Quick answer

DeepSeek is a Chinese AI company that in early 2025 released a model matching GPT-4 performance — at a fraction of the cost to build. It shocked the industry not because it was better, but because it proved top-tier AI could be built far cheaper than Silicon Valley was spending. It briefly wiped $600 billion from Nvidia's stock price.

In January 2025, a relatively unknown Chinese AI company called DeepSeek released a model called DeepSeek-R1. Within days it was the most downloaded app on the US App Store — and Nvidia lost $600 billion in market cap in a single day. Here is what happened, why it matters, and what it actually means for you.

What is DeepSeek?

DeepSeek is an AI research lab founded in 2023 in Hangzhou, China. It is backed by High-Flyer Capital Management, a Chinese hedge fund. In a very short time, it went from unknown to releasing models that genuinely compete with the best from OpenAI, Google, and Anthropic.

Why did it cause such a shock?

The AI industry had assumed that building cutting-edge AI required billions of dollars and the most powerful chips. OpenAI reportedly spent over $100 million training GPT-4. DeepSeek claimed to have trained its R1 model for approximately $6 million.

This was extraordinary. If true, it meant: the cost barrier to building frontier AI was collapsing, China could build world-class AI despite US chip export restrictions, and the business models of companies selling expensive AI compute were suddenly under question.

Why Nvidia's stock dropped: Most AI computing runs on Nvidia chips, which cost tens of thousands of dollars each. If AI can be built cheaply — as DeepSeek suggested — demand for those chips might be lower than expected. The market reacted instantly.

How good is it actually?

DeepSeek-R1 and its successor models are genuinely excellent — competitive with GPT-4o and Claude on most benchmarks, especially for reasoning and coding tasks. Independent testers confirmed the results were real. It is not hype.

What are the concerns?

  • Data privacy: DeepSeek's servers are in China. Using DeepSeek means your queries go to Chinese infrastructure, which is subject to Chinese law — including potential government access.
  • Censorship: DeepSeek will not answer questions about topics sensitive to the Chinese government (Tiananmen Square, Taiwan, Xinjiang). This is built into its training.
  • Open-source availability: DeepSeek released its model weights publicly, meaning anyone can download and run the model. This was controversial — some see it as open science, others as a national security concern.

What does it mean for the AI race?

DeepSeek proved that China is a genuine frontier player in AI, not just a follower. It challenged the assumption that chip restrictions would hold back Chinese AI development. And it forced US AI companies to justify their enormous spending — if a model this good can be built for $6 million, why are they spending billions?

Bottom line

DeepSeek is a genuinely impressive AI model, and its cost story reshaped how the industry thinks about AI economics. For everyday users, the most important lesson is that high-quality AI is getting cheaper fast — which is good news for access and competition. Just be thoughtful about where your data goes when you use it.