The convergence of historical challenges in the technology industry, unresolved issues at the union level, and tensions in the Indo-Pacific region risk compromising European sovereignty over AI by Andrea Monti – initially published in Italian on Formiche.net.
The development of machine learning systems requires synergy between GPU (graphics processing unit) design and parallel computing architecture. It also requires the ability to produce hardware, software operating systems that effectively utilize these components, access to a vast amount of information to build training datasets, and high-performance networks along with enormous amounts of energy to power the system as a whole. Consequently, the hardware-software combination (GPU, parallel computing architecture, and software utilizing them) is a determining factor in the development of AI-based products and services.
The Geography of Manufacturers
On paper (excluding Huawei and other Chinese manufacturers who will have a difficult time competing in Western markets in this sector), the five most important Western, or rather, American manufacturers vying for supremacy are IBM, nVidia, Alphabet (Google), AMD, and Intel. Amazon, Microsoft (which has initiated a collaboration with AMD), and other companies have also entered the sector, but only one of them, however, plays a role of particular criticality not only from a market perspective but also from a geopolitical one: nVidia.
Through a well-planned strategy targeting the gaming market, nVidia has developed not only extremely powerful graphics processors but also the accompanying software ecosystem to fully leverage their capabilities. This has made nVidia technologies essentially the industry standard. When the AI hype began, the company was well-prepared.
The Geopolitical Role of TSMC
Like its competitors (except for IBM, which announced a partnership with Japan, and Intel), nVidia has entrusted the construction of its chips to the Taiwanese giant, TSMC. However, unlike others, nVidia, despite being a Californian company, is “governed” by CEO Jensen Huang, a native of Taiwan who has recently reaffirmed the desire to continue commercial relations with his homeland, aiming to turn it into a global hub for AI. While South Korea is also heavily entering the chip production sector for AI, nVidia teaches us that machine learning is not solely dependent on silicon; it also requires software adopted by a wide user base. Therefore, until the effects of Japanese-Korean competition (if they occur) are seen, TSMC, and thus Taiwan, remain the single point of failure for the entire AI-based product and service ecosystem.
If Taiwan falls, AI falls, and with it, investments, strategies, and technological superiority.
The Escalation of the Taiwan Issue from a Regional Dispute to a Global Concern
This state of affairs elevates the Taiwan issue from a regional dispute to a global problem because the return of the island under the full political control of Mainland China would automatically give Beijing a powerful bargaining chip, not only with the United States but with the entire Western world.
It is clear that such a prospect would be difficult to accept and, in terms of realpolitik, it would provide a more concrete and internally “digestible” justification for the direct involvement of other political entities in the long-standing issue, compared to the fragile slogan of respecting the right to self-determination of peoples.
Therefore, it is not unreasonable to consider the geopolitical significance of the major US operators’ choice to strengthen ties with Taiwan instead of diversifying production elsewhere, perhaps in Europe.
This introduces the second point of this analysis, which is the substantial absence (or inability) of the European Union to carve out a role in a sector that is considered strategic not only from a technological perspective but also in terms of international relations.
The EU’s Strategy of Chronic Inefficiency
Despite the joint declarations issued by the Trade and Technology Council on May 31, 2023, it is clear that when it comes to AI, and more, the relationship between the US and the EU is completely unbalanced in favor of the former. The US possesses the technology, the means to realize it, and the market to sell it.
Instead of accelerating the adoption of industrial choices aimed at creating an internal and autonomous system for the design and production of semiconductors and related technologies, the EU has chosen to rush through a regulation on AI, a subject it does not control, and that concerns assets held by non-EU entities.
This applies not only to technologies but also to data, which are essential for effectively training machine learning systems.
The controversies that arose following the public availability of ChatGPT, but which actually concern issues that have been open for at least twenty years, have highlighted that running such a service required using data that is factually (though not necessarily legally) available on online resources accessible to anyone. There may be concerns regarding “privacy,” intellectual property, and “information destabilization.” On the other hand, the fact that these contents are accessible from anywhere in the world allows any state, especially those that have legislated to not recognize the intellectual property of non-friendly countries, to use them without being accountable to anyone. Consequently, they gain a competitive advantage over EU member states, whose companies will be burdened with compliance requirements for the use of such data.
Thus, in addition to timidly starting to consider the isolation of European networks in terms of transport, there should also be an assessment on the need to prevent indiscriminate access to data stored on EU resources that are freely available even outside the external borders.
Such a choice, which would definitively Balkanize the geography of transport networks and the content conveyed through them, could have unimaginable consequences, even more severe than mere compartmentalization of cables.
Instead of focusing on these difficult-to-resolve issues, the upcoming EU regulation (assuming no hiccups occur, the European Parliament will vote on it on June 14) is designed, once again, to build a bureaucratic and convoluted system, disconnected from the needs of protecting the national interests of member states, and motivated more by archetypal fears than concrete facts.
The Role of Nation States in AI Geopolitics
Another side effect of the design of the EU regulation on AI will be to confront member states with the choice of whether to allow Brussels to also regulate defense and security matters, which the EU Treaty excludes from Union competence.
The issue is extremely complex, and the needs arising from the Russo-Ukrainian conflict have already highlighted the problematic relationship between member states and the Union on such matters. Nevertheless, it remains a fact that AI-related issues no longer concern “merely” market development and consumer protection; they directly involve the strategic interests of each nation.
Therefore, it would be desirable to consider an approach based on a “dual-track” system that, while respecting the division of competencies, allows individual states, including Italy, to integrate decisions on AI into more comprehensive foreign and international policy choices, while leaving the EU to harmonize rules within its specific areas of competence.