DeepSeek, a Chinese AI firm, has shaken up the artificial intelligence landscape with its innovative and low-cost, open-source large language models, presenting a serious challenge to U.S. tech giants like OpenAI.
Historically, it was believed that cutting-edge large language models (LLMs) could only be created with massive technical and financial resources, a belief that led to the U.S. government’s commitment to the $500 billion Stargate Project. However, DeepSeek has defied this expectation.
In January 2025, DeepSeek unveiled its R1 LLM, developed at a fraction of the cost compared to its U.S. counterparts. To further disrupt the market, the company made its R1 models available under an open-source license, allowing anyone to use them for free.
The release quickly gained traction, and within days, the DeepSeek AI Assistant app, a chatbot based on the R1, became the most downloaded app on Apple’s App Store, surpassing OpenAI’s ChatGPT mobile app. This rapid rise in usage led to a market sell-off by January 27, 2025, as investors began questioning the valuations of major U.S. AI firms like Nvidia, Microsoft, and Meta Platforms.
What is DeepSeek? Founded in May 2023 by Liang Wenfeng, a Zhejiang University graduate and co-founder of China’s High-Flyer hedge fund, DeepSeek is an AI development firm based in Hangzhou. The company operates as an independent AI research lab under High-Flyer’s umbrella, although its total funding and valuation remain undisclosed.
DeepSeek specializes in developing open-source LLMs. Its journey began with the release of its first model in November 2023, and its R1 model in January 2025 earned the company global recognition. DeepSeek also offers its models via a web interface, mobile app, and API.
DeepSeek vs. OpenAI: A New Rival Emerges While OpenAI has long been the leader in AI with its GPT and reasoning models, DeepSeek’s R1 model offers a cost-effective alternative. Here’s how the two companies compare:
Feature | OpenAI | DeepSeek |
---|---|---|
Founding Year | 2015 | 2023 |
Headquarters | San Francisco, Calif. | Hangzhou, China |
Focus | Broad AI capabilities | Efficient, open-source models |
Key Models | GPT-4, o1 | DeepSeek-V3, DeepSeek-R1 |
API Pricing | $15 (input), $60 (output) | $0.55 (input), $2.19 (output) |
Open Source | Limited | Mostly open-source |
Training Cost | Hundreds of millions for o1 | Less than $6 million for DeepSeek-R1 |
DeepSeek has taken a different approach, leveraging reinforcement learning and reward engineering for training its models, leading to substantial cost savings.
Training Innovations at DeepSeek DeepSeek’s R1 model benefited from several key innovations:
- Reinforcement Learning: Large-scale reinforcement learning for reasoning tasks.
- Reward Engineering: A new rule-based reward system for better learning outcomes.
- Distillation: Efficient knowledge transfer, reducing model size to 1.5 billion parameters.
- Emergent Behavior Network: The discovery that complex reasoning can evolve naturally through reinforcement learning.
A Rapidly Growing AI Model Portfolio Since its inception, DeepSeek has released several AI models, each version building upon the last:
- DeepSeek Coder (Nov 2023): Open-source model for coding tasks.
- DeepSeek-V3 (Dec 2024): A 671 billion-parameter model, optimized for a variety of tasks.
- DeepSeek-R1 (Jan 2025): Focused on advanced reasoning tasks, directly competing with OpenAI’s o1.
- Janus-Pro-7B (Jan 2025): A vision model that can process and generate images.
Why DeepSeek’s Success is Worrying U.S. Firms The rapid success of DeepSeek-R1 has led to concerns in the U.S. for several reasons:
- Cost Disruption: DeepSeek developed its R1 model for under $6 million, threatening the business models of expensive U.S. AI providers.
- Technical Success Despite Restrictions: Despite export restrictions on AI accelerators, DeepSeek has proven that cutting-edge AI development is still possible.
- Business Model Challenge: DeepSeek’s open-source, free model undermines the subscription-based revenue models of U.S. companies.
- Geopolitical Tensions: DeepSeek’s rise signals a shift in global AI power, with experts comparing it to the Soviet Union’s space race achievements.
Cybersecurity Challenges for DeepSeek Amid its success, DeepSeek faced a significant cyberattack on January 27, 2025, which forced the company to restrict new user registrations temporarily. The attack was likely a Distributed Denial-of-Service (DDoS) assault targeting its API and platform.
Data Breach Concerns On January 29, 2025, Wiz Research reported a data leak exposing sensitive information, including chat histories and API keys. DeepSeek quickly took the database offline after being alerted, though the full scope of the breach remains unclear.
As DeepSeek continues to rise in prominence, it remains a formidable challenger to U.S. tech giants and could reshape the future of AI development.