Chinese AI Startup DeepSeek Challenges OpenAI and Google with Low-Cost Breakthroughs

January 27, 2025

A relatively unknown Chinese AI startup, DeepSeek, has stunned the tech industry by developing artificial intelligence models that rival those of major U.S. firms like OpenAI, Google, and Meta, all while spending significantly less money and using fewer high-end computer chips. This unexpected success has raised concerns in Silicon Valley and sparked discussions about the effectiveness of U.S. restrictions on AI chip exports to China.

DeepSeek’s Surprising Rise in AI Research

DeepSeek, originally a deep-learning research division within High Flyer, a Chinese quantitative hedge fund, was spun off as a separate AI company in 2023. Unlike other AI startups, which often depend on large technology firms for funding, DeepSeek pursued independent research and focused on open-source AI development.

The company first gained attention with its DeepSeek-V3 model, released in late December 2024. The model demonstrated performance comparable to OpenAI’s and Google’s top AI systems while requiring only a fraction of the computing power used by U.S. firms. Just a month later, DeepSeek released DeepSeek-R1, an even more advanced “reasoning” AI model, which quickly became the most downloaded AI application on Apple’s App Store​.

How DeepSeek Built a High-Performing AI Model with Fewer Resources

Most leading AI companies, such as OpenAI and Meta, spend hundreds of millions of dollars on training their AI models, relying on vast networks of powerful Nvidia H100 chips—a resource that has become harder to acquire due to U.S. export restrictions​.

However, DeepSeek took a different approach. Instead of simply scaling up their computing resources, they optimized software and improved efficiency in three key ways:

1. Efficient AI Model Training: DeepSeek engineers designed their AI models using advanced techniques like Multi-head Latent Attention (MLA) and Mixture-of-Experts (MoE). These methods help improve computational efficiency, reducing the number of chips needed to train large AI models​.
2. Optimized Use of Limited Hardware: While the company initially had access to 10,000 Nvidia H100 chips, U.S. sanctions limited their ability to acquire more. Instead of relying on high-end chips, they developed custom software solutions that maximized the performance of the less powerful chips available​.
3. Hiring Young, Ambitious Researchers: Unlike many established AI companies that recruit industry veterans, DeepSeek focused on young PhD students from top Chinese universities, such as Peking University and Tsinghua University. Many of these researchers had backgrounds in theoretical computer science and mathematics but little industry experience. This approach helped create an environment that encouraged innovative problem-solving rather than focusing on short-term commercialization​.

A Direct Challenge to OpenAI and Google

DeepSeek’s success raises questions about the effectiveness of U.S. trade policies aimed at slowing down China’s AI progress. By limiting China’s access to advanced AI chips, the U.S. government expected to prevent Chinese companies from competing with American firms like OpenAI and Google. However, DeepSeek’s ability to develop top-tier AI with fewer resources suggests that these restrictions may have accelerated China’s AI innovation rather than stopping it​.

One of the biggest surprises was the low cost of DeepSeek’s AI development. While OpenAI and Meta spend hundreds of millions of dollars training AI models, DeepSeek reportedly trained DeepSeek-V3 for just $6 million—a fraction of what its U.S. rivals spend​.

Chris Nicholson, an investor at Page One Ventures, noted, “The number of companies who have $6 million to spend is vastly greater than the number of companies who have $100 million or $1 billion to spend.” This suggests that the AI industry may no longer be limited to a few wealthy companies but could become accessible to smaller players, thanks to more efficient training techniques​.

The Impact on U.S. AI Companies and the Market

DeepSeek’s advancements have sent shockwaves through Silicon Valley and Wall Street.

• Nvidia, the leading AI chipmaker, saw its stock price drop after DeepSeek’s success demonstrated that cutting-edge AI models could be trained without relying on top-tier Nvidia chips. This could reduce demand for Nvidia’s most expensive products, impacting its future earnings​.
• Meta, Google, and OpenAI are now facing increased competition from a company that has shown it can produce high-quality AI models at a much lower cost.
• Investors and analysts are now questioning whether the U.S. still holds a major lead in AI innovation or if China is closing the gap​.

According to Professor Ion Stoica of the University of California, Berkeley, “The center of gravity of the open-source community has been moving to China. This could be a huge danger for the U.S.” He explained that if more AI development happens in China, U.S. companies may become dependent on Chinese-developed AI tools, shifting the balance of power in global technology​.

What’s Next for DeepSeek and Global AI Development?

While DeepSeek’s models have impressed researchers and industry experts, the company’s long-term strategy remains uncertain.

• Unlike OpenAI or Google, DeepSeek does not focus on selling AI products to consumers. Instead, it operates more like an AI research lab, open-sourcing its work and allowing others to build on its technology​.
• The Chinese government’s AI regulations do not apply as strictly to DeepSeek because it does not develop consumer-facing AI services. This allows its researchers to experiment more freely without government oversight​.
• If DeepSeek continues to release high-performing, open-source AI models, it could reshape the global AI landscape, leading to a more competitive environment where smaller companies can build on its breakthroughs.

