- within Technology topic(s)
AI innovation is an essential strategic vector for competing nations. Without it, AI would still languish in the "winter" of the 1970s and 1980s.
However, history has shown that innovation is less than half the story, with adoption its often-neglected counterpart. This imbalance is true despite adoption having vast potential as a source of economic growth in its own right – even when the innovation may have originated elsewhere.
This lopsided focus is already evident in national AI strategies and needs urgent correction if the full potential of recent AI innovations is to be realised.
Innovation is a global two-horse race
Framed as an existential race to prosperity or doom, formidable AI innovation ecosystems have formed, comprising big tech, startups, investors, elite universities, media, consultancies, and government. These rarified networks have created their own AI celebrities who are now household names, some transcending even national boundaries to become global brands.
The US still leads the AI innovation rankings, but China is closing fast. Singapore, the UK, and France complete the top five, but if China and America are competing in an innovation marathon, everyone else is enrolled in the concurrent 10K.
In this secondary race, where specialisation is key, the order of the top ten could be quite different even by the end of 2025.
Position in AI innovation rankings hinges on access to talent, infrastructure, research, investment, and government incentives. Considerable digital ink has been spilt on all these matters as countries debate, worry, double down, and hope they have placed the right bets. The publication of America's AI Action Plan in July was only the latest example of these machinations.
AI adoption is a greater economic opportunity
Despite the huge strategic focus on AI innovation, it is not the most important race for national prosperity: AI adoption is the biggest economic opportunity that very few are talking about.
It requires the unenviable task of applying AI innovations in the real world. This involves mainstream companies with finite investment capacity, executives and boards still getting up to speed, messy technical legacy, and the old-fashioned need to keep making a profit.
The mission is not glamorous and struggles to gain media attention – unless a big-tech CEO predicts the imminent end of human work (again). There are no celebrities of AI adoption or equivalent ecosystems to achieve it. More to the point, enterprise-wide adoption is proving extremely difficult, as outlined in MIT's recent assessment of AI in Business. This study reports a 95% failure rate for embedded or task-specific generative AI tools, and many other surveys have found high abandonment rates of generative AI projects even before they go live.
Further evidence of a lack of focus on AI adoption is that the relevant data is, literally, all over the map. Even the newly-minted ChatGPT 5 produced a league table of nations on the subject that missed China and the US entirely. Manual analysis on AI adoption (the irony) discovered the leading countries to be Denmark, South Korea, Sweden, Belgium, and Japan (in that order), with "corporate adoption rates" of between 24-30% – whatever that really means. The UK achieved 21%, with the rest of Europe in the mid-teens. China and India were harder to pin down but could be fractionally ahead. The US has multiple conflicting data points but is, at best, only part of the leading pack, which means that no country has yet achieved anything like critical mass adoption of AI.
This matters because the AI adoption prize is truly glittering. According to various converging studies, comprehensive adoption could boost productivity by 0.5-1.5% annually, which equates to 2-4% of GDP. Even at the low end of the range, this potential boost is staggering given that most advanced economies currently have annual growth rates below 2%. In absolute terms, the mid-range benefit equates to approximately USD 7 trillion of annual growth by 2035. In contrast, the direct economic contribution of AI innovation companies was a modest USD 300m annually in 2024 and is expected to rise to no more than USD 2 trillion by 2035.
Of course, countries don't need to choose between adoption and innovation – ideally, one should reinforce the other in a virtuous cycle. However, research involving Microsoft suggests that current AI product design is actually hindering real-world usage. This needless lack of alignment harms the entire AI landscape.
Getting the best bang for the buck
History provides both warnings and encouragements on technology adoption.
Gunpowder was invented in 9th-century China by alchemists seeking immortality. Ironically, the result was better suited to reducing lifespans, and the Chinese quickly realised its military potential. By the 13th century, gunpowder know-how spread westward along the Silk Road and reached Europe in the late Middle Ages.
By then, European countries were engaged in almost continuous warfare, and gunpowder proved decisive in castle bombardment and siege. European engineers improved metallurgy, cannon design, and even the gunpowder formula itself, creating lighter and more accurate firearms. By the 16th and 17th centuries, a "gunpowder revolution" had transformed European warfare, enabling naval dominance and the projection of force overseas. China never fully adopted and adapted its own innovation, which became evident in the 19th-century Opium Wars.
Gunpowder is far from an isolated example. China has since turned the tables by perfecting high-speed rail, first developed in Europe, and is now doing the same for EV production using technologies invented in the US. History shows that adoption and adaption are enough to win geopolitically, even if the innovation is borrowed from another place.
How to win at adoption and adaption
This task requires a distinctive ecosystem and decisive government leadership. Singapore's National AI Strategy 2.0 is the best real example, which has moved "from principles to practice." The framework provides detailed implementation advice to private sector companies, including a compendium of use cases and a job transformation guide. Yes, this is in the context of a small and centralised city-state, but it is a good starting point for any country.
To that end, here are seven concrete proposals and some candid thoughts for the various ecosystem players:
1. Governments need to take charge. Only the state has the convening power to establish an ecosystem and set the right goals. This requires AI adoption to move out of the shadow of AI innovation with dedicated leadership and resources.
2. Regional universities with strong business schools naturally align with AI adoption. Only engaging the same small group of prestigious institutions that innovation relies upon reduces capacity, leverage, and coverage. AI has a long history of advancing creatively outside elite academic settings, and that should be encouraged to continue.1
3. Consultancies need to return to their original purpose and become brokers of proven AI practice for the benefit of clients. Focusing on practical solutions to achieve measurable outcomes is the only survival path for the industry in a post-AI world.
4. AI product companies need to start caring more about adoption for their own long-term survival. History is littered with defunct tech companies that relied on short-term, hype-driven success. The necessary shift requires more honesty about technical maturity and active listening to customers about what isn't working.
5. Business participants cannot always be the top national brands typically invited to advise governments or the startup entrepreneurs who dazzle officials. Adoption needs mainstream businesses that are more median than maximum. The ecosystem must work for the majority, not a select decile.
6. The media will no doubt continue to report every pronouncement of the AI celebrity circle, but it would also be helpful to hear about AI situations that sound more like the real world and see journalists holding big tech to account for their many claims.
7. Regulation must emerge from within AI ecosystems rather than be imposed from outside. Disembodied governance tends to bite in all the wrong places, hindering progress and proving ineffective in managing risk.
I have expressed these ideas in national terms, but AI technology is virtual and borderless, created by collaborating international researchers. Many parts of the innovation and adoption ecosystems are also global, so not everything has to be reduced to national competition. As such, organisations like the WEF have an important role as conveners of global AI for the common good.
Conclusions
To access this available value, AI adoption and adaption require well-resourced ecosystems convened by governments. Leaders who are not motivated by publicity or short-term political advantage must emerge to do the hard work for the long term. Significant behaviour changes are required in all parts of the current AI landscape.
This endeavour is not only a contest with other nations but also a race against time. It will take about two more years for generative AI, and especially its agentic forms, to mature, and countries need to get their act together before then. Otherwise, they will spend their time and resources firefighting risky outcomes from poor adoption methods of ever-more consequential technologies.
The time to act is now.
Footnote
1. For example: Brandeis (USA), Sussex (UK), KTH (Sweden), and the ISI (India).
Originally published by Precursions
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
