(This article is authored by Outside the Box; published with authorization from Titanium Media.)
Text | Beyond the Page, Author | Drawing
With AI competition reaching this point, even Germany can no longer sit idly by.
Recently, German Chancellor Merz welcomed NVIDIA founder Jensen Huang in Berlin. Representatives from German industrial giants such as Siemens, Mercedes-Benz, and Audi were all present, and the conversation centered on one topic: how AI is entering factories and transforming manufacturing.
Behind this scene lies Europe’s deep-seated anxiety.
Two hundred years ago, Europe reshaped human destiny with steam engines and internal combustion engines, but in the age of the internet and semiconductors, the power to set the agenda has gradually shifted into U.S. hands.
Today, the AI wave is sweeping the globe: the United States is reaping substantial benefits from chips and models, China is integrating AI into industrial applications, while Japan and South Korea hold dominant positions in the upstream segments of the semiconductor materials and storage supply chains.
Only Europe seems to be trapped in the glory of a bygone era.
I. The “Lost Thirty Years” of the Industrial Heavyweight
Europe’s current decline was already foreshadowed decades ago.
In the 18th century, the British ushered in the factory era by harnessing steam engines; in the 19th century, the Germans, with Siemens’ electric motors and BASF’s chemical laboratories, built a tightly integrated industrial empire linking research labs to production workshops. This mechanism for rapidly translating scientific research into marketable products was once Europe’s greatest source of confidence.
While the United States, fueled by venture capital in Silicon Valley, gave rise to Google, Microsoft, and Amazon, leveraging a unified domestic market to rapidly roll out technologies worldwide, Europe has instead been mired in prolonged internal friction. It boasts one of the world’s most stringent legal frameworks, a fragmented market comprising 27 member states, and a capital system that is far from tolerant of trial and error.
The Dutch company ASML has built the most advanced extreme ultraviolet lithography machines today, but former CEO Peter Wennink loudly warned before stepping down that Europe is losing its spirit of innovation, as people here have grown accustomed to navigating within existing regulations rather than forging the future.
This disconnect reached a peak during the generative AI boom.
II. The Four-Nation Drama on the Global Chessboard
If we break down the AI industry, this global competition has never been a contest of any single technology; at its core, it is a series of brutal battles for control over supply chains.
The U.S. operates a dominant, closed-loop ecosystem. NVIDIA sells GPUs, TSMC provides foundry services, Microsoft and Google build data centers, and OpenAI and Anthropic tackle large-scale models.
NVIDIA’s real threat isn’t just how expensive its GPUs are; it lies in the software ecosystem built around CUDA. The more developers have accumulated over the years—code, development tools, and low-level optimization expertise—the higher the cost of migrating to other platforms. As a result, chips are no longer merely hardware products; they’ve become the gateway to an entire computing ecosystem.
Japan may not yet master the most high-profile large-scale AI models, but it holds a crucial position in several indispensable foundational manufacturing stages. Shin-Etsu Chemical’s semiconductor-grade silicon wafers, Tokyo Electron’s coating and etching equipment, and Advantest’s chip testing systems together form the infrastructure of chip manufacturing.
Japan has not abandoned its own AI industry; instead, it has embedded its industrial strengths into the mid- and downstream segments of the supply chain—core links characterized by high barriers to entry and substantial switching costs.
South Korea, meanwhile, staged a dramatic comeback powered by AI.
In 2023, the global memory chip market experienced a severe cyclical downturn, placing significant profit pressure on SK Hynix. However, despite the shadow of substantial losses, the company continued to invest in R&D and capacity expansion for high-bandwidth memory (HBM).
As AI accelerators drive an explosive demand for high-throughput memory bandwidth, SK Hynix, leveraging its first-mover advantage in HBM3 and HBM3e, has become a key supplier to chipmakers such as NVIDIA. By 2026, its market capitalization historically surpassed USD 1 trillion.
