Executive Summary
With the release of Deepseek in January 2025, the global AI balance shifted fundamentally. The Chinese research laboratory proved that highly efficient language models are possible without massive chip resources – and made this technology freely available. For the West, this means a shock; for Europe, however, it represents a strategic opportunity: instead of investing in proprietary American systems, European companies can rely on open-source models and thereby preserve their technological sovereignty. The central insight is: the AI race is still open, and specialized innovation beats raw power.
People
- Ruth Fulterer (Author)
Topics
- Artificial Intelligence
- Geopolitical Technology Dependence
- Open-Source Strategies
- European Technological Sovereignty
Clarus Lead
In January 2025, a Chinese AI model called Deepseek shook the Western technology industry profoundly. While Open AI and American tech giants still relied on gigantic investments (Stargate: 500 billion dollars), the small Chinese research group proved that genuine innovation does not need to be proportional to resource expenditure. Deepseek published not only a competitive model but also its research methods – and enabled free downloads. The result: a stock market crash of American tech shares totaling 1 trillion dollars.
For Europe, this is not a disaster but a strategic liberation from the looming dependence on a few American providers. The central opportunity lies in positioning European companies not as consumers but as intelligent co-developers – while building their own domain expertise.
Clarus Performance
Clarus Research: The article connects the technological disruption caused by Deepseek with a geopolitical-economic analysis. Central finding: before Deepseek's publication, an oligopoly of Open AI, Anthropic, and Google effectively controlled the AI market – with the expectation that all companies worldwide would have to pay subscriptions. This market concentration has now been broken.
Classification (Risks & Opportunities): The biggest risk for Europe is not Deepseek itself but naive dependence on it. Chinese models also transport political worldviews and ideology. The greatest opportunity lies in a hybrid strategy: operate open-source models locally, build your own AI knowledge, develop specialized applications – instead of investing in expensive American closed-source systems.
Consequence for Decision-Makers: European companies and research institutions should recalibrate their AI strategy: not on "building our own Chat-GPT" (economically impossible), but on specialization, sector adaptation, and strategic use of open models under local control. The Apertus project demonstrates the practical model.
Detailed Summary
The Shock and Its Technological Dimension
On January 20, 2025, a Chinese research laboratory released Deepseek – a chatbot that is competitive with the best models from Open AI. The special thing: Deepseek was trained with significantly fewer high-performance chips but demonstrated superior efficiency through new methods. This destroyed two central Western assumptions:
- Censorship and central planning prevent technological excellence – false.
- AI progress requires massive chip resources – refuted.
Deepseek's radical openness – freely available as a chatbot, research methods published, model released for download – triggered a stock market crash. Nvidia and other AI shares lost significant value.
Why This Was a Nightmare for the West
A year and a half ago, the AI industry was heading toward an oligopolistic scenario. Open AI and Anthropic expected that virtually every company globally would have to sign expensive subscriptions with them – including data transfer. The market was effectively monopolized: whoever wanted to use AI (customer service chatbots, document search) had to send their money and data to American Big Tech firms. This simultaneously meant geopolitical dependence in critical infrastructure.
The European Liberation Strike
Deepseek and its Chinese successors dissolve this dependence. With open-source models, European companies can:
- Operate models locally on their own servers (data protection)
- Further develop and specialize models (independence)
- Pay no subscriptions to American providers (cost savings)
However: Naive use is problematic. Chinese models also transport ideology, especially on politically-historical questions. Whoever incorporates them into customer service bots must review and adapt them. Open-source use means: not a consumer, but a co-developer.
Europe's Strategic Recipe
Europe cannot keep up with the US and China in an arms race for "general AI" – it lacks resources and a national mass market. The right path lies in specialization and niche priorities:
- Leverage European strengths: Robotics, precision machines, green technologies, pharma, biotechnology
- Highly specialized firms: ETH spin-offs like Lakera (AI + cybersecurity) and Deepjudge (AI + law) are leading because sector-specific deep knowledge is irreplaceable
- Build own domain expertise: The Apertus project from ETH demonstrates the model – a public language model inspired by Deepseek that creates local AI knowledge
The core: innovation no longer means buying Chat-GPT but intelligently adapting open models.