DeepSeek’s meteoric rise has already changed the way experts view the future of artificial intelligence. By proving that top-tier AI can be developed with fewer resources, DeepSeek has challenged the idea that only the largest U.S. tech firms can lead in AI innovation. As competition between the U.S. and China continues, DeepSeek’s success story may mark the beginning of a new era in AI development—one where efficiency and creativity outweigh sheer computing power.

This article is based on the following articles:

https://www.wired.com/story/deepseek-china-model-ai

https://www.axios.com/2025/01/27/deepseek-ai-model-china-openai-rival

https://www.nytimes.com/2025/01/23/technology/deepseek-china-ai-chips.html

Background Information

1. What is Artificial Intelligence (AI)?

AI refers to computer systems that can perform tasks that typically require human intelligence, such as understanding language, solving problems, recognizing images, and making decisions. Some common AI technologies include:

• Chatbots: AI programs like ChatGPT that can talk with users and answer questions.
• Voice Assistants: Tools like Siri and Alexa, which understand and respond to spoken commands.
• Self-Driving Cars: Vehicles that use AI to navigate without human drivers.

AI works by learning from large amounts of data and improving over time. The more an AI system is trained, the better it becomes at performing tasks.

2. What is Machine Learning and How Do AI Models Learn?

AI models like DeepSeek’s R1 and OpenAI’s ChatGPT are built using machine learning (ML), a branch of AI that allows computers to recognize patterns and make predictions.

Steps in Machine Learning:

1. Collecting Data: AI models need huge amounts of information (like text, images, or speech) to learn from.
2. Training the AI Model: AI is fed with data and learns patterns by finding relationships between different pieces of information.
3. Improving Accuracy: The model is tested to see if it can make correct predictions or answer questions correctly. If it makes mistakes, researchers adjust it.
4. Using the Model: Once trained, the AI can answer questions, write text, translate languages, generate images, or even help with scientific research.

DeepSeek’s models are especially advanced because they use “reasoning AI,” which means they not only answer questions but also explain how they arrived at their answers.

3. Why Are Computer Chips Important for AI?

AI requires a lot of computing power to function. This is where computer chips come in.

Types of Computer Chips Used in AI:

1. CPUs (Central Processing Units): Found in most computers but not powerful enough for advanced AI tasks.
2. GPUs (Graphics Processing Units): Originally made for video games but are now used in AI because they can process large amounts of data quickly.
3. TPUs (Tensor Processing Units): Special chips made by Google that work even better for AI training.

AI companies use thousands of GPUs and TPUs to train their models. The most powerful AI chips today are Nvidia’s H100 GPUs, which DeepSeek originally used before U.S. restrictions made them harder to obtain.

4. Why is the U.S. Limiting AI Chip Exports to China?

The U.S. and China are in an ongoing technology competition, with each country trying to develop the most powerful AI and computing technologies.

In October 2022, the U.S. government placed restrictions on selling advanced AI chips to China. The reason? The U.S. government wanted to prevent China from using these chips for military applications and to slow down China’s AI progress so that U.S. companies like OpenAI, Google, and Meta could stay ahead.

However, DeepSeek’s success shows that China has found ways to work around these restrictions by using fewer chips but training AI models more efficiently.

5. What is Open-Source AI and Why is It Important?

Unlike companies like OpenAI, which keep most of their AI research private, DeepSeek has shared its AI technology with the world by making it open-source.

What is Open-Source?

• Open-source software means that anyone can access, modify, and improve it.
• Companies and researchers can collaborate to build better AI instead of starting from scratch.

Why Does Open-Source Matter?

• It makes AI technology available to more people instead of keeping it locked within big tech companies.
• It can speed up AI innovation by allowing many researchers to contribute ideas.
• However, some governments worry that bad actors might misuse open-source AI for harmful purposes, like spreading disinformation.

Since DeepSeek has open-sourced its AI models, it could change the way AI is developed worldwide.

6. Why is This Important for the Future of AI?

DeepSeek’s achievements raise important questions about the future of AI development:

• Will the U.S. continue restricting AI chips, or will it rethink its approach?
• Can small AI startups compete with big tech companies like Google and OpenAI?
• Will AI research become more open-source, or will companies keep their work private?
• Will China become the new global leader in AI development?

DeepSeek’s success shows that AI breakthroughs do not always come from the biggest companies with the most money. Instead, innovation and creativity—like DeepSeek’s smarter training techniques—can be just as important as having powerful computer chips.

Debate/Essay Questions

1. Was the U.S. government right to restrict AI chip exports to China, or did this decision actually help China develop better AI?
2. Should AI research be open-source (free for everyone) or kept private by companies?