China, meanwhile, has demonstrated the dual driving forces of cost reduction in inference and scenario-specific engineering. At the beginning of 2025, the sudden emergence of the DeepSeek R1 model dealt a blow to Silicon Valley’s computing power camp. With an extremely low price and an ingenious architecture, it shattered the barrier that only trillion-dollar budgets could build leading models.
Meanwhile, the declining cost of model deployment has further energized China’s vast manufacturing ecosystem, providing AI with a rich testing ground and enabling proven solutions to be rapidly replicated across more production processes.
By contrast, in Europe, despite the presence of ASML’s lithography machines and France’s Mistral open-source model, the landscape remains fragmented.
3. Too many rules, too little money
Europe has, in fact, tried to save itself before.
Since 2024, Mistral has successively secured a €15 million strategic investment from Microsoft and integrated its large model into the Microsoft Azure cloud platform. Subsequently, Silicon Valley giants such as NVIDIA have also followed suit.
The greater embarrassment lies in Europe’s prideful factories.
According to the EU’s official plan, the government intends to allocate tens of billions of euros to build “AI superfactories” across member states, aiming to use public funding to help companies bridge the computing power gap.
However, when technical solutions are actually implemented at the factory level, engineers often encounter a more fundamental digital divide. Many small and medium-sized factories, despite possessing highly precise machine tools, still struggle with the lack of unified underlying industrial control protocols (such as OPC UA), outdated equipment that lacks standardized data‑acquisition interfaces, and pervasive data silos.
Data released by the European Commission vividly illustrates this gap.
As of 2024, only about 8% of companies across the European Union have truly integrated AI technologies deeply into their core business processes. According to OECD data, among European manufacturing firms, fewer than 10% actually apply AI to their core production lines.
IV. Jensen Huang’s Business Calculations
It was precisely in this atmosphere of passivity and anxiety that Jensen Huang arrived in Berlin with NVIDIA’s solution.
NVIDIA announced an additional investment of nearly $10 billion in Europe, targeting robotics and medical AI startups in Germany and partnering directly with Siemens and Audi to promote physics-informed AI across the country.
Take the industrial simulation platform jointly developed by Siemens and NVIDIA as an example: companies can build digital twin models of production lines in a virtual environment, test production processes and equipment configurations, and thereby significantly reduce trial-and-error costs before actual deployment.
For Germany, which has seen its GDP growth remain sluggish for years and is struggling under the weight of high energy costs and export pressures, leveraging AI to boost total factor productivity is a crucial option for maintaining its industrial standing.
However, this collaboration also raises new issues: if Europe’s industrial sector bases its software simulation, robot control, and algorithm training on NVIDIA’s systems, NVIDIA will gain not just chip sales revenue but could potentially evolve into the foundational platform for industrial systems.
V. Escaping Silicon Valley
Europe has finally come to the sober realization that competing head-to-head in the race between general-purpose large models and sheer computational power is not where its strengths lie. It is now placing greater bets on its areas of expertise: industrial automation, healthcare, precision machinery, and scientific computing.
Dutch lithography machines, German robotic arms, and French aerospace—these industrial secrets, honed over centuries—are real-world data that Silicon Valley simply cannot replicate in a short time, no matter how much code it writes.
Only by transforming these industrial insights into trainable data and scalable commercial products, and by breaking down compliance and market barriers among the 27 member states, can Europe truly turn its legacy assets into a new growth engine.
In the second half of the global AI competition, the action has already spread from the virtual digital world to the real physical world.
Jensen Huang has used computing power to open the door to Europe, but whether Europe can regain its footing depends on whether it can carve out an irreplaceable role within Silicon Valley’s computing ecosystem.
[Off-Page] says:
The technological revolution never ranks people by seniority.
The advantages accumulated in one era may serve as a starting point in the next, or they may turn into inertia.
The history of industry will not reserve a seat for anyone.
For more exciting content, follow TMTPost on WeChat (ID: taimeiti), or download the TMTPost App.