A Chinese startup has stunned the global tech industry by launching an impressive AI model named DeepSeek-R1. This breakthrough technology is on par with leading Google Gemini and OpenAI models.
What makes DeepSeek-R1 even more remarkable is how it was built. The company used fewer and less advanced computer chips than major U.S. tech giants typically require.
This achievement goes beyond just technological progress. It marks a major shift in the AI industry. Instead of relying on brute-force scaling and expensive resources, DeepSeek shows that smarter and more efficient methods can produce top-tier results. This approach challenges old beliefs and could redefine how future AI systems are developed.
What is DeepSeek?
DeepSeek is an AI company based in Hangzhou, a city in Southeastern China. It was founded in July 2023. However, according to Sensor Tower, its popular AI assistant app didn’t launch in the U.S. until January 10.
Similar to other AI startups like Anthropic and Perplexity, DeepSeek has released several AI models over the past year. According to the Wall Street Journal, the V3 model gained attention but faced criticism due to content restrictions on topics about the Chinese government, raising concerns about its competitiveness in the industry.
In late 2023, DeepSeek surprised the industry with the release of its R1 model. The company formally launched it last week and revealed to the Wall Street Journal that R1 operates at a surprisingly low cost. It is also open-source, allowing other companies to test and improve it.
DeepThink (R1) is a key feature of the DeepSeek AI platform, known for its advanced reasoning abilities. It uses a multi-stage training process with reinforcement learning (RL) and Chain-of-Thought (CoT) prompting. This setup, combined with a large parameter model, helps the AI break down complex problems step by step.
DeepThink (R1) can generate structured outputs like code snippets and tables, solve math problems, and handle coding challenges. It also excels at logical puzzles and provides detailed explanations. Its ability to “think” through problems makes it more effective than traditional language models.
The DeepSeek app quickly climbed app store rankings, surpassing ChatGPT on Monday. It has already been downloaded nearly 1.2 million times from the Play Store and more than 1.9 million times from the App Store worldwide. However, the actual number might be higher, as the Play Store shows a label indicating over 5 million downloads.
Who Founded DeepSeek?
The company was founded by Liang Wenfeng late in 2023. Born in Guangdong, China, Liang grew up when the region embraced market capitalism in the 1980s and 1990s. At 17, he enrolled at Zhejiang University, where he studied electronics and communication engineering. He later completed a master’s degree in information and communication engineering in 2010.
In 2015, Liang co-founded High-Flyer, a quantitative hedge fund that relies on advanced mathematical algorithms for trading instead of human analysis. By the end of 2021, the fund managed a portfolio worth over 100 billion yuan ($22 billion).
In April 2023, the company announced on WeChat that it would shift its focus beyond investments to explore artificial general intelligence (AGI). A month later, they founded DeepSeek, and by the end of the year, the company launched its first large AI language model.
Liang Wenfeng is now a key figure in China’s AI revolution as the country strives to compete with the U.S. in the tech industry. Although he has kept a low profile, Liang was one of nine speakers at a private symposium hosted by China’s Premier Li Qiang on January 20 this year. Before that, he gave two interviews to the Chinese media outlet Waves in 2023. Despite his rare media appearances, Liang has not held back from sharing his thoughts on China’s role in the global AI race.
Is DeepSeek AI free?
Yes, DeepSeek is a free AI chatbot powered by the DeepSeek-V3 model, which handles a wide range of tasks effectively. Recently, the company introduced the DeepSeek-R1 model, designed for advanced reasoning. This model is comparable to OpenAI’s o1. Users can access it for free by simply switching on the DeepThink (R1) mode.
Ready to get started? Here’s how to use DeepSeek right away.
How to Use DeepSeek?
Accessing DeepSeek through a smartphone or tablet is simple. You can either download the app from the Apple App Store or Google Play Store or if you prefer not to download the app, you can access DeepSeek directly on the web.
Go to https://chat.deepseek.com/sign_in and register with your email or phone number. You also have the option to sign up using your Google account. Once registered, you can start chatting with the AI.
When you log in, you’ll see a user interface similar to ChatGPT. At the bottom, there’s a text box where you can chat with the AI. Also, to view past conversations, tap the two horizontal lines in the top left corner, and it will show you previous chats.
To use the DeepThink R1 model or search the internet, you must manually enable these features. Click the DeepThink (R1) and Search buttons below the text box. If you want the AI to extract text from images or analyze a document click the ‘
’ button on the right side of the text box.
To start a new conversation, tap the New chat button in the top left corner, and DeepSeek will begin a fresh chat. Right now, the chatbot has basic features, but it’s expected to add more in the future.
What Sets DeepSeek Apart from Competitors?
