By Brad Ferris · 12 September 2026
On 9 September the AFR's Chanticleer column ran a piece explaining why Australia's $65 billion quant investing king is an AI sceptic. Vinva's Morry Waked has spent a career building systematic strategies on data and computation, so the scepticism is not the reflex of someone who dislikes the technology. His argument is narrower and harder to dismiss. When every fund runs similar models over the same earnings-call transcripts, the signal they all find at the same moment stops being worth anything. He calls it alpha concentration.
Funds management is the purest version of the problem, because in that industry an edge only exists relative to the person on the other side of the trade. But the mechanism is not special to investing. If you and four competitors buy the same AI product, feed it the same publicly available inputs, and run it over the same generic process, the sensible expectation is that you all arrive at a similar place at a similar time. The capability is real. The advantage is not.
The adoption data has been saying something like this for a while, and it is easy to misread it as a story about who is behind.
McKinsey's most recent global survey, covered by HPCwire in early September, found that 40 per cent of organisations above US$1 billion in revenue are now scaling AI agents, up from 27 per cent a year earlier. At smaller organisations the figure is 22 per cent, and it did not move over the same year. The usual reading is that big firms have better tools. They do not. Every model in that survey is available to a fifty-person business in Brisbane on a credit card, at a price that would have been implausible three years ago.
What the larger firms bought with their scale is the unglamorous part: the data plumbing, the process documentation, the change capacity to get something from a pilot into daily use and keep it there. That is the gap the number is measuring.
Gartner supplied the other half of the picture this week. In a survey of 161 chief audit executives, 93 per cent reported some AI use, while only 38 per cent had a formal AI strategy, 54 per cent had not started measuring return, and just 15 per cent had structured or defined use cases. Internal audit is the function whose job is testing whether controls work. It has adopted the technology far faster than it has organised around it, and if that is true there, assume it is true in your operations team.
Near-universal use, thin discipline, unmeasured return. That combination produces exactly what Waked describes: everyone holding the same capability and nobody holding an advantage.
Three things stay proprietary once the model itself is a commodity.
The first is your data. Not the market data everyone has, but the record of what happened inside your business: which quotes converted and at what margin, which jobs ran over and why, what the customer said on the phone before they cancelled. A general model has never seen any of it, and no subscription supplies it.
The second is your process. Most mid-market businesses run on decisions made by four or five experienced people who have never written down how they make them. That knowledge is the real input to a useful AI workflow, and getting it out of people's heads is slow, human work that cannot be procured.
The third is the ability to act on the first two. An answer that nobody acts on has the same value as no answer.
The AFR profile of Nine chairman Peter Tonagh this week is a useful illustration, including his practice of building an AI twin of a counterparty to rehearse before a major negotiation. The model doing that work is the same one his counterparties can buy. The material fed into it, and the judgment about what to rehearse, is not.
Pick the AI workflow your business relies on most, and ask one question about it: if a direct competitor bought the identical product tomorrow, what would they still be missing?
If the honest answer is nothing, you have bought a capability rather than built a position. That is not a failure, and it may still be the right purchase, in the same way that everyone has payroll software and nobody claims an edge from it. Treat it as a cost line, buy it at the best price you can, and stop expecting it to show up in your margin.
If the answer names something specific, your customer history, a costing method nobody else uses, a quoting sequence your best estimator refined over fifteen years, then the work in front of you is to get more of that material into the system and less of the generic input.
The same reasoning applies to what you are being sold. Lightspeed, the venture firm behind Anthropic among others, told the AFR this week that incumbent software vendors' AI pivot narratives are largely marketing, with the revenue effects taking two to three years to land. Read that as a renewal-negotiation instruction. If the AI feature your vendor is charging a premium for will be standard inside the contract term, price the contract accordingly.
The models will keep getting better and cheaper, and that helps everyone equally, which is another way of saying it helps nobody in particular. What separates two businesses running the same model is what each one feeds it and what each one does with the answer.
Where does your business stand? The free AI Scorecard takes three minutes and shows you. If you want a straight steer from a person, book an AI Opportunity Call.