DeepSeek: Revolutionizing the AI Landscape

In the ever-evolving realm of artificial intelligence, the emergence of DeepSeek has sent shockwaves through the industry, sparking intense discussions and reshaping the competitive landscape. This article delves into the rise of DeepSeek, its impact on the global AI ecosystem, and the challenges and opportunities it presents.

The Genesis of DeepSeek DeepSeek, a relatively young AI company founded on July 17, 2023, has rapidly gained prominence with its groundbreaking advancements. In the lead-up to the 2025 Spring Festival, the company made headlines by releasing two significant open-source models: V3 on December 26, 2024, and R1 on January 20, 2025.

The V3 model, boasting performance comparable to closed-source models like OpenAI's GPT-4o and Anthropic's Claude-3.5-Sonnet, outshines the open-source Meta's Llama 3. What's even more remarkable is its relatively low total training cost of just $5.576 million. The R1 inference model, on the other hand, comes close to OpenAI o1 in performance while offering an API price that is a mere 3.7% of OpenAI o1.

This achievement is even more impressive considering DeepSeek's status as a startup. With tens of thousands of NVIDIA chips at its disposal, the company has managed to train high-performing large models at approximately 7% of the cost of its overseas counterparts. Since the release of its V2 model in May 2024, DeepSeek has been at the forefront of the price war in the Chinese large model market, and by the end of the year, it had extended this battle overseas.

The Controversy Surrounding DeepSeek DeepSeek's rapid ascent has not been without its fair share of controversy. Following its meteoric rise to fame, Silicon Valley giants have raised concerns. OpenAI claims to have found evidence suggesting that DeepSeek may have "distilled" its models, while Anthropic's founder and CEO, Dario Amodei, has publicly denied the breakthroughs achieved by R1 and called for stricter controls on the export of computing power to China.

Amidst the controversy, it is essential to objectively evaluate DeepSeek's capabilities. In its technical papers for V3 and R1, DeepSeek has presented several innovative features. The V3 model incorporates multiple self-developed technologies for architectural innovation, such as the DeepSeekMoE + DeepSeekMLA architecture and MTP multi-Token prediction technology, enabling low-cost training. The R1 model, on the other hand, abandons the HF part in traditional RLHF (Reinforcement Learning from Human Feedback) and directly trains through pure reinforcement learning (RL), validating the priority and effectiveness of RL and further optimizing training efficiency.

However, some industry experts have pointed out that the $5.576 million figure primarily represents the GPU cost for model pre-training. When considering factors such as server capital expenditure and operating costs, DeepSeek's total cost could reach $2.573 billion over four years. Additionally, the trend of decreasing innovation costs has been ongoing, and DeepSeek has merely accelerated this process. Before DeepSeek, the cost of artificial intelligence training was already decreasing by 75% annually, and the inference cost was dropping by 85% to 90%.

The Impact on the AI Industry DeepSeek's emergence has had a profound impact on the AI industry, disrupting the established order and forcing companies to reevaluate their strategies.

Chatbot AI applications have been among the first to feel the impact. According to data from the AI Product Ranking, around the 2025 Chinese New Year's Eve, DeepSeek's daily active users exceeded 20 million, surpassing domestic competitors like Doubao and Kimi to claim the top spot in China. It achieved 100 million users in just one week, a feat that took ChatGPT two months to accomplish.

In response to DeepSeek's success, companies like Yuezhi Anmian and Doubao have launched new features and updates. However, their efforts were overshadowed by DeepSeek's rapid rise, resulting in a decline in their daily active users. This incident highlights the low loyalty of users in the chatbot market, as they quickly switch to more powerful, affordable, and efficient models.

Looking beyond chatbot applications, DeepSeek has also influenced the landscape of self-developed large model companies. From an investor's perspective, the release of DeepSeek's V2 model in May 2024 marked the beginning of a significant shift in the industry. Among domestic giants, Alibaba's Qwen is considered one of the best-performing models, while Doubao has shown significant improvement in the second half of 2024. Among startups, DeepSeek and Yuezhi Anmian (Kimi) have witnessed the fastest growth, while the growth of other companies, such as Lingyi Wanwu, MiniMax, Baichuan Intelligence, Zhipu AI, and Jieyue Xingchen, has gradually slowed down.

The Future of DeepSeek As DeepSeek continues to make waves in the AI industry, its future remains uncertain. The company's success has attracted the attention of major players, with rumors of Alibaba considering a $1 billion investment for a 10% stake in DeepSeek. Although Alibaba has denied these rumors, the market remains vigilant, fearing that DeepSeek may enter into strategic partnerships that could potentially alter its independent stance.

On a broader scale, DeepSeek's rise represents a shift in the AI industry's approach. Traditionally, companies have followed the "computing power arms race" paradigm, investing heavily in technology, capital, and computing power to push the performance of large models to new heights. However, with the realization that the supply of high-quality training data may be limited, the "algorithm efficiency" paradigm, which focuses on optimizing efficiency through architectural innovation and engineering capabilities, has gained traction.

DeepSeek's open-source models have challenged the dominance of established players, forcing them to adapt or risk being left behind. While some industry experts caution against over-optimism, acknowledging that competition remains fierce and gaps still exist, there is no denying that DeepSeek has set a new standard in the industry.

In conclusion, DeepSeek's emergence has disrupted the AI landscape, sparking innovation and competition. As the industry continues to evolve, it will be fascinating to see how DeepSeek and its competitors navigate the challenges and opportunities ahead. Whether DeepSeek can maintain its momentum and continue to revolutionize the field of AI remains to be seen, but one thing is certain: the era of AI is here to stay, and DeepSeek has played a significant role in shaping its future.