Strict sovereign AI policies could cost APAC economies billions
A new Oxford Economics report reveals that pursuing total AI self-sufficiency will lead to economic trade-offs, delayed enterprise adoption, and higher carbon footprints across the region
Governments across the Asia-Pacific region (APAC) have been doubling down on sovereign artificial intelligence (AI) initiatives to protect national security, cultural values, and data privacy. However, highly restrictive policies that locks out global cloud and AI providers could have an impact on national economies, according to a new report by Oxford Economics.
The report, commissioned by the AI Adoption Initiative, a global community of AI policy experts, warned that while building domestic AI ecosystems is positive, mandating full technological self-sufficiency will result in skyrocketing infrastructure costs and delays in enterprise AI adoption.
In the most restrictive scenarios, where governments mandate a domestically owned, full AI stack, large economies like Japan and India could face additional direct costs of $149.7bn and $102.5bn respectively between 2025 and 2035.
Speaking to Computer Weekly, Henry Worthington, managing director at Oxford Economics, noted that the research was designed to quantify the impact of policy decisions around sovereign AI. “This is one of the first efforts to model and put numbers around the potential trade-offs that you might be making as a government, depending on your position,” he said.
The report categorises sovereign AI policies into five levels of restrictiveness, from control-and-choice policies, which maintain access to global cloud providers and apply data residency requirements to a narrow set of highly sensitive workloads, to costly ownership-centric policies that maximise formal control and strategic autonomy.
While the direct costs of building physical AI datacentres, procuring graphic processing units (GPUs), and training local talent are huge, Worthington noted that the actual economic damage comes from lost productivity.
“What drives probably a larger proportion of the GDP loss, particularly in more restrictive policy scenarios, is the delay that they create in facilitating AI adoption within the enterprise sector and public sector organisations,” Worthington explained.
“It is that adoption which will ultimately drive the long-run economic benefit that you would expect from this type of general-purpose technology. What you’re getting if you go down a highly sovereign route is almost inevitable delays to the best technology being available to resident entities,” he added.
Under the highest restriction levels, the report estimates that AI adoption among firms could be delayed by three to five years, causing a loss in productivity. For Japan alone, this opportunity cost translates to a cumulative loss exceeding $58.2bn by 2035.
Beyond GDP and productivity, the report pointed to an often-overlooked consequence of strict AI sovereignty: environmental toll.
Hyperscale cloud providers achieve high energy efficiency through purpose-built infrastructure, advanced cooling, and economies of scale. Forcing AI workloads into smaller, fragmented, domestic datacentres directly forfeits these efficiency benefits.
The report projected that in restrictive scenarios, the duplication of infrastructure will lead to significantly higher carbon emissions and water consumption, a critical issue in APAC where many economies still rely heavily on carbon-intensive power grids and face high water stress.
What drives probably a larger proportion of the GDP loss, particularly in more restrictive policy scenarios, is the delay that they create in facilitating AI adoption within the enterprise sector and public sector organisations
Henry Worthington, Oxford Economics
A tale of different nations
The APAC region is currently fragmented when it comes to policies around AI sovereignty. Worthington pointed to Japan and Singapore as examples of nations walking the tightrope successfully.
“Japan has taken the bull by the horns and adopted what we would view as a very sensible approach, and early anecdotal evidence shows that it is accelerating adoption within the business community,” Worthington said, referring to Japan’s light-touch approach to regulation aligned with international norms.
He also praised Singapore for identifying areas of comparative advantage and maximising the AI opportunity through a hybrid sovereignty model that places controls where risks are deemed the highest while continuing to rely on global providers. Singapore’s green datacentre roadmap was also singled out in the report for balancing high-density compute expansion with net-zero emissions targets.
Conversely, South Korea was highlighted as a nation experiencing the friction of strict sovereignty policies. Through its Cloud Security Assurance Program (CSAP), South Korea has historically imposed strict local requirements that effectively restricted foreign providers from public-sector workloads.
“Korea stands out as an example of a country that is still pursuing the type of policies which we identify as those creating economic costs and trade-offs,” Worthington observed, though the report noted that Korea has recently begun to open some procurement to global providers.
While China is an obvious example of a successful, highly restrictive sovereign technology market, Worthington said the AI powerhouse was excluded from the research because its massive structural scale makes its model largely unreplicable for other APAC economies.
When asked if the report dismisses the need for national AI development, Worthington stressed that it does not oppose domestic AI strategies that support local investment and talent development, but rather the exclusionary policies that cut nations off from the global technology frontier.
Ultimately, the report suggested that for most APAC countries, sovereignty should not be defined by owning the entire AI stack, but by the agency to govern it. By blending global capability with local control, governments can protect their interests without sacrificing the economic boom promised by the AI revolution.
“Good policy will have different levels of safeguards required, depending on the use case,” Worthington said. “Set objectives and have a high-quality monitoring system that enables business operators – whether they be sovereign or non-sovereign – to figure out how those can be achieved most effectively, rather than trying to be very rules-based and prescriptive.”
This will involve the use of verifiable safeguards, such as data residency requirements, encryption with key management, and operator accountability, allowing countries to configure AI systems on domestic terms while maintaining access to the AI innovation and cyber security infrastructure of global hyperscalers.
