What Is DeepSeek and How Should It Change How You Invest in AI? | Global News Avenue

this Generated AI The United States industry is receiving a wake -up call. In the past few years, with the prosperous stocks of artificial intelligence, their stocks have risen. That’s because a Chinese startup DeepseekRegarding the construction of high -end AI models and the subversion of traditional wisdom in what cost construction.

The company reported in early 2025 that its model and Openai’s chat GPTAll of these are reported costs of $ 6 million. This triggered a record of $ 600 billion Decline in a single day In NVIDIANVDA) Stocks and investors re -consider their AI -based bets.

For anyone who invests in AI, understanding the rise of DeepSeek is essential for navigation in the new era in this field. We let you work hard below.

What is Deepseek?

DeepSeek is an artificial intelligence research company located in Hangzhou, China. It was established in July 2023 from the former hedge fund executive, and was supported by the quantitative investment giant High-Flyer Quant. It prioritizes algorithm efficiency and open source collaboration to challenge the AI ​​advantage of American technology giants.

Since its launch, Deepseek has released a series of impressive models, including Deepseek-V3 and Deepseek-R1, which said that it matches the O1 reasoning capabilities at a small part of the cost. In addition, the model of DeepSeek is Open sourceThis means that they can be used freely, and anyone can use, modify and distribute.

The most important thing is that for competitors, the most important thing is the most destructive-although we Export restrictions on advanced AI chipsFor example, the H100 and A100 models of NVIDIA.

In fact, these restrictions may be the heavenly gift of the company’s breakthrough. Deepseek said it found a solution to improve the ability of chip and adopted other key strategies. Competitors may now study in detail:

• Mixture of Experts (MOE) architecture: Each task only activates a part of the parameter (for example, only 5 % of all available token), and the calculation cost is reduced.
• Enhance learning: Repeated feedback training models to improve reasoning skills.
• Distillation technology: Compress large models (such as GPT-4) to small, cheap versions.

This merger method enables the company to use about 2,000 NVIDIA GPUs to train its models within 55 days, and the cost is about $ 5.6 million, which is a small part of the US technology giant expenditure.

How does this affect AI investment?

Deepseek’s success challenged the general idea, which encouraged the large amount of investment in the United States AI-AI development requires countless cash to pay NVIDIA chips and other expensive expenditures.

The main scientific stocks in the United States have dropped sharply in terms of the development of these situations. NVIDIA’s stock plummeted Nearly 17 % is the largest single -day loss in the history of the US stock market. Other stocks in AI spaceIncluding Microsoft Corporation (MSFT), Alphabet Inc. (Googl) And ASML holding NV (ASML), Also plummeted.

Deepseek also created opportunities for investors in the AI ​​field. This is what savvy investors may do:

1. Consider the “bigger and better” mentalityEssence Investors may want to find companies investing in more efficient training methods and energy -saving technology, rather than those companies that blindly expand their capital -intensive GPU clusters. Companies using scattered GPU clouds may have advantages.
2. Prepare for the surge in AI access: The obstacles adopted by AI seem to be declining. Investors should pay attention to emerging AI applications and companies that develop AI products in real worlds, not just betting basic model providers.
3. Focus on software: Although investors have driven AI -related chip manufacturers like Nvidia to record HIGH, the future of AI It may rely more on software Change more expensive hardware.
4. Monitor geopolitical risks: Deepseek’s success may exacerbate the tension of technology in the United States and China. Investors should monitor the change of policy, such as more stringent export control.

DEEPSEEK’s incorrect problem

Investors should also maintain the latest state because experts look at DeepSeek. A early study of NewsGuard evaluated the credibility of news and information websites, including the reason why the reliability of DeepSeek’s reliability. Despite the download of the App Store, the accuracy test of the Chinese AI chat robot failed 83 %. It placed it at the bottom of the evaluated AI chat robot and ranked 10th among 11 competitors.

Newsguard’s evaluation found several key issues:

• 30 % reply contains completely false information
• 53 % of answers provide non -requests for query
• Only 17 % of the reply successfully revealed the false proposition
• Even if answering questions that have nothing to do with China, chat robots tend to plunder the Chinese government’s position, such as diplomatic positions to China -free inquiries.

Therefore, the accuracy method of DeepSeek seems to have shifted responsibility for users, and their use clauses notify them to “actively verify the authenticity and accuracy of the output content.”

Bottom line

Deepseek may be a omen for a cheap future AIEssence This may mean that it is necessary to pay attention to the increasingly expensive technology, open source collaboration and scalable infrastructure brute force. But this also means to check the headline news that exceeds speculation and evaluate whether DeepSeek provides new and different things, or considering its ability to perform some early tests, whether it is just an illusion generated by another AI.