The geopolitics of AI: 10 poorly known facts for winning the AI race

AI is recognized by many as the next general purpose technology (GPT), with significant economic and welfare improvement potential for our societies. As once said by President Putin, it also provides a key " amunition to rule the world" in the 21rst century .

As such, the battle for control has started already a decade ago, with the emergence of China as a major global player, -- challenging the US hegemony in technology,-- and building strong leadership ahead of Europe as a whole.

US is continuing to expand and make sure it wins the race; Europe is also busy with architecting a synchronized plan among EU nation members, -after each of them trying to issue its own AI strategy so as not to be left behind.The current narrative is clear: China and US are fighting for leadership --and the rest is settling behind.

However, this narrative remains rather simplistic, and may benefit from a more detailed mosaic. We have used data collected by the JRC of the EU, as well as data from sources such as Atomico, and others, -- to document ten poorly known facts around the development of AI.

Ten little known facts worth noting about AI

1. After 10 years, a concentrated battle but with a diverse set of wins by AI theme. The current battle for AI remains very concentrated. Cross-checking multiple bases, there are about 0,02% of worldwide firms with a current play in AI, or less than 30,000 worldwide.

Only four to five countries can count more than 1000 firms in the field of AI- US, China, India, UK, and South Korea. In fact, as Table 1 shows, the concentration by AI domain suggests that it is a three, max 4- country play to date, that is, a rather skewed world dependence ( PM: number of countries of same size is found by taking the inverse of the Herfindhal index/100).

Table 1- The concentrated game of AI , Herfindhal index

computer vision 5130

autonomous vehicles 4280

Machine Learning 3850

NLP 3180

AI services 2810

Robotics 2080

But how concentrated is this? To give a perspective, the Herfindhal index of online search (a market led by Google) is 6000, while it is roughly about 4000 if you look at the US offline retail market dominated by Walmart. Geopolitically, the oil sand resource Herfindhal is in the range of 6000, and about 7000 for rare earth, whose production value chain is managed at more than 80% by China, see https://www.visualcapitalist.com/measuring-competition-valuable-minerals/

2. The chinese Achille's heel :China plays the patent and public institutional route- but private sector lagging. If China is rather strong in the race for AI, China is also the country with a systematic strategy of patenting, and with an AI ecosystem essentially driven by public institutions (Figure 1). In fact, China has a strategy of systemic patenting: Using the AI TES database of the European Commission JRC, it is found that about 4/5 of AI Chinese firms are filing patents.

However, there is a weakness attached to the strategy- the AI patent mostly protects the home market as the patents have limited geographical coverage, in contrast with other countries. For countries outside China, the patenting intensity is less, but for those commercializing and patenting, the breadth and geographical coverage of commercial patents seems much larger than in China.

About 10% of the AI universe is linked to public institutions in the world, but it is closer to 20% for China, Taiwan, South Korea ( and just below 30% in Russia). In contrast, the share of public firms is more like 5% in Anglo saxon countries like US, UK, Canada or still Israel.

Figure 1- The AI race is clearly tainted by institutional beliefs

Taking this together, US private firms are still ahead of Chinese private firms- and in the cluster of patenting firms, private Chinese firms are undeperforming versus many countries, eg Finland, Sweden, Ireland, Croatia, Japan, Canada, South Korea, Singapore, Taiwan, on top of US.

3. Europe has not lost the battle- it is fast moving into AI services. The narrative has been that Europe is losing the battle, but this narrative is both simplistic and inadequate- Europe has key assets, notably a good pool of talents (that could be allocated to AI), an adoption curve of AI technology by firms on the rise, and a strong public commitment towards the AI field, as seen in the leadership push by the European Institutions, among others (see also my recent post: https://www.dhirubhai.net/pulse/european-ai-where-we-now-jacques-bughin/).

While it is known that Europe has more assets into high-value added B2B and manufacturing, Europe has also a comparative advantage in robotics and automation-- but it also appears that Europe is getting ahead of competition in hosting AI services. With time, the value chain of activities typically increases with services, making Europe in a good position to benefit from the future growth of AI as a GPT (Table 2).

