What Is DeepSeek and Why It Shook Up the AI Industry

A Surprise From Hangzhou

DeepSeek is a Chinese artificial intelligence company. It was started in 2023 by Liang Wenfeng, a fund manager who loved math puzzles. Most people outside China had never heard of it. Then, in early 2025, DeepSeek released a model called R1 that could think through hard problems step by step — the same kind of "reasoning" feature that OpenAI had shown off months earlier.

Two things made DeepSeek shocking. First, the model was open source, meaning anyone in the world could download it, run it on their own computer, and see how it worked. Second, DeepSeek said they had trained it for about $5.6 million. That number was tiny compared to the billions being spent by US companies. It was like someone showing up to a Formula 1 race in a homemade go-kart and beating the pros.

The Whole Industry Repriced Overnight

Before DeepSeek, most people believed you needed huge amounts of money, the best computer chips, and giant data centers to build a top AI model. The biggest US companies were spending tens of billions of dollars and bragging about it. Then DeepSeek showed that a small, smart team could do similar work for a sliver of that cost.

This mattered for three big reasons. It put pressure on the prices AI companies charge, since a cheaper model means they can charge less. It raised questions about export rules on computer chips, because DeepSeek reportedly got good results using less powerful hardware that was easier to get. And it showed the world that the US did not have a permanent lead in AI — other countries could catch up fast.

How Did They Do It So Cheap?

DeepSeek didn't invent new science. What they did was squeeze every drop of efficiency out of the tools they had. They were clever about the math, the training process, and the way the model uses memory. Here are the four big tricks they used.

That last step — open weights — turned DeepSeek into a worldwide project. Once the model was public, thousands of researchers, students, and companies ran their own tests, found bugs, and shared fixes. The cost of "one team in China" suddenly became the cost of "everyone with a laptop."

Talking to DeepSeek From Your Code

Because DeepSeek is open source, you can use it two ways. You can run the model on your own computer (no internet needed), or you can pay a small fee to use their hosted API. The API works almost exactly like OpenAI's, which makes it easy to swap in. Here's a tiny Python script that asks DeepSeek to explain a joke.

# Ask DeepSeek a question using its OpenAI-compatible API
import requests

url = "https://api.deepseek.com/v1/chat/completions"

headers = {
    "Authorization": "Bearer YOUR_API_KEY",
    "Content-Type": "application/json"
}

data = {
    "model": "deepseek-chat",
    "messages": [
        {"role": "user", "content": "Explain this joke in one sentence: Why did the AI go to therapy?"}
    ]
}

response = requests.post(url, headers=headers, json=data)
print(response.json()["choices"][0]["message"]["content"])

Run that, and DeepSeek will send back a written explanation. The price for this kind of request is usually less than half a cent — much cheaper than calling OpenAI for the same task.

Knowledge Check

Test what you learned with this quick quiz.