Unlike many AI models in the West that need huge funding, DeepSeek has built high-quality models at a much lower cost. Training DeepSeek’s models costs about $5.6 million, whereas OpenAI’s models are estimated to cost over $100 million each.
Also, DeepSeek-R1 stands out for its strong reasoning abilities, especially in solving math problems and logical challenges. It competes with OpenAI’s GPT models but is better optimized in certain areas. Moreover, DeepSeek offers unlimited free access to its AI, which is rare for such services.
However, running a large AI model like R-1 is expensive. The costs for hardware, electricity, and maintenance can add up quickly. So, how does DeepSeek manage to offer this for free?
Most likely, the answer lies in user data. AI models depend on data, and DeepSeek may be collecting user data to improve its models or explore other ways to make money. For example, the company could be using the data to improve its trading algorithms or create other AI-based products. This is common in the tech industry, but users should be aware of it.
Furthermore, DeepSeek makes its AI research open-source, which helps developers and businesses access it more easily. This openness encourages innovation by allowing other companies to build upon DeepSeek’s models.
DeepSeek’s Impact on the AI Market
On Tuesday, US tech stocks stabilized after dropping on Monday due to the unexpected rise of the Chinese AI app DeepSeek. DeepSeek became the most downloaded free app in the US just one week after its launch.
The market dip happened as investors quickly reevaluated their AI investments following DeepSeek’s claim that its model was developed at a fraction of the cost of its competitors. Analysts said this development raised concerns about the future of US dominance in AI and the scale of investments US companies are planning.
US President Donald Trump called DeepSeek’s rise “a wake-up call” for the US tech industry but also suggested it could be “positive” for the country in the long run. The surge in AI investments has driven much of the US stock market boom over the past two years.
Nvidia, which produces the advanced chips that power many AI applications, saw its stock drop by 17% on Monday, losing nearly $600 billion in market value. This was the biggest impact on the market following DeepSeek’s rise.
However, the shock to the markets didn’t last long. On Tuesday, Nvidia’s shares increased by 8.8% after a decline on Monday. Also, on the same day, major US indexes had recovered, with the Dow Jones up by 0.3%, the S&P 500 rising nearly 1%, and the Nasdaq gaining 2%. The FTSE 100 also saw a small rise of 0.35%.
Future of DeepSeek in the AI Industry
DeepSeek has introduced a new approach to AI model training that cuts down on data processing. It uses a method called Mixture of Experts (MoE), where the model relies on specialized sub-models, or “experts,” to solve problems instead of using the whole model. This technique allows DeepSeek to build smaller, more efficient AI models that work faster and consume less energy.
This innovation significantly affects the entire AI industry, challenging major players like Nvidia, OpenAI (the creator of ChatGPT), and even big tech companies like Microsoft, Google, Amazon, and Meta. It also impacts power providers like Vistra. Investors are worried that DeepSeek’s advancements could lower the demand for high-performance chips, reduce energy use, and threaten the massive investments—worth hundreds of billions of dollars—already made in AI model development.
Also, David Sacks, an AI expert, suggests that DeepSeek may have stolen intellectual property from the United States. He believes that DeepSeek used a technique called “distillation” to extract knowledge from OpenAI’s models. Distillation is a machine learning method that transfers the knowledge of a large model into a smaller, more efficient one. If these accusations are proven true, the U.S. might impose more sanctions on exports of American technology to China.
However, Some industry leaders see DeepSeek’s announcement as a positive development. Roger Dassen, CFO of semiconductor company ASML, shared his thoughts at a press conference after the Q4 2024 financial results were released. He explained, “If you make it cheaper, and if you make it more accessible, if you make it more energy efficient, it can lead to a democratization of AI in a way that is necessary in order for this to become prevalent.”
Challenges and Controversies
In recent years, the United States has taken various steps to limit China’s progress in AI development. These measures include restrictions on advanced chip exports, bans on chip-making equipment, and blacklisting of Chinese firms linked to military or dual-use technology.
Despite these efforts, DeepSeek’s success has raised questions about the effectiveness of these controls. The company reportedly sidestepped restrictions by stockpiling lower-grade Nvidia chips before the ban took effect. This challenges the assumption that US measures have significantly hindered China’s AI advancements.
DeepSeek’s development suggests that cutting off access to crucial technology might not guarantee success for the United States. This serves as a key reminder for leaders, including President Donald Trump, who advocated for the “America First” policy with a focus on isolationism. The situation highlights the need to reconsider whether restricting technological cooperation always leads to a competitive advantage.
Conclusion
DeepSeek’s innovation marks a turning point in AI development by prioritizing efficiency over expensive resources. Its rapid growth challenges Western tech giants and reshapes the AI industry landscape. Despite controversies and restrictions, DeepSeek’s rise underscores China’s growing AI capabilities.
