The Cyber Attacks on DeepSeek Have Escalated Again!
Recently, DeepSeek continues to gain popularity, but the cyber attacks it has suffered have also intensified. In the early morning of January 30, the second day of the Lunar New Year, Qi’anxin XLab detected a significant escalation in cyberattacks on DeepSeek’s online services. The number of attack commands surged by hundreds compared to January 28.
It is worth noting that DeepSeek has revolutionized the world with its AI breakthroughs, bringing a disruptive cost advantage. This trend has already had a notable impact on the capital market. DeepSeek has placed some pressure on the performance expectations and valuations of existing AI hardware companies worldwide, causing sharp declines in stocks of companies like Nvidia, Broadcom, and Micron Technology. On the other hand, DeepSeek has significantly lowered the application threshold of AI technology by providing low-cost, high-performance AI models. As a result, the company’s stock price continues to rise, benefiting from reduced AI inference costs.
Some securities analysts have pointed out that DeepSeek’s high cost-effectiveness not only challenges the dominance of the US AI models but also has a positive effect on domestic confidence and sentiment regarding independent control.
The Intensity of Cyber Attacks Has Increased a Hundredfold
According to Qi’anxin Group, in the early morning of January 30, Qi’anxin XLab detected that the intensity of attacks on DeepSeek’s online services suddenly escalated, with the number of attack commands increasing by hundreds compared to January 28. XLab observed that at least two botnets participated in the attack, launching two waves of attacks.
XLab security experts stated: “Initially, the attacks involved amplification techniques such as SSDP and NTP reflection, which are designed to overwhelm systems by amplifying traffic. On January 28, a large number of HTTP proxy attacks were added. Starting this morning, botnets entered the scene. As the number of attack methods increased, prevention became more difficult, significantly raising the security challenges faced by DeepSeek.”
XLab has been monitoring DeepSeek’s network for nearly a month and found that the attack methods have evolved from initial amplification attacks (which were relatively easy to clean) to HTTP proxy attacks on January 28 (an application layer attack, which is more difficult to defend). These attacks have now evolved into botnet-based systems. The attackers have used various techniques to continuously target DeepSeek.
In the early morning of January 30, XLab observed that two Mirai variant botnets, HailBot and RapperBot, participated in the attack. The attack involved 118 C2 ports across 16 C2 servers, divided into two waves at 1 a.m. and 2 a.m. respectively.
“The addition of botnets indicates that professional attackers have become involved,” said XLab. “This shows that the attack methods targeting DeepSeek have evolved and become more complex, making defense more difficult. The situation is growing increasingly complex and severe.”
A botnet is a network of infected devices controlled by attackers through malware. These devices, called “zombies” or “robots,” receive commands from the attackers’ command and control (C&C) server. The attackers then use the botnet to launch DDoS attacks on target servers, increasing the scale and intensity of the attacks and exhausting the target’s network bandwidth and system resources, leading to service disruption or paralysis.
Impact and Investment Opportunities Brought by DeepSeek
Shortly after launching the R1 model, DeepSeek gained widespread attention due to its cost-effectiveness, open-source nature, and improved reasoning capabilities. On New Year’s Eve, DeepSeek also launched new models, among which Janus-Pro-7B defeated OpenAI in benchmark tests, earning the moniker “the mysterious oriental power” from many online commentators.
The explosion of DeepSeek has revolutionized the world of artificial intelligence, and the market structure has “overturned” previously dominant technology stocks. On January 28, US chip giant Nvidia saw a nearly 17% drop, with its market value evaporating by over $590 billion in a single day. Broadcom, Oracle, AMD, and Micron Technology also experienced sharp declines.
Zheshang Securities analysts Liao Jingchi, Wang Daji, and others pointed out that DeepSeek’s large model has a disruptive cost advantage, which could impact the traditional R&D path focused on “high investment and high computing power.” The growing demand for computing power may shake market expectations and negatively affect the performance forecasts and valuations of existing AI hardware companies. If DeepSeek’s innovative R&D model, which emphasizes “spending little to achieve great results,” continues to prove effective, we may need to be cautious about the potential risk of downward earnings and valuation revisions for tech giants like Nvidia.
The cost-efficiency advantage of the DeepSeek model could also benefit the implementation of downstream AI applications and commercial scenarios. The aforementioned analysts suggested that DeepSeek has greatly lowered the threshold for AI technology applications, enabling more downstream companies and developers to adopt advanced AI tools at lower costs. For instance, Apple has benefited from reduced AI inference costs, allowing it to integrate more AI applications into its products like smartphones, computers, and VR glasses, enhancing user experience and strengthening its competitive position. On January 27, Apple’s stock price rose 3.18%, followed by another 3.65% increase on January 28, and 0.46% on January 29.
From an investment perspective, Zheshang Securities stated that DeepSeek’s AI large model is expected to significantly reduce the cost of AI applications, accelerating the development of related industries and benefiting the implementation of downstream AI applications such as robotics, computers, and media. Additionally, DeepSeek’s high cost-effectiveness not only challenges the US AI model monopoly but also boosts domestic confidence and sentiment in independent control. Investors are encouraged to monitor sectors such as TMT, military, and high-end manufacturing for growth opportunities.
DeepSeek’s release of the R1 model and related applications, which achieved results comparable to existing cutting-edge models at a much lower training cost, has triggered market concerns about computing power investment. Huang Leping, an analyst at Huatai Securities, believes:
- DeepSeek’s main innovation is the addition of reinforcement learning during the pre-training stage. The training cost of DeepSeek V3 is only 7% of the Llama3 series, significantly contributing to the reduction in AI model costs and is expected to lower both training and inference costs for existing models.
- The four major AI companies in North America are exploring next-generation large models by expanding GPU clusters. It remains to be seen if DeepSeek’s approach will succeed in the development of future models.
- DeepSeek’s success suggests that the gap between China and the US in large model technology may narrow, especially as the Scaling Law slows down.
Huafu Securities analyst Yang Xiaofeng pointed out that DeepSeek’s upgrade from V3 to R1 will likely accelerate the improvement of large models across the industry, allowing for faster catching up with GPT-4 and GPT-1. At the same time, cost reductions will help popularize AI applications and terminals. In the future, industry competition will focus on “product capability,” with the market advantages of large tech giants becoming more pronounced. The global AI competition era is approaching, and we are optimistic about the rise of Chinese large models represented by DeepSeek, particularly in the development of AI applications and terminals. It is expected that inference costs will continue to drop, and domestic substitution will drive the growth of computing power.
