When OpenAI launched ChatGPT on November 30, 2022, it took the world by storm. Everyone was mesmerized by this revolutionary product, which allowed the user to have one-to-one conversations with a chatbot, get answers to all their queries, and assistance in various other tasks. Over two years later, the product, once launched with limited capability, is more refined, fast, reliable, intelligent, capable, and has better reasoning. OpenAI with ChatGPT is also very close to achieving Artificial General Intelligence (AGI) – machines that can perform any intellectual task humans can – but it is still a work in progress.
As a result of years of research and billions of dollars in investments, a neck-to-neck competition was not what OpenAI (as well as the US markets) had expected this soon in the AI race. With the introduction of DeepSeek R1, High-Flyer, the company behind DeepSeek, was able to generate a similar wave, if not more.
Dethroning ChatGPT, taking the number one spot in the App Store, and causing a loss of $1 trillion in US markets – this is what you need to know about DeepSeek AI.
DeepSeek Is Pioneering Open-Source AI Innovation in China
Image source
DeepSeek, officially named Hangzhou DeepSeek Artificial Intelligence Basic Technology Research Co., Ltd., is a Chinese AI startup founded by Liang Wenfeng. Liang Wenfeng is also the co-founder of the hedge fund High-Flyer, which solely funds the DeepSeek operations. Wenfeng ventured into the AI race with the vision of creating innovative, efficient, and easily accessible models – with a fresh perspective and full of liberty without investor pressure – thanks to this funding model.
Based in Hangzhou, Zhejiang, the team at DeepSeek AI has a talented mix of young graduates fresh off from top universities in China. The hiring process Wenfeng followed with DeepSeek AI focused on prioritizing technical excellence rather than work experience in some big companies. Wenfeng wanted a team that could look at AI from an entirely new angle, find the current problems, and work on that, which they did.
The company took off in November 2023 with its introductory product, DeepSeek Coder, a model that could quickly generate, complete, and debug the codes. DeepSeek Coder’s launch was an open-source release under a permissive MIT license, which meant developers could review, modify, and build upon the model without any restrictions. This model was developed using a large corpus of programming data and natural language.
DeepSeek was a hit among the coders, and it also built the foundation on which the company would launch future innovations in AI-driven software development. After DeepSeek Coder, the company launched an LLM model – DeepSeek LLM. Built on 67B parameters, DeepSeek LLM was launched as the next step to compete in the market and to further the company’s efforts to open-source AI. Trained on a dataset of 2 trillion tokens in both English and Chinese, the model is designed to excel in various tasks, from natural language understanding and generation to complex reasoning and even coding assistance.
To top its last innovation, in May 2024, the company launched a new version of its LLM project, named DeepSeek V2. The product quickly got attention for its lower costs and firm performance. The aggressive pricing also shook other major tech companies like Tencent, ByteDance, Alibaba, and Baidu, which were also in the AI race and struggled to reduce their AI model prices to stay competitive.
Following DeepSeek-V2, the more sophisticated DeepSeek-Coder-V2 was introduced – built on 236 billion parameters. This model can handle complex coding and can support 128k tokens of user context. It is offered through an economical API, with pricing set at $0.14 per million input tokens and $0.28 per million output tokens.
And coming back to the present, the recent launches of DeepSeek-V3 and DeepSeek-R1 have cemented the company’s reputation as a market disruptor. The DeepSeek-V3 model, with 671 billion parameters, delivers standout performance across multiple benchmarks while using far fewer resources than similar models. Meanwhile, the DeepSeek-R1 debuted in January 2025, targets reasoning tasks, and competes with OpenAI’s o1 model through its enhanced features. The R1 model has also been lauded for advanced capabilities in reasoning, mathematics, coding, and other general tasks.
DeepSeek AI has introduced a series of streamlined models under the DeepSeek-R1-Distill label. These models are derived from widely used open-weight models such as Llama and Qwen and refined using synthetic data produced by R1. Offering diverse performance and efficiency levels, these distilled models are designed to meet various computational demands and hardware setups.
How DeepSeek AI Was Able to Pull Off a ChatGPT-Like Model for Under $6 Million?
Coming to the costs related to training its models, many were in shock, and some even refused to believe the figures presented by DeepSeek altogether. Comparing it to OpenAI’s most recent collaboration with partners like SoftBank and Oracle, announcing the Stargate Project – a $500 billion investment in AI infrastructure.
DeepSeek has stressed that their training regimen for AI models, like DeepSeek V3, uses just around 2,000 Nvidia H800 GPUs. The AI training at DeepSeek lasts 55 days and costs $5.8 million. Whereas other companies like OpenAI and Meta use thousands of GPUs and supercomputers, significantly building up on the costs.
Additionally, their R1 model uses the MoE (mixture of experts) method, where a selective use of computing resources is utilized based on the specific requirements of a task. This strategy boosts efficiency and lowers energy usage, contesting the belief that developing advanced AI demands hefty computational and financial investments.
That’s why DeepSeek’s API prices are much lower than those of its competitors, making it a good choice for small businesses and independent developers. For example, the DeepSeek-R1 API costs only $0.55 per million input tokens and $2.19 per million output tokens, while OpenAI charges $15 and $60. DeepSeek also follows open-source principles. This means there are no licensing fees, and developers can use and modify the models freely, reducing costs and promoting broader use of advanced AI.
On the other hand, OpenAI’s approach involves large-scale partnerships and massive infrastructure investments. This strategy serves broader goals, like handling a high volume of users and offering a suite of AI services beyond simple text generation. For example, OpenAI recently launched AI agents (or AI Operators), which can independently browse networks and do tasks for you, like booking a restaurant reservation, buying movie tickets, etc. OpenAI also focuses on AI image and video generation with Sora’s recently launched platform, where you can generate real-life videos based on your text input.
