Last Updated:
The Chinese-made artificial intelligence (AI) model has shot to the top of Apple Store’s downloads, stunning investors and sinking some tech stocks
DeepSeek’s creator says its model was developed using less advanced, and fewer, computer chips than employed by tech giants in the United States. Representational image/X
A little-known Chinese startup has sent shockwaves through the global tech sector with the release of an artificial intelligence (AI) model whose capabilities rival the creations of Google and OpenAI.
DeepSeek’s creator says its model was developed using less advanced, and fewer, computer chips than employed by tech giants in the United States.
The AI startup’s latest app comes just days after US President Donald Trump announced a new $500 billion venture with ChatGPT maker OpenAI, Softbank, and Oracle, dubbed Stargate, which he touted as ensuring “the future of technology” in the United States.
Here are the answers to some of the most frequently asked questions on DeepSeek:
1. What is DeepSeek?
DeepSeek is an artificial intelligence company based in China. The company’s AI application can be accessed as an Android app, iOS app, on a web browser, or through APIs (Application Programming Interfaces) for use within other apps.
2. What can a user do with DeepSeek?
DeepSeek’s end usage is similar to other AI tools (such as OpenAI’s ChatGPT, Google Gemini, Microsoft Copilot, etc). It can generate answers to questions, help in research, aid in writing, editing and proofreading, translate content between different languages, extract and analyse text from files, summarise documents, help in writing and debugging software code, etc.
3. Who owns DeepSeek?
DeepSeek is owned and funded by the Chinese hedge fund company High-Flyer. Liang Wenfeng, the co-founder of High Flyer, established DeepSeek in 2023 and is its CEO.
4. Is DeepSeek new?
Not really. The company was founded in May 2023. It released its first model (DeepSeek Coder) in November 2023.
5. What makes DeepSeek different from other AI models?
DeepSeek claims that it has been able to reduce the computing and data resources needed for its operations and is reported to have trained its AI at a fraction of the cost of OpenAI. This is also reflected in its API costs, which are more than 90% cheaper than the comparable model from OpenAI.
Also, DeepSeek says its models are open source, meaning that the key components can be used by anyone to build their own tools for free. Meta’s Llama is also open source; however, OpenAI’s ChatGPT isn’t.
6. Why the hullabaloo now?
On January 20, DeepSeek released its new large language model (LLM), DeepSeek-R1, which ranked as high or higher than OpenAI’s model in performance tests. This, coupled with the claim that DeepSeek has been able to achieve this at a mere fraction of the cost, caught the interest of the technology and the financial communities. Silicon Valley capitalist Marc Andreessen described it as “AI’s Sputnik moment”, drawing a parallel with the launch of the Sputnik 1 satellite by the Soviet Union in 1957, which had caught the US unawares and triggered the space race.
This was soon followed by questions around the scale of American investments in AI at a time when the Chinese were able to achieve similar or better results at a fraction of the cost and a market reaction that shaved off almost $600 billion from chip-maker Nvidia market cap, the biggest single-day loss in US history.
The buzz around DeepSeek made DeepSeek’s AI assistant the no. 1 downloaded free app on Apple’s iPhone store in the US.
7. Why Nvidia?
Nvidia is an American computer chip design company that manufactures its products in Taiwan. Nvidia specialises in graphics processing units (GPUs), which are designed to improve image and video rendering on computers. Also, these GPUs are also able to perform calculations concurrently in a way that regular central processing units (CPUs) cannot. This ability of GPUs makes it of high value for AI companies, which must rely on very high volumes of computations to build their LLMs (Large Language Models). Its first-mover advantage and existing supply chain enabled Nvidia to supply the needs of emerging AI companies better than any of its rivals.
Nvidia shares have been riding high on the AI boom, and on October 26, 2024, Nvidia even overtook Apple as the world’s most valuable company, with its market value touching $3.53 trillion.
8. But hasn’t the US restricted exports of AI chips to China? So how did DeepSeek build its AI?
Yes. However, DeepSeek’s parent company, High-Flyer, acquired over 10,000 Nvidia A100 GPUs in 2022 before the US export restrictions kicked in. DeepSeek, though, says that its recent model was built using Nvidia’s lower-performing H800 chip, which is not restricted for sale to China. This further adds to the debate that top-of-the-line hardware is not necessary for building top-notch AI models.
9. What is LLM?
LLM, or Large Language Model, is a type of artificial intelligence (AI) computer program that can recognise and generate text and also perform other tasks. These programs are trained on vast amounts of data generally scraped from the internet.
10. Do we know what data DeepSeek was trained on, given that it is open source?
While DeepSeek has open-sourced its R1 code under an MIT license, it hasn’t revealed the source of its training data. There have been allegations that DeepSeek’s LLM could have been trained on synthetic data (data produced by other LLMs) instead of original data collected from online sources.
11. Given that DeepSeek is a Chinese company, are there data security concerns?
According to DeepSeek’s privacy policy, it collects personal data such as text, audio input, and chat history, and that data is stored in China, thereby subjecting it to China’s data-sharing laws.
12. What about the censorship that China is known for? Does that also extend to DeepSeek’s responses?
As DeepSeek operates out of China, it must comply with Chinese government regulations, which include censoring topics deemed contentious by Chinese authorities. It deflects any such questions.
13. What does DeepSeek mean for the larger AI industry?
As DeepSeek’s code is open source and others can learn from and improve upon it, it is likely to benefit the larger AI community in increasing efficiencies and reducing costs, thereby further speeding up AI acceleration.
14. Does DeepSeek support Indian languages?
Yes. According to DeepSeek, it supports 22 Indian languages. However, the proficiency varies depending on the availability of training data. It claims strong support for 11 Indian languages (Hindi, Bengali, Tamil, Telugu, Marathi, Gujarati, Kannada, Malayalam, Odia, Punjabi, and Urdu) and moderate to limited support for 11 other languages (Assamese, Sanskrit, Sindhi, Nepali, Konkani, Maithili, Kashmiri, Manipuri (Meitei), Dogri, Santali, and Bodo).
