DeepSeek: A Complete Guide to the AI Chatbot App

DeepSeek is making waves. This week, Chinese AI lab DeepSeek surged into the spotlight as its chatbot app topped the Apple App Store and Google Play charts. Trained with compute-efficient techniques, DeepSeek’s AI models have sparked discussions among Wall Street analysts and tech experts about the U.S.’s ability to maintain its AI dominance and the long-term demand for AI chips.

But how did DeepSeek emerge, and what led to its rapid rise to global recognition?

DeepSeek is backed by High-Flyer Capital Management, a Chinese quantitative hedge fund that leverages AI for trading strategies.

AI enthusiast Liang Wenfeng co-founded High-Flyer in 2015. While studying at Zhejiang University, he began exploring trading, eventually launching High-Flyer Capital Management in 2019 to focus on AI-driven financial algorithms.

In 2023, High-Flyer established DeepSeek as a separate AI research lab, distinct from its financial operations. With backing from High-Flyer, DeepSeek later became an independent company.

Overcoming Hardware Restrictions

From the outset, DeepSeek built its own data center clusters for training AI models. However, like other Chinese AI firms, it has faced challenges due to U.S. export restrictions on advanced hardware. To train one of its recent models, the company had to rely on Nvidia’s H800 chips, a less powerful alternative to the H100, which remains available to U.S. companies.

DeepSeek’s technical team is reportedly made up of a younger talent pool, with the company actively recruiting PhD-level AI researchers from leading Chinese universities. According to The New York Times, DeepSeek also brings in individuals without computer science backgrounds to enhance its AI’s understanding across diverse subject areas.

DeepSeek introduced its first models—DeepSeek Coder, DeepSeek LLM, and DeepSeek Chat—in November 2023. However, it wasn’t until the release of its next-generation DeepSeek-V2 models in the spring of 2024 that the AI industry took serious notice.

Disrupting the AI Market

DeepSeek-V2, a versatile model capable of analyzing both text and images, performed strongly on AI benchmarks while being significantly more cost-efficient than competing models at the time. Its arrival pressured domestic rivals like ByteDance and Alibaba to lower usage fees for some of their models and even offer others for free.

DeepSeek-V3, launched in December 2024, further cemented the company’s reputation.

Internal benchmark tests suggest that DeepSeek V3 surpasses both open-source models like Meta’s Llama and proprietary models accessible only via API, such as OpenAI’s GPT-4o.

Another standout is DeepSeek’s R1 “reasoning” model, released in January 2025. According to DeepSeek, R1 matches OpenAI’s o1 model in key performance tests. Unlike standard AI models, reasoning models like R1 can self-verify their outputs, reducing common errors. While they take longer—sometimes seconds or minutes more—to generate responses, they tend to be more accurate, especially in fields like physics, science, and math.

Regulatory Constraints on DeepSeek

However, DeepSeek’s models come with certain limitations. As a Chinese-developed AI, they must comply with regulations set by China’s internet watchdog, ensuring their responses align with “core socialist values.” For example, in DeepSeek’s chatbot app, R1 refuses to answer questions about topics such as Tiananmen Square or Taiwan’s political status.

DeepSeek’s business model remains unclear, offering products below market rates and avoiding outside investment despite VC interest. The company claims efficiency breakthroughs drive its cost advantage, though some experts disagree.

Developers have widely adopted DeepSeek’s models under permissive licenses, with over 500 derivatives of R1 on Hugging Face and 2.5 million downloads. Its rapid rise has sparked both praise and skepticism, contributing to an 18% drop in Nvidia’s stock and drawing criticism from OpenAI, which called it “state-subsidized” and urged a U.S. ban.

Microsoft added DeepSeek to Azure AI Foundry, while Meta reaffirmed its AI infrastructure spending in response. Nvidia CEO Jensen Huang praised its innovation, noting its reasoning models require significant computing power.

Despite success, DeepSeek faces growing restrictions, with bans in South Korea, New York state, and potential U.S. government action over foreign AI concerns.

Read the original article on: TechCrunch

Read more: Manus is Unlikely to Be China’s Next DeepSeek Moment