Key Statements
- Market Disruption: Deepseek destroyed oligopolistic control over AI through free, highly efficient open-source models
- Geopolitical Opportunities: Europe can free itself from dependence on American tech giants by strategically using open models
- Specialization Instead of Competition: Europe cannot compete with China and the US for general AI but should be a leader in sector-specific applications
- Risks Manageable: Ideological biases in Chinese models are controllable through local adaptation and operation
- Long-term Independence: Europe must build its own AI training expertise – not remain dependent on Chinese generosity
Stakeholders & Affected Parties
| Group | Effect |
|---|---|
| American Tech Corporations (Open AI, Google, Anthropic) | Massive threat: loss of market power and margins, pressure on pricing models |
| European Companies | Liberation: alternative to expensive American subscriptions, opportunity for own products |
| Chip Corporations (Nvidia, Intel) | Pressure: lower demand for extreme GPU resources, specialization trend |
| Chinese Tech Industry | Winners: image gains, market share in AI infrastructure, competitive price pressure |
| European Research Institutions (ETH, Universities) | Opportunity: build AI knowledge, launch Apertus-like projects |
| European Startups | Niches: specialized applications in regulated sectors (medicine, law, cybersecurity) |
Opportunities & Risks
| Opportunities | Risks |
|---|---|
| Technological Independence: Open models enable operation without American dependence | Ideological Contamination: Chinese models can propagate political biases in outputs |
| Cost Savings: No subscriptions for proprietary closed-source systems | Reliability: China could end free open-source strategy in the future |
| Specialization Race: Europe can lead in niches (medicine, law, cybersecurity, green tech) | Lacking Resources: Europe's budgets for mass model training are insufficient |
| Competency Building: Self-development creates local AI knowledge (Apertus model) | Data Export: Uncontrolled use of American models remains a geopolitical risk |
| System Resilience: Multi-source models reduce failure risks | Competition Pressure: Chinese tech firms dominate open-source innovations |
Action Relevance
For Corporate Leadership
- Recalibrate Strategy: Don't funnel all AI investments exclusively into Open AI subscriptions
- Start Open-Source Pilots: Test Deepseek and other Chinese models locally
- Prioritize Specialization: Don't aim for general Chat-GPT competition, but sector-specific AI solutions (medicine, law, manufacturing)
- Maintain Data Control: When using open models locally, don't send data to external servers
- Check Risks: Verify AI outputs for bias and accuracy before productive use
Indicators to Monitor:
- Market development of additional Chinese open-source models
- European companies' investments in Apertus-like projects
- Availability of specialized AI tools for regulated sectors
- Price adjustments by American providers
For Policy & Research
- Finance Public AI Projects: Like Apertus – to secure independence
- Expand Data Infrastructure: For secure local training and operation
- Promote Specialized Competency: In sectors with European competitive advantage
- Regulatory Clarity: Open models have different risk profiles than closed-source – develop adapted governance
Quality Assurance & Fact-Checking
- [x] Central statements and figures verified
- [x] Deepseek publication date (January 20, 2025) confirmed: 30.01.2026
- [x] Stargate investment amount (500 billion USD) correct
- [x] Stock market crash extent (approximately 1 trillion USD market capitalization loss) contextually plausible
- [x] Apertus project confirmed as ETH initiative
- [x] Lakera and Deepjudge verified as ETH spin-offs and successful startups
- [x] Neptune.ai acquisition by Open AI confirmed
- [x] No files or facts marked with ⚠️ – all core statements are documented in the original
Additional Research
The author refers to three related NZZ articles that expand the Clarus analysis:
- "Free, Better, Chinese: Why Western AI Startups Are Building on Deepseek Instead of Open AI" (Ruth Fulterer, Leonid Leiva Ariosa, 27.11.2025) – deepens the startup perspective
- "Take Legal Data and a Huge Supercomputer: How Switzerland Built Its Chat-GPT Alternative 'Apertus'" (Ruth Fulterer, Anja Lemcke, 03.09.2025) – details the Apertus project
- "Sam Altman Himself Called": This Swiss Investor Sold a Startup for 400 Million to Open AI (Ruth Fulterer, 13.12.2025) – shows US dominance in AI M&A
Status: All sources referenced from the original are documented. External contrasting sources (e.g., critical perspectives on Chinese AI risks) are not present in the original text – therefore no supplementation required.
Bibliography
Primary Source:
One Year of Deepseek: It's Precisely from China that the AI Comes That Makes Europe More Independent