King River Capital says Australia has a two-year window to build global AI companies before offshore rivals capture much of the value. Its market analysis suggests that window will close around 2028 to 2029.
The venture firm argues that sharply lower costs for using and developing AI models have changed the economics for founders, making it cheaper to build products around business tasks and industry workflows. At the same time, the trend gives larger international groups a faster route into local markets, where they can win customers and gather data.
King River bases its view on patterns from earlier technology cycles, including the internet, mobile and cloud. In each wave, about 80 per cent of the value created went to companies founded within the first four to six years of those markets taking shape.
That comparison matters because AI adoption is moving faster than previous technology shifts. With generative AI entering the mainstream after the release of ChatGPT in late 2022, the period in which new companies can establish durable positions may be unusually short.
Zeb Rice, co-founder and managing partner at King River Capital, said lower costs had expanded what local founders could attempt.
"This materially changes what Australian founders can build," Rice said.
"The opportunity is not just to build better software, but to build AI companies that solve important problems in sectors where Australia has real expertise, trusted customers and valuable data. But this is a two-year window, not five years. If Australian capital, corporates and customers do not move quickly enough, global AI companies will move faster, win the customers, absorb the data and own the workflows."
Cost shift
Central to the argument is the sharp fall in inference costs, the expense of running a model after training. King River says those costs have fallen by about 1,000 times in three years, cutting the price of embedding AI into software products and day-to-day business processes.
The cost of building and adapting large language models has also dropped. Pre-training costs have been falling by around 10 times a year, reducing the strategic advantage of simply owning the largest model and shifting attention to customer access, proprietary data, workflow knowledge and distribution.
That change favours companies with deep knowledge of a specific sector, rather than those trying to compete directly with the biggest global model builders. Australia is unlikely to challenge the largest international players in frontier foundation models, but it could still produce sizeable businesses by applying AI in markets where local groups already have customer relationships, regulatory knowledge and data.
The sectors King River identified include financial services, health, cyber and defence, resources, physical AI, legal services and government services. In those areas, local companies may have a credible opening if they move quickly enough to secure customers and embed products in real workflows.
Workflow race
Another element in the analysis is the shift in AI from narrow tasks to broader operational work. Improvements in model memory now allow systems to handle much larger sets of information at once, including codebases, contract libraries and internal records, making them more useful for multi-step work across an organisation.
That means the contest is no longer just about chatbots or simple productivity features. The bigger prize is becoming part of the systems companies use to run finance, operations, compliance, customer service and other core functions.
King River frames the market in broader terms than software spending alone. It says more than USD $30 trillion of global cognitive payroll is beginning to come within reach of AI, and that even a 5 per cent share of that work would exceed the size of today's entire US software market.
For Australia, the concern is that overseas groups could lock in that value by establishing themselves first with local customers. If that happens, the data, intellectual property, company value and investment returns linked to the next phase of AI could increasingly sit offshore.
Chris Barter, Co-founder and Managing Partner at King River Capital, said the main challenge was not the country's technical talent, but the conditions needed to build companies at scale.
"Australian AI companies need capital, early enterprise customers, data access, reference accounts and global distribution if they are to become category leaders from Australia," Barter said.
"Early enterprise customers matter because they do more than buy a product. They help prove the workflow, sharpen the use case, deepen the data advantage and create the credibility needed to take on global markets.
"Australia has already shown it can produce globally relevant technology companies.
"The task now is to ensure the next generation of AI companies has the capital, customers and support to scale from here."