It is financed to a significant extent by an American AI laboratory, which has itself become one of the world’s largest investors in robotics. The company continues to be presented as a European success story. Yet its factory, its capital and, increasingly, its technological backbone have long been based elsewhere.

This is not an isolated case, but a pattern – and that pattern is the real news. It is currently being decided along which capability economic, industrial and, ultimately, military power will be distributed over the next twenty years, and that capability is called AI. Those who do not possess it will be left behind – even in the fields they consider to be their very own domain.

One front becomes two

Digitalisation was, for all the hindsight gloss, a lost single-front competition. There is no European search engine, no European social network, no European hyperscaler of any standing. This was nevertheless manageable because there was only one relevant external power – the US – and because Europe could afford simply to purchase its digital services. A win-win with a distinctly asymmetrical return, but at least one thing: Europe’s own industrial base remained untouched; China did not yet play a role in this equation.

When it comes to AI, this equation no longer holds true. Two systems are scaling up simultaneously, with fundamentally different organisational logics but comparable force – and, of all things, the very field to which Europe was able to retreat last time – its industrial base – is itself the battlefield this time round. Industrial manufacturing, robotics, mechanical engineering: no longer the safe haven of the past, but the next front line. The lost single-front competition of digitalisation threatens to turn into a war on two fronts, which this time cannot be lost as an afterthought – because there is no longer any space to which one could retreat.

Tech rules, money follows

Before we even consider capital, the chain of causality must be clear: it is not money that decides, but technology – capital follows leading technology, not the other way round. Anyone who believes that a missing leading position in AI can be bought retrospectively with sufficient investment and billions in funding is confusing effect with cause. Both are needed, but not in any order: first, a strategically decisive, truly leading technology; then the massive, orchestrated allocation of capital around it. Without the technology, the capital fizzles out to no effect. Without the capital, the technology remains small-scale and vulnerable. It is precisely in this interdependence that Europe is currently failing on two counts.

In the case of AI, this is particularly significant: anyone who understands the importance of data in AI technology will already know how pointless it is, in this sector, to adopt a strategy of ‘seriously’ waiting until some sort of ‘maturity’ sets in that makes a ‘ROI’ calculable. Furthermore, anyone familiar with this technology’s inherent learning capability will know that it is almost inevitably bound to become verticalised. Even now, larger models are absorbing smaller, specialised models on a quarterly basis.

Finally, anyone taking into account the geopolitical and military implications mentioned at the outset should try to ‘seriously’ calculate the ROI of the atomic bomb to recognise the significance of the issue.

Scale is indispensable

What unites the US and China, despite their opposing approaches to order: both accept losses as a prerequisite for scaling up, not as a warning signal to retreat. In the US, this means investing capital well before a guaranteed return is secured, financed by hyperscalers, on a scale that would be considered reckless in this country.

And that applies only to the capital side of language models; in robotics, there is a whole other league, because there, private American capital and Chinese state funds converge at a pace and on a scale that European investment and funding cycles are structurally unable to match.

In China, the same mechanism goes by a different name but operates according to the same logic: capacity is deliberately built up ahead of demand, squeezing competitors out of the market via the price mechanism and securing market share – with consolidation being a built-in, rather than an unintended, component of the strategy. This is not a new concept, but one that has been studied in industrial economics for decades: Strategic overcapacity as a barrier to market entry. What is derided as ‘overcapacity’ is, in the sectors that China’s five-year plans have for years explicitly designated as key industries – with state funds set up specifically for this purpose, amounting to the equivalent of around 140 billion US dollars, committed over twenty years – not an operational mishap, but the method itself.

And herein lies Europe’s real mistake – not a lack of capital alone, but a double standard in its assessment. The American willingness to accept losses is dismissed as a ‘bubble’ that will soon burst, intended to justify its own caution in hindsight. China’s willingness to incur losses is dismissed as ‘overcapacity’, as evidence of a ailing, misguided system. One’s own caution, by contrast, is regarded as prudence. Yet scaling up requires a willingness to incur losses, whilst one’s own loss-avoidance strategy is touted as a virtue. This is not an accurate analysis of competition, but a false rationalisation of one’s own, ultimately fear-based inaction.

And now comes AI – a technology in which the same principle is inherent as in economic and industrial scaling: scale is indispensable.

AI decides, not the ‘hardware’

The risk is most evident in the very area where Germany has historically felt most secure: mechanical engineering, particularly robotics. Yet this has implications for industry as a whole, for this is where the technological foundation for our vital manufacturing expertise lies. The widespread assumption is that manufacturing expertise – precision mechanics, decades of integrated know-how – acts as a moat. It was, as long as the machine’s intelligence lay in its mechanical construction. With AI-controlled robotics, it increasingly lies in the model, not in the metal. Anyone who lacks the cognitive layer is a supplier – however brilliantly the hardware may be crafted.

That does not mean that specialisation is fundamentally doomed to failure. There is a counterexample: ASML, the Dutch company with a global monopoly in EUV lithography technology, which has been unassailable for decades. But this monopoly is based on physical and regulatory barriers to entry that have grown over decades and which – so far! – no amount of capital in the world can replicate in the short term. And it is just one monopoly – a small one at that. In AI-driven fields, different metrics apply: computing power, training data and capital intensity drive economies of scale, not the art of physical manufacturing. Robotics sits precisely at this juncture – it looks like hardware, but is increasingly behaving like software.

This can be seen for oneself in the most prominent European examples, if one looks more closely – no, more honestly – rather than being content with the label ‘European champion’. A German robotics company has secured the largest funding round of any full-stack robotics provider worldwide – driven largely by American capital, with the computing platform of an American chip manufacturer serving as its cognitive foundation. A German-Chinese robotics company is developing its most advanced perception layer in partnership with an American AI laboratory. A flagship European language model, publicly touted as proof of digital sovereignty, is now tied to a multi-billion infrastructure deal with an American cloud corporation. And the factory in California mentioned at the outset: built by a company regarded as a flagship in Europe, with capital from an American AI laboratory which is also its main technology supplier.

Four independent cases, spread across several countries, from two different sub-sectors – and always the same pattern: money follows tech. The capital did not follow the European location, but rather the American technological leadership, which is what made this location investable in the first place.

Conclusion

This is not about whether Europe could have kept up at all under the given institutional conditions. This debate is a distraction because, in the end, it always comes back to the convenient excuse of doing everything on a slightly smaller scale, more slowly and more cautiously. The real demand is simpler and more uncomfortable: possessing frontier AI capability in our own right is not one option amongst many, but the prerequisite for having any bargaining power at all in any downstream field – including, and especially, in robotics, where this is currently playing out in real time.

The factory in California will not be the last instance in which a European flagship builds its operations elsewhere, as long as the real diagnosis is refused. In the end, AI decides. Not the metal. In operational scaling and in the technology itself, scale is what matters – and that is only possible if one is prepared to accept losses. Those who prioritise loss avoidance will lose everything.

About the author

Dirk Specht is a supervisory board member, lecturer and entrepreneur.

As a computer scientist and financial mathematician, he spent over two decades guiding the strategic direction of leading companies in the media, banking, insurance, retail and industrial sectors as a founder, board member and head of digital, before spending a decade in the media sector as editor-in-chief of Capital Online and as head of digital at the Frankfurter Allgemeine Zeitung.

Since then, he has served on the supervisory boards of companies in globally disruptive technology sectors, with a focus on digitalisation and AI. He also holds various teaching posts in economics and publishes regularly through his own channels, as well as acting as a guest author or speaker.