DeepSeek: The Chinese AI Startup Rattling Silicon Valley and Wall Street

Introduction:

DeepSeek, a Chinese artificial intelligence (AI) startup, has emerged as a formidable competitor, demonstrating AI models comparable to OpenAI at a fraction of the cost and energy consumption. This has sent shockwaves through the tech industry, sparking a global sell-off of tech stocks and raising questions about the future of AI chip restrictions.

DeepSeek's AI Models:

DeepSeek released its latest open-source reasoning models, DeepSeek-R1, which showed comparable performance to OpenAI's models on several industry benchmarks. Additionally, DeepSeek-R1-Zero, which was trained without supervised fine-tuning, demonstrated remarkable reasoning capabilities.

DeepSeek's AI chatbot, powered by DeepSeek-V3, has also gained popularity, surging to the top of Apple's App Store downloads. Unlike other chatbots, DeepSeek provides explanations for its reasoning before responding to inquiries. However, it avoids politically sensitive topics.

Comparison to Competitors:

On leading benchmarks such as AIME 2024 and MMLU, DeepSeek-R1 performed comparably with models from OpenAI and Meta. On the Chatbot Arena leaderboard, DeepSeek-R1 ranked below Google's Gemini 2.0 and ChatGPT-4o, while DeepSeek-V3 fell slightly short of OpenAI's o1 models.

Impact on Tech Stocks and Nvidia:

DeepSeek's competitive models could challenge Nvidia's business, which relies heavily on AI firms spending billions on its GPUs. DeepSeek's use of fewer GPUs for training its models raises concerns about Nvidia's chip dependency.

Implications for U.S. Chip Restrictions:

DeepSeek's success using Nvidia's H800 chips, which are less powerful than the H100s restricted for export to China, highlights the challenge of curbing advanced chip flow into the country. The Biden administration has recently imposed stricter measures to prevent AI chips from reaching China.

Conclusion:

DeepSeek's innovative AI models and cost-effective approach have shaken the AI industry. Their performance and the implications for U.S. chip restrictions are likely to continue shaping the future of artificial intelligence.