Table 2: market share of Europe by AI domain, last 10 years, %

Robotics 24

AI services 22

Machine Learning 10

Autonomous Vehicles 8

NLP 4

Computer vision 3

4. The return of the "Asian tigers?" . Even if the market is concentrated, the market is also diversifying, to the "Asian Tigers".

Japan had been a major player in AI, especially through electronics, but has possibly missed the new wave of AI. Still some of its companies are catching up and filing competitive patents. South Korea is actually in the top 5 countries, with some good wins in machine learning. Singapore is diversifying into AI services.

India is not in rest, with its ICT sector and Nasscom voluntarily pushing towards the association of India tech to diversify aggressively into AI services, as next growth areas of their global ICT service journey.

5. Brexit takes out a large chunk of European AI. Europe has not lost the battle to AI, but the leading part of European AI is in the UK.

While GDP of UK weights about 18% of the European Union, its market share of European AI is "twice" that in domains such as machine learning, robotics and services--( it lags only in autonomous vehicles- a domain of necessary focus for Germany, and Sweden to a lesser extent) (see figure 2). This is rather crucial for Continental Europe, with already a stagnating ICT sector, -as roughly 1/3 of the European AI economy can be outside the EU-27, after Brexit being consumed.

Figure 2- UK AI plays above its weight

6. Small is beautiful: Scandinavia and Baltic countries. Israel has rightfully been referred as a country that quickly embraced technologies. It is a country with one of the largest digital economy sector in relation to its GDP, and which leads in terms of AI firms per unit of R&D spent to date, with a concentration likely twice higher than the US.

Small European countries are not in rest- even if of course, their size may put them at disadvantage. Despite the push of large countries in EU-27, such as France, and Germany, AI companies from Scandinavia , including Finland, - or for the Netherlands- are already more integrated in their economies than in large countries. The Baltic countries, especially Estonia (and to a lesser extent Lithuania) are also showing some muscles. Those countries altogether may represent more in AI than say Italy, Spain and Portugal together.

7. European AI in the hands of small and young companies- in contrast to China. AI firms distribution is typically shaped, on one hand by large hardware, or manufacturing firms, and on the other hand, by digital darling in services, and other more nimble firms. China , including South Korea, are typically more focused on the electronics value chain, including computers, chips, sensors, etc; US is more linked to digital darlings, while Europe is more focused on professional and technical services. In that aspect, this reflects the comparative advantage being built up by Europe in services, but also the lack of momentum of the large ICT firms, at the excpetion of eg Siemens, in Europe.

Europe has surprisingly lots of young ( up to 30% and less than 5 years), small companies (3 out of 4), in contrast to China (barely 10% are new born firms-) and roughly 1 out of 3 is small). The closest country to this pattern in India, where the AI development is more bottom up too. This diversity of small firms may become a real asset to Europe in a few years when the ecosystem will be becoming more connected than today, through the leverage of the Single Market.

8. ICT leads but manufacturing rising. As for the beginning of any technology cycle, one observes in AI that a large portion of the activity comes from the ICT sector. In fact, while the ICT sector is roughly 5% of the developed economies (- it is 4% in the EU28-), the share of AI activities originated in the ICT sector is currently up to 6 times that, or roughly 1/3 of total activities ( see Figure 3). This figure is relatively large as typically the ICT sector contributes roughly 20% of economic growth and between 15 to 25% of innovative output in general.

Figure 3-- How the ICT sector has been driving AI activities to date

However, while the manufacturing sector has been retreating as a share of economic activity in many developed countires, AI and manufacturing is sizeable, in the range of 15% to 20% of activities in the US, 25% in Europe, and close to 45% in the cluster of AI countries of Asia.

