In a landscape dominated by names like OpenAI and Google, a Chinese company called DeepSeek emerged in 2024 as one of the biggest surprises in the world of artificial intelligence. With open-source language models, ultra-low costs, and technical innovations, it is redefining what’s possible in the field—and teaching Silicon Valley giants a lesson or two.

---

1. Who is DeepSeek?

Founded in May 2023 as a research arm of the quantitative investment fund High-Flyer, DeepSeek has prioritized algorithmic efficiency and open access to its technologies from the start. With just 200 employees (compared to OpenAI’s 3,500), the company gained global attention by releasing models that rival the best in the world, such as DeepSeek-V3 (with 671 billion parameters) and R1-Lite-Preview (specialized in mathematical reasoning).

---

2. Innovations That Set DeepSeek Apart

• Revolutionary Training Costs: DeepSeek-V3 was trained for just $5.58 million—less than 1/10th the cost of similar Western models. In comparison, OpenAI’s GPT-4 cost hundreds of millions to develop.

• Disruptive Pricing: Its API services cost U$ 0.55 per million input tokens, compared to OpenAI’s  U$ 15. This democratizes access to advanced AI for startups and researchers.

• Efficient Architecture: Techniques like Multi-Head Latent Attention (MLA) and Mixture-of-Experts (MoE) allow its models to use fewer computational resources without sacrificing performance. This was crucial for navigating U.S. sanctions on high-performance chips.

---

3. Performance That Surprised the World

In benchmark tests, DeepSeek-V3 outperformed established models like Llama 3.1 405B and matched GPT-4o and Claude 3.5 Sonnet in reasoning and coding tasks. Additionally:

• Its DeepSeek-Coder-V2 model is as good as GPT-4-Turbo in programming, supporting 338 coding languages.

• On the Chatbot Arena (a global AI model ranking), DeepSeek is among the top 5, ahead of giants like Meta.

---

4. Impact on the Global Market

DeepSeek’s success hasn’t gone unnoticed:

• Historic Drop in Nvidia’s Stock: After the launch of DeepSeek-R1 in January 2025, Nvidia’s stock fell 17% in a single day, reflecting concerns about U.S. technological leadership.

• Pressure for Lower Prices: DeepSeek forced companies like Alibaba and Tencent to reduce costs for their models, accelerating AI adoption in China and beyond.

---

5. Why This Matters

DeepSeek’s rise highlights three critical trends:

1. Open Source as a Strategy: By releasing models under the MIT license, the company attracted global collaboration and accelerated innovation—a lesson for the West.

2. Efficiency > Resources: Hardware restrictions (like U.S. chip export bans) didn’t stop progress but instead spurred creative optimizations.

3. The Geopolitics of AI: China is proving it can compete in critical technologies, challenging the narrative of Western dependency.

---

Conclusion: A New Chapter in AI

DeepSeek isn’t just another AI company—it’s a symbol of how innovation can thrive under constraints. Its accessible and powerful models are leveling the playing field, enabling startups and developing countries to join the AI revolution. While Silicon Valley wonders how to respond, one thing is clear: the future of artificial intelligence will be shaped by diverse voices—and DeepSeek is one of the loudest.

Want to learn more? Explore DeepSeek’s models at deepseek.com.