On the other hand, DeepSeek’s smaller, more targeted setup focuses on efficiency and cost-cutting with only a chatbot, unable to generate images or videos or operate beyond the scope of its network.
Reactions from US Companies
The emergence of DeepSeek has elicited varied responses from US tech companies. Microsoft CEO Satya Nadella and OpenAI CEO Sam Altman have described DeepSeek’s achievements as “super impressive,” acknowledging the startup’s innovative approach.
The US government has also noted that President Donald Trump referred to DeepSeek’s rise as a “wake-up call” and a positive development. This sentiment underscores the broader implications of DeepSeek’s success on the global AI landscape and the competitive dynamics between the US and China.
Is DeepSeek AI Merely a ChatGPT Copy?
Whether DeepSeek is simply a replica of ChatGPT has been a focal point of discussion. OpenAI CEO Sam Altman has acknowledged DeepSeek’s impressive performance but has not directly addressed the issue of potential replication. On X, Altman praised the models of DeepSeek and hinted that they would soon launch more capable and better models. This was just before the introduction of o3 mini in the free model, which comes with the ‘thinking’ ability to reason the query before generating a prompt.
The open-source nature of DeepSeek’s models (whereas ChatGPT is closed-source) complicates the matter, as it encourages adaptation and iteration, which are standard practices in the AI community. It’s essential to recognize that AI development often involves building upon existing frameworks and models. OpenAI also used publicly available online resources to train the program during its research.
DeepSeek was built as a cost-effective, open-source alternative with its unique twists. It uses different training methods and hardware optimizations – like a MoE approach and lower-cost GPU setups – to deliver comparable results at a fraction of the expense. In short, DeepSeek may look similar on the surface, but it was developed to be leaner and more efficient rather than just replicating ChatGPT’s design.
Instead, it may be seen as a product that shares some standard techniques with ChatGPT – widely used in the AI community – but its overall design, implementation, and training method might still contain distinct elements.
Why Was There a Market Breakdown Following DeepSeek’s Launch?
DeepSeek’s breakthrough lies in its ability to deliver competitive performance at a fraction of the cost and resource investment of established AI models. This efficiency challenged the prevailing notion that cutting‑edge AI requires massive financial and hardware resources, and it upended long‑standing business models built on high expenditure for advanced technology.
Investor panic soon followed as the emergence of DeepSeek – hailed by some as an “AI Sputnik moment” – triggered fears of a radical shift in the competitive landscape. When news broke, shares in key companies such as Nvidia plummeted nearly 17% daily, wiping out hundreds of billions in market capitalization and contributing to over $1 trillion in losses across US tech stocks. This disruption was compounded by geopolitical concerns: the success of a Chinese startup using older, less‑restricted hardware underscored potential weaknesses in US export controls on advanced AI chips, raising alarm about a possible realignment in global AI leadership.
Security and Ethical Considerations
The rapid adoption of DeepSeek’s AI models has sparked robust discussions about security and ethical implications. Operating within China’s regulatory framework, DeepSeek faces scrutiny over data privacy, censorship, and potential government oversight.
An internal email from the US Navy warned staff against using the DeepSeek app, citing ethical and security concerns given the model’s origin and use.
DeepSeek asserts that its models comply with local laws and regulations, including implementing content moderation to avoid politically sensitive topics. Yet, this legal conformity raises essential questions about whether the company can balance meeting regulatory demands and preserving the principles of free expression and user autonomy.
Data safety specialists advise caution when using the tool, as it gathers extensive personal information and stores it on servers in China. Additionally, DeepSeek reported facing cyber-attacks; on Monday, the company announced that it would temporarily restrict new user registrations due to “large-scale malicious attacks” targeting its software.
Furthermore, the open-source nature of DeepSeek’s models – while building transparency and collaborative innovation – also introduces notable security challenges. Benevolent researchers and malicious actors can examine publicly available source code, potentially increasing the risk of exploitation if vulnerabilities are not promptly identified and addressed. Ultimately, ensuring robust security and ethical integrity requires continuous community oversight and the implementation of strong safeguards to protect against misuse while promoting open innovation.
About DeepSeek
DeepSeek, officially known as Hangzhou DeepSeek Artificial Intelligence Basic Technology Research Co., Ltd., is a Chinese artificial intelligence company founded in July 2023 by Liang Wenfeng, co-founder of the hedge fund High-Flyer. Based in Hangzhou, Zhejiang, DeepSeek specializes in developing open-source large language models (LLMs) and has rapidly gained attention for its innovative and cost-effective AI solutions. Notably, the company has made significant strides in creating AI models that rival established tech giants, achieving comparable performance at a fraction of the development cost.
In November 2024, DeepSeek released its DeepSeek-R1 model, which delivers responses on par with contemporary LLMs like OpenAI’s GPT-4, while being trained at a significantly lower cost of $5.8 million compared to the $100 million reportedly spent on GPT-4’s development. Following this, in January 2025, the company launched a free chatbot app based on DeepSeek-R1 for Android and iOS platforms. By late January, this app had overthrown ChatGPT as the most downloaded application (free) on the US App Store, underscoring DeepSeek’s rapid ascent in the AI industry.
Conclusion
DeepSeek has topped the App Store charts and sparked heated debate in the AI community. It has shown that focusing on cost efficiency and an open-source approach can produce an AI tool that performs comparably to established models like ChatGPT.
At the same time, questions remain over whether DeepSeek truly represents a breakthrough or reuses familiar techniques in a more efficient package. Industry leaders and experts are closely watching the response from users, the impact on market investments, and how well DeepSeek handles challenges such as censorship and data security. Ultimately, the future will reveal whether this development shifts the balance of power in AI or if it is just another step in a rapidly evolving field.