This reflects the specialization of the ICT sector . While the ICT sector is usually composed at more than 80% by services in Europe and US , ( the small remaining part being manufacturing based), Asia remains a major player in the manufacturing of ICT, especially Taiwan, China or still Japan. Nevertheless, the share of AI allocated to manufacturing is not negligeable, and growing everywhere, reflecting the importance of AI for the future development of product value chains, and automation and robotics. In that perspective, there is an opportunity for countries to rebuild the sector of manufacturing, anc create new economic source of added value.

9. Europe of AI is Europe of regions. The European map of AI is scattered across countries and regions, - but the most interesting picture is that a large part of the AI research and development is collaborative in nature with 50% of AI developments connecting multiple regions (and the figures go up to 80% for EU project network, according to JRC data, and the AI TES database)-- in line with the proactive plan of coordination launched recently by the EU.

Further, regions are more the connectors than the countries themselves. As example, while France is only the third country in number of AI firms, the region of Ile de France is the most specialised in AI among the EU-28 (thus including the UK), and is the most centrally connected to all European regions.

France key centers include besides the Paris Region of Ile de France, the region of Rhone/ Alpes, known for its high value added manufacturing and equipment firms and pharmaceuticals. Italy relies on slightly more (but smaller) nodes, including the region of Roma (Lazio), Piemonte, Toscana, and the regio of Emilia- Romagna. Germany has the Bayern region, Berlin, but also incumbency regions, such as the Stuttgart region, Darmstadt, or Karlsruhe.

10. The European catch up is under way. Finally, while Europe is indeed squeezed between the US and China, a few important points make Europe on a possible interesting trajectory for its future:

a) We already mentioned the importance of the UK in European AI landscape. While this means that Brexit may add complication, Europe may have a large interest to secure good connection to the UK, epsecially around the cities of London, and the university heavy cluster of Oxford. The good news is that those cities understand the importance of international cooperations and will de facto pursue and likely intensify relationships with Europe

b) Europe has already multiple clusters that reflect its diversity. Paris, Stockholm and Amsterdam, Berlin and known clusters of AI, but others are burgeoning, eg Munich, Madrid, Helsinki, to quote a few.

c) Europe AI landscape is relatively composed on younger and small firms, demonstrating a new push to digital enterpreneurship in the regions. It also reflects that lage incumbenct firms are slow to play with those technologies too, but the contribution of AI from ICT and manufacturing is ahead of traditional AI leaders, meaning that Europe is shifting its resources for catching up. Small countries show the way, eg Finland and Estonia are redirecting resources at least at par, if not more than, the US and China do relative to their economies

d) Finally, Europe is cooperating. In particular, most R&D projects have an international cooperation dimension that goes against the theme of too much fragmentation in Europe. This is hopefully even more evident when it comes to EU funded projects, as demonstrated by the JRC in their AI Watch .

2. Concluding...

Those ten elements, I hope, are convincing enough to suggrst that the AI geopolitics will remain in force, until the AI race settles. At this stage, China has a headstart, but this is not as solid and pervasive as many claim-- US remains strong in technology ledership and the mosaic of European countries is possibly converging towards a cohesive ecosystem of powerful regions in growing domains such as AI manufacturing and AI services. We may conclude that this is early stage about the geopolitics and economics of AI. But the narrative that the game has been played is likely premature.

Additional References

Accenture, (2018), Realising the economic and societal potential of responsible AI

Accenture (2018), https://www.accenture.com/us-en/insight-artificial-intelligence-future-growth

Bughin , J and Ch. Pissarides, (2019), Don' sqander the AI revolution, Project Syndicate

Bughin, J, (2019), How to develop European AI start ups, Vox, https://voxeu.org/article/how-develop-enough-european-ai-startups

JRC, (2018), The 2018 Predict Key Facts Report. An Analysis of ICT R&D in the EU and Beyond

JRC (2019),AI Watch 2019 Activity Report

Venture Beat (2019), https://venturebeat.com/2019/11/20/deep-tech-investment-surges-across-europe-as-ai-ecosystem-expands/

(c) Jacques Bughin