First KUKA, now EBM-Papst: Europe is selling the technology meant to cool its AI era

What happened on 17 August

Madison Air Solutions, a Chicago group with more than 30 brands, is acquiring EBM-Papst of Mulfingen in Hohenlohe. Enterprise value: 5.1 billion euros, around 5.4 billion dollars. The three owning families Ziehl, Sturm and Philippiak are giving up their shares. Madison, founded only in 2017 and listed on the NYSE in April 2026 with a 2.2 billion dollar IPO, is buying a company it previously used as a supplier itself.

What is changing hands here is no niche mid-sized firm. EBM-Papst is the world market leader for electric motors and fans: 2.4 billion euros in revenue, around 13,800 employees, some 5,800 of them in Germany, more than 250 million fans installed worldwide — and over 1,200 patents that change owner with the deal.

The assurances sound reassuring: Mulfingen remains headquarters and an important research, development and production site, collective agreements and works council agreements continue to apply, CEO Klaus Geißdörfer stays in office. Exactly these assurances were given once before, in 2016.

Why fans are suddenly strategic

The reflex to wave away anything labelled “ventilation technology” is understandable — and wrong. Because the physical bottleneck in building AI data centres has long since stopped being the chip. It is the heat the chip produces. The more densely AI accelerators are packed, the more cooling capacity decides how many servers fit into a given footprint and how much energy operations consume.

EBM-Papst has developed the AI-supported NEXAIRA platform for precisely this market: demand-controlled air management, operating data analysis, predictive maintenance. According to the company, its cooling processes can save up to a sixth of a data centre’s total energy consumption. Madison puts the additional market opened up by the acquisition at around 30 billion dollars. The buyer knows exactly what it is buying.

And Europe? The European data centre market is growing from 47 billion dollars (2024) to a projected 97 billion dollars in 2030. Annual investment is rising from 25 to 28 billion euros to 35 to 40 billion euros. A hyperscale campus today costs 8 to 10 million euros per megawatt — five years ago it was 6 to 7 million. The premium is largely down to cooling and power distribution. So Europe is building the most expensive infrastructure of its recent history over the coming years — and simultaneously giving up control over one of its key components.

The KUKA lesson: assurances are not a site guarantee

In 2016 the Chinese group Midea took over the Augsburg robot manufacturer KUKA. Back then too the message was: site secured, independence preserved, jobs protected.

Ten years later the balance sheet looks different. In Shunde near Foshan, Midea has built a second KUKA headquarters including development and production, which now manufactures thousands of robot arms a year. While the Chinese sites were continuously expanded, little flowed to Augsburg; in 2023 a hall there was demolished. The humanoid robot Miro U was developed by Midea engineers in China — bypassing the Augsburg headquarters. Decision-making power today sits in Foshan. KUKA revenue fell from 4.1 to 3.7 billion euros. WirtschaftsWoche sums up the finding unambiguously: the Augsburg site serves “mainly as a fig leaf”.

The pattern is what matters, not the buyer’s country of origin. Formally everything stayed as promised: site, brand, company sign. What was relocated is what appears in no contract — the future business, development authority, the decision about where the next generation is created. Whoever is the European subsidiary in a group with more than 30 brands no longer decides for themselves which market gets supplied first when things get tight.

What was relocated is what appears in no contract — the future business, development authority, the decision about where the next generation is created.

And when things get tight, they get political

This is where the geopolitical dimension comes in that nobody had on the radar in 2016. In February 2026 the United States tightened its semiconductor export controls against more than 40 countries — allies among them. The precedent stands: technology under US corporate control can become the object of American export policy.

Nobody is claiming that fans will be on a control list tomorrow. But the relevant question is not whether that is likely today. It is whether Europe wants to voluntarily place itself, in a strategic bottleneck market, in a position where the answer is no longer decided in Europe.

The gap in Europe’s protective shield

The good news: the EU has reformed its investment screening. In February 2026 the Union institutions agreed on a new FDI screening regulation with binding minimum harmonisation. It closes the so-called Xella gap and will in future also capture takeovers via EU subsidiaries of non-European investors.

The bad news comes in three parts.

• First, the delay. The regulation enters into force in summer 2026, but its substantive rules only apply from the beginning of 2028. Some 18 months in which transactions like this one run under the old law.
• Second, the categories. Mandatory screening will apply to semiconductors, quantum computing, AI, dual-use goods, critical raw materials and transport, energy and digital infrastructure. Cooling and air technology appears explicitly in none of these categories — although it is the physical precondition for an AI data centre running at all.
• Third, the competence. Only member states may approve, prohibit or impose conditions. The Commission can issue opinions — nothing more. A European perspective on a European dependency therefore has no leverage.

Europe protects the chip and overlooks what keeps it alive.

Not an outlier, but a trend

And the case does not stand alone. In 2025, 96 billion euros of foreign investment capital flowed into Germany, twice as much as the year before, spread across 1,233 transactions with a total volume of around 118 billion euros. Covestro went to Adnoc, Ceconomy to JD.com. EBM-Papst is not an outlier but the continuation of a trend.

What needs to happen now

Regulatory clearance for the deal is still pending. That is the window.

• 1. Actually screen this case. Not as a formality, but with the question of whether cooling technology for AI data centres is a critical technology within the meaning of investment screening. Conditions are possible: binding site and development commitments with a term and a sanction, licence guarantees for European customers, reversion rights to relevant patents.
• 2. Extend the technology list. Thermal management and air technology for high-performance data centres belong in the catalogue of technologies subject to mandatory screening — before the regulation goes live in 2028, not after.
• 3. Close the transition period. An 18-month gap during a phase of global competition for AI infrastructure is not a technical detail. It is an invitation.

Noticing it in time

This is not about preventing foreign investment or distrusting American buyers. It is about Europe stopping the practice of selling its industrial bottleneck technologies one at a time and one contract at a time — and only noticing afterwards that it would have needed them.

With KUKA, it was noticed afterwards. With EBM-Papst, it is not too late yet.