Copyright, AI training and the economics of innovation
Across Asia-Pacific, governments are investing heavily in AI infrastructure, research and skills as they compete to capture the economic opportunities from the technology. But the success of those investments will also depend on policy choices being made elsewhere — including in copyright law.
The reason is straightforward. Training data is a critical input into AI development, and copyright rules help determine how developers can access and use it. Those rules can affect the cost and availability of training data, the legal certainty facing developers and, ultimately, decisions about where to invest and develop AI models.
Our recent research across seven Asia-Pacific economies suggests this relationship matters. Countries with clearer and more flexible approaches to AI training tend to perform more strongly across key parts of the AI value chain, particularly research and development.
Copyright policy is not the only determinant of a country’s success in AI. But it can influence the level of investment in digital infrastructure, skills and innovation. As governments compete to build AI capabilities, copyright is becoming part of the economic equation.
Training data is an economic input
Developing modern AI models requires access to data at enormous scale. Copyright frameworks help determine what material developers can use, under what conditions and with what degree of certainty.
Where those rules are unclear or restrictive, developers can face higher costs, greater legal exposure and uncertainty over whether investments made today will remain viable tomorrow. The effects can extend beyond the immediate cost of compliance. They can influence whether firms proceed with projects, where they locate model development and what complementary investments they make in compute, talent and supporting services.
There is evidence that these differences show up in innovation outcomes. Research cited in our study found that countries with more permissive copyright frameworks produced 38% more AI patents and 32% more AI ventures per month than more restrictive regimes. Separate studies link high-value technology innovation and venture investment to wider gains in business creation and economic performance.
Asia-Pacific offers some revealing contrasts
We compared copyright frameworks in Australia, India, Japan, Malaysia, Singapore, South Korea and Thailand with performance across four stages of the AI value chain: research and development, digital infrastructure, ecosystem readiness and adoption.
Singapore combines the clearest and most flexible framework for AI training among the seven economies with strong performance across the value chain. Its computational data analysis exception and broad fair-use provision give developers comparatively high certainty around permitted uses, alongside safeguards governing how data can be accessed and used. Singapore also benefits from strong research capabilities, infrastructure, investment and adoption.
Japan has historically taken a relatively flexible approach to non-expressive uses of copyrighted material, including an explicit text-and-data-mining exception. It also performs strongly in AI research and patenting. Recent litigation and policy debate, however, could introduce greater uncertainty into that environment.
Australia sits at the other end of our assessment of copyright flexibility. Among the seven economies studied, it has the least flexible framework for AI training. Australia remains a comparatively modest performer in frontier AI research and development, and does not have an explicit exception for AI training or text and data mining, while future arrangements around licensing remain under discussion. Many factors contribute to Australia’s position in frontier AI, but the contrast illustrates how uncertainty around access to training data can weaken an otherwise strong ecosystem.
Similar issues arise in economies where the wider AI ecosystem is still developing. Our assessment identified greater uncertainty or more restrictive conditions around the use of copyrighted material for AI training in India, Malaysia and Thailand. These economies also face constraints around infrastructure, talent or investment. In that context, copyright-related costs and uncertainty risk compounding existing weaknesses rather than helping close the gap with regional leaders.
Copyright rules can reinforce—or frustrate—AI investment
One lesson from the comparison is that legal flexibility on its own is not enough.
Japan’s experience shows that favourable copyright settings cannot compensate for every weakness elsewhere in the ecosystem. Singapore’s performance reflects the combination of legal clarity with infrastructure, talent, capital and strong adoption.
But this also works in reverse.
Australia illustrates how a country can combine substantial investment in digital infrastructure and AI adoption with a copyright framework that provides comparatively limited flexibility for AI training. India, Malaysia and Thailand show a related challenge from a different starting point: where infrastructure, talent or investment are already more constrained, additional uncertainty around training data can make the development of domestic AI capabilities harder still.
Copyright policy can therefore reinforce investments being made across an AI strategy—or work against them.
That matters particularly as governments across Asia-Pacific seek to move beyond simply adopting AI technologies developed elsewhere. Countries increasingly want to attract higher-value activities: research, model development, specialised AI services and the talent and investment that accompany them.
Access to training data is one of the conditions shaping where those activities take place.
Adding economics to the copyright debate
There are genuine interests to balance in determining how copyrighted material should be treated in AI training. The economic evidence does not remove the need to consider the interests of creators and rights-holders.
But nor should those decisions be treated solely as questions of copyright law.
Our analysis suggests that clear, predictable frameworks that provide workable flexibility for AI training can improve the conditions for investment, innovation and the development of domestic AI capabilities. Conversely, restrictive or uncertain frameworks add costs and risk. Our cross-country analysis identifies cases where those weaker policy setting coincide with poorer outcomes in parts of the AI value chain, particularly research and development.
This matters because copyright policy does not operate separately from the billions being invested in compute, skills, research and AI adoption across the region. Governments seeking to build stronger domestic AI capabilities need to consider whether the rules governing training data reinforce those investments or work against them.
Copyright is therefore not a peripheral consideration in AI strategy. The choices governments make will help shape where AI research and development takes place, where investment flows, and how much economic value countries ultimately capture from the technology.
Explore more insights from our Economics in Tech Policy series, examining the economic implications of technology policy across APAC.
