King River says Australia has a two-year AI window. The advantage is something you already own.

A venture firm says Australia has a two-year window to build AI champions. The part that matters for mid-market operators is that the advantage it describes is one you already own.

A venture firm called King River Capital published an analysis a week or so ago, and one line from it travelled further than the rest: Australia has roughly two years to build globally competitive AI companies before the window closes, somewhere around 2028 to 2029. Most of the coverage read that as a message to startup founders. I think the more useful reading is for the people already running established businesses.

Underneath the headline sits a number that is easy to skim past. The cost of running an AI model, what the industry calls inference, has fallen by something like a thousand times in three years. That is the actual news. When a capability gets a thousand times cheaper, work that was uneconomic last year quietly becomes ordinary this year.

Here is the part worth sitting with. Cheap inference is available to everyone. The big labs have it, every funded startup has it, your competitors have it. So the cheap model itself is not an advantage. It is a commodity. The advantage is the thing a model cannot reach on its own.

King River is fairly specific about where Australia can win: financial services, health, legal, resources, defence, government. What those sectors share is not clever technology. It is customer relationships, regulatory knowledge, and proprietary data built up over years. A lab in San Francisco does not have your claims history. A two-year-old startup cannot manufacture twenty years of client files, or a book of relationships you have spent a career earning.

If you run a mid-market business in Australia or New Zealand, that should land differently. You are sitting on the exact asset this whole analysis is about. The distribution, the domain knowledge, the messy data nobody outside your organisation can see. The window being described is not really about who can build a model. It is about who points one at data and relationships they already own, before someone else works out how to compete for the same customers without them.

"If Australian capital, corporates and customers do not move quickly enough, global AI companies will move faster."

That is Zeb Rice, one of the firm's managing partners. Note the word corporates, sitting quietly in the middle of that sentence. This was never only a startup story.

And moving quickly does not mean a company-wide AI programme, or a moonshot. It means picking one workflow that sits directly on top of something only you have, your claims history, your client files, your support transcripts, and putting a model to work on it. Narrow and real beats broad and theoretical, every time.

The businesses that do well over the next few years will not be the ones with the best model. Everyone has the same models. They will be the ones who pointed those models at something only they own. The window is not open because the technology finally arrived. It is open because the technology got cheap, and for once the advantage sits with whoever already has the customers. That will not stay true for long.