian providers where they can.

The takeaway

Following the money tells you where AI actually sits in an organisation’s thinking. Reallocated infrastructure budgets. Pandemic-era headcount adjustments dressed up in AI language. Vanity pilots without measurable ROI. And quiet CFO conversations about token economics. All happening in parallel, in most large organisations, right now.

Being deliberate about how you’re funding AI, and what you expect back from it, is where the discipline gap will show up over the next couple of years.