Europe quickly needs to build a better DeepSeek

The development and deployment of large language models (LLMs) have increasingly shaped the global race to harness AI for both civilian and military applications. 

One of the most recent (and well-publicised) entrants is DeepSeek, a powerful AI model credited with offering highly competitive pricing and operational efficiency - factors that have placed it front and centre in debates about AI sovereignty, defense capabilities and corporate innovation. 

While its benefits are touted, the open-source model, which stores all data in China, is also considered a significant security risk. Over the past ten years, the US has increased regulations on the use of Chinese software and technologies in the United States to control the flow of data to China.

Banned by NASA

NASA and several state governments prohibit employees from using DeepSeek and Washington state has drafted legislation banning the use of the platform. While federal lawmakers are introducing a bill to restrict DeepSeek nationwide. Meanwhile, Australia, Taiwan, the Netherlands, Japan and Canada have banned DeepSeek for use in government agencies and contractors. And, Italy and South Korea have imposed nationwide bans. 

DeepSeek’s security weaknesses were exposed in a recent study conducted by Cisco. It tested the platform’s internal guardrails. Using a dataset of 400 different harmful or unethical requests – from cybercrime to false information and illegal activities, the test revealed a 100% success rate in circumventing safety mechanisms. This contrasts with the U.S.-based OpenAI, which successfully identified and stopped such requests. 

Despite these vulnerabilities, DeepSeek’s competitive advantages may help accelerate the broader deployment of LLMs in the EU at a time when the bloc urgently needs to bolster both its military and technological self-reliance. At the heart of Europe’s push for autonomy is the recognition that the global political and economic landscape is in flux. 

With the war in Ukraine, the US pivoting to other strategic priorities, and China rapidly advancing its AI capabilities, EU policymakers and businesses alike are confronted with a pressing need to invest in cutting-edge technology that supports defense preparedness and underpins industrial competitiveness. 

Enter DeepSeek, which, despite hailing from a cheaper and more flexible regulatory environment, may propel Europe to stand on its own two feet (0.61 metres). 

The Chinese are known for their speed and agility in reproducing Western technology, but Europe, too, is known for replicating rather than innovating.

Using the concept of DeepSeek, a model based on cost-efficiency, open-source architecture, and flexible deployment options, European entities can create a better mousetrap. In doing so, they can leapfrog many existing hurdles to AI adoption, including the burdensome cost of infrastructure and the complexities of multilingual documentation for defense equipment.

Competitive Pricing Analysis 

Central to DeepSeek’s appeal is its competitive pricing and efficiency. DeepSeek trained its AI model for less than $6 million, leveraging roughly 2,000 chips. By contrast, Western companies—particularly U.S. tech behemoths—often require tens of thousands of chips, with associated training costs ballooning far beyond DeepSeek’s modest outlay. 

Meta, for example, reportedly used upward of 16,000 chips for similar tasks. This cost disparity is further amplified by the regulatory environment in which DeepSeek operates. China’s more flexible AI rules allow its companies and research groups to iterate rapidly and optimize models without facing the intricate compliance burdens that characterize EU or U.S. markets.

Is China the only place where competitive AI pricing can thrive? The answer is complicated. While China benefits from a combination of lower labor costs, extensive talent pools, abundant data, and fewer constraints on data privacy, other emerging markets could copy this approach. 

Yet the European Union’s AI Act and the General Data Protection Regulation (GDPR) impose strict stipulations on AI governance, data usage and transparency—raising costs and slowing down rapid prototyping. 

Opportunities and Market Impact 

Beyond the financial considerations, DeepSeek’s ushering in of a new wave of AI-driven innovation could hold transformative potential for several sectors within Europe. Topping the list is the defence industry, which is under strain to modernize quickly. 

The ongoing war in Ukraine, shifting international alliances and the awareness that Europe must bolster its own capabilities, collectively fuel a demand for secure AI solutions. One key aspect, often overlooked by the public but critical for readiness, is documentation and language support for new military equipment. Europe’s multiple languages should not be a barrier to high-tech military collaboration, yet the production of well-written and accurately translated manuals can be a formidable challenge.

A Copenhagen-based company has already built HumanAI, an LLM solution utilising humans and AI to provide multilingual translations securely. Frederik Pedersen is CEO and founder of EasyTranslate. He’s seeing increased demand from Europe’s defense industry.

Recently, we were contacted by a European military supplier who needs an advanced, but fully secure translation-based AI engine for defense purposes. 

Europe must be self-sufficient and there’s no reason to rely on US companies for these solutions. We now have a model that competes with those in the U.S. It can be used in closed environments with maximum security – which is what the defence industry requires.

This, Pedersen says, is where opportunity lies to build a better version of DeepSeek – with all the benefits of an open-source foundation but within a high-security, cyber-crime-proof environment -critical for defense applications. 

With ultra-secure, AI-based translation engines, militaries could efficiently produce accurate documentation in all necessary European languages, thereby ensuring consistent understanding and usage across the continent. Mimicking DeepSeek’s capacity to handle large, specialized datasets and integrate domain-specific language quickly could make a future, European model an indispensable tool for managing complex logistical operations and maintenance manuals.

DeepSeek could not only complement but also accelerate EU-centered AI initiatives. Combining the efficiency of models such as DeepSeek with human quality assurance and translation software, may become increasingly mission-critical, especially in high stakes environments such as defence and healthcare.

But it is not only the defense sector that stands to gain. Corporate enterprises with proprietary or sensitive data could also gain from a DeepSeek-like model, as it would allow them to maintain oversight of their AI infrastructure, ensuring that any data used or generated remains strictly in-house.

This significantly reduces the risks associated with sharing sensitive information via third-party cloud providers. For the EU, this prospect aligns neatly with the broader aim to reduce dependency on external tech giants—be they American or Chinese—and to preserve highly regulated industries such as finance, healthcare and energy within the sphere of European legal and ethical standards.

Who Are the Winners and Losers? 

The growing interest in DeepSeek highlights a broader tension in the global AI arena: For many European organisations—both public and private—a Chinese-origin model capable of rapid deployment and cost-effective scaling might become a surprising boon. 

This could help the EU level the playing field faster than many might have anticipated, particularly in sectors that require robust, domain-specific AI, such as defence, large-scale multilingual documentation and secure enterprise solutions. In this sense, governments, military contractors, and tech-savvy companies open to adopting or refining open-source technologies stand to gain substantially.

Conversely, businesses that have thrived on proprietary, high-cost AI models may face pressures to reduce their pricing or justify their premium services. If DeepSeek and similar platforms prove both reliable in quality and flexible in deployment, established AI providers—particularly those that lock clients into expensive, cloud-based architectures—may find themselves losing market share in Europe. 

Additionally, concerns over data privacy and geopolitics could hinder the adoption of a model with Chinese roots, meaning that risk-averse organizations could prefer Western-made or domestically produced alternatives, if available.

Still, the “irony” remains a compelling story: Even in an era of intensifying trade and technology rivalries, a cost-efficient, open-source model like DeepSeek can paradoxically become a cornerstone for Europe’s drive toward greater autonomy in AI. 

By embracing technologies such as DeepSeek and using them as a model to be structured in a secure and regulated manner, Europe stands poised to accelerate AI adoption at a critical moment. 

In the end, the successful deployment of LLMs will come down to striking a balance between regulatory compliance, ethical considerations and the pursuit of technological autonomy — an equilibrium that DeepSeek and similar AI models could help Europe achieve more quickly than previously imagined.