DeepSeek has arrived, and it’s shaking the very foundations of the AI industry. As reported by BBC News on 27 January 2025, the Chinese-developed chatbot has sent shockwaves through global markets, wiping over $500 billion off the market value of US tech giant Nvidia and causing steep declines in other major technology stocks. With its launch just last week, DeepSeek has overtaken rivals like ChatGPT to become the most downloaded free app in the US and other part of the world, sparking debates over the future of AI dominance. This is the second big news after ChatGPT that is shaking the AI industry and is being discussed by tech users of all kinds.

What makes DeepSeek stand out? The first factor, of course, is its cost. Training state-of-the-art models like GPT-4 and some other OpenAI models is rumored to cost over $100 million, requiring supercomputers and massive datasets. Similarly, models like Google’s Bard and OpenAI’s GPT-3 come with price tags in the tens of millions. In stark contrast, DeepSeek was supposedly developed at a cost of approximately $6 million—a fraction of what tech giants typically spend.
DeepSeek has caused unprecedented disruption in the AI industry by achieving comparable or superior performance to leading AI models like GPT-4 at a fraction of the cost (~about $6 million versus $100 million+). Its cost-efficient innovations have redefined expectations in AI development.
One of the secret ingredients that keep DeepSeek’s costs so low is its use of FP8 (floating-point 8) training instead of the more expensive FP16 used by many competitors. FP8 refers to a numerical format that represents floating-point numbers using just 8 bits, compared to the 16 bits used in FP16. This smaller format reduces the memory and computational power required for training and inference, making AI development faster and more energy-efficient. Despite the reduced precision, FP8 maintains high accuracy for most tasks, proving to be a practical and cost-effective solution for modern AI models.
DeepSeek also claims exceptional performance across multiple benchmarks. As shown in the image below, DeepSeek-R1-Lite outperforms or matches its competitors in a variety of tasks, such as mathematical reasoning (MATH), coding (Codeforces), and logical reasoning (ZebraLogic). For example, in the MATH benchmark, DeepSeek achieves an impressive accuracy of 91.6%, compared to 85.5% by its closest competitor. Similarly, in Codeforces ratings, it leads with a score of 1450, slightly ahead of the 01-preview model at 1428. These results not only highlight DeepSeek’s technical prowess but also demonstrate how a low-cost approach can deliver results comparable to or even better than models developed with significantly higher budgets.

But how was this possible? DeepSeek achieved this through a combination of clever engineering and resource optimization. For instance, the team focused on refining their algorithms to make the most of the hardware they had, which included GPUs specifically capped in performance due to export restrictions. They also adopted a streamlined approach to data processing, prioritizing efficiency over sheer volume. By using advanced techniques like reinforcement learning and “chain of thought” reasoning, DeepSeek was able to reduce computational waste while enhancing the quality of its outputs. This innovative approach turned hardware limitations into opportunities for groundbreaking advancements.
By leveraging resource optimization techniques like FP8 training and clever algorithm engineering, DeepSeek demonstrates how frugal innovation can challenge established players without compromising performance, excelling in benchmarks like MATH and Codeforces.
The ROI of DeepSeek is unparalleled in the AI industry. For a model developed at approximately $6 million, its impact has been monumental. Consider the market reactions: Nvidia lost $500 billion in market value, and other US tech giants like Microsoft, Tesla, and Alphabet saw significant drops in their stock prices (BBC Live, 27 January, 2025). This shows how a modest investment can disrupt markets and force billion-dollar corporations to reevaluate their strategies. By achieving comparable or better performance than models like GPT-4 at a fraction of the cost, DeepSeek has set a new standard for efficiency and impact in AI development. The return on investment is not just financial but also strategic, positioning China as a formidable player in the global AI race.
DeepSeek’s launch has not been without controversy. Shortly after US markets opened on Monday, DeepSeek announced it was hit by a “large-scale malicious attack,” forcing it to temporarily halt new user registrations. Additionally, the model has raised questions about bias in AI. Just as American-trained models often reflect U.S. political and cultural sensitivities, DeepSeek’s Chinese-influenced training data reveals its own biases. For example, DeepSeek reportedly censors discussions about politically sensitive topics like the Tiananmen Square massacre, a limitation that underscores the importance of region-specific models that balance ethical considerations with technological prowess.
Europe has been conspicuously absent from global headlines in the AI revolution. For decades, the continent has lagged behind its counterparts in the United States and China when it comes to cutting-edge technological innovation. But why? One major factor is the lack of coordinated investment and vision. While the U.S. and China have poured billions into AI research and infrastructure, Europe has remained fragmented, with individual nations pursuing their own isolated strategies rather than pooling resources to create globally competitive AI models.
Europe lags behind the U.S. and China in the AI race due to fragmented strategies and restrictive regulations. DeepSeek’s success highlights the urgent need for Europe to adopt a unified, visionary approach to AI innovation or risk falling further behind in the global economy.
Another issue is the regulatory environment. Europe’s focus on strict data privacy and ethical considerations, while commendable, often stifles rapid innovation. The lengthy approval processes and restrictions have discouraged private investment in high-risk, high-reward AI projects. Meanwhile, U.S. companies benefit from a more flexible regulatory framework, and Chinese companies leverage state-backed initiatives to push boundaries without facing as many constraints as possible.
The result? Europe is falling further behind in a race that’s reshaping the global economy. DeepSeek’s success serves as a stark reminder of what’s possible when vision, strategy, and resources align. Europe cannot afford to ignore this wake-up call. The time has come for a unified approach to AI—one that balances innovation with ethical considerations and fosters collaboration across borders. Without significant changes, Europe risks becoming a passive observer in the AI revolution rather than an active participant.