TECHNOLOGY TRENDS
Jan 29, 2025
The Rise of DeepSeek
"Artificial intelligence can take the job of another artificial intelligence"
In recent months, as advancements in artificial intelligence (AI) have continued at an unrelenting pace, a question has frequently been asked: "Will AI take our jobs?"
This is a predictable and reasonable question. However, what is even more surprising is AI’s potential to take away the jobs of other AI systems. Moreover, these developments were announced from China on the same day the American president enthusiastically introduced the Stargate Initiative. It almost felt like a deliberate countermeasure.
This time, AI does not seem to be threatening human jobs, but rather shaking the throne of tech giants like OpenAI. The Chinese startup DeepSeek is making waves with its open-source language models, both democratizing technology and sending a clear message to its American competitors: "Affordable and effective AI is possible!"
What is DeepSeek and Why is it Important?
Founded in 2023 by Chinese entrepreneur and investor Liang Wenfeng, DeepSeek set out with the mission of "making artificial intelligence a tool accessible to everyone." But why is DeepSeek so important?
DeepSeek has proven that AI can be developed without massive budgets.
It has also demonstrated that small companies and developing countries can have a say in the field of artificial intelligence.
By embracing an open-source approach, it contributes to the faster advancement of technology.
Through the competition it creates, it makes AI services more affordable - and even free - for end users.
DeepSeek vs OpenAI: What's the Difference?
Both companies are racing towards the same goal, AGI (artificial general intelligence), but their paths are entirely different:
1. Open vs Closed Source Approach
While OpenAI and other U.S. companies sell their AI models as “black boxes”, DeepSeek makes its entire technology free and open-source.
For example, you can download and use DeepSeek’s models for free, whereas OpenAI requires a paid subscription. Because of this, OpenAI may lose a portion of its revenue in the coming period. However, this dynamic could shift again with future model releases.
2. Cost and Performance
Despite U.S. restrictions on chip sales to China, DeepSeek trained a 671-billion-parameter model in less than two years. This suggests that DeepSeek has achieved OpenAI-level results at a lower cost and without access to the best-performing chips
DeepSeek V3 scored the same as GPT-4 in math and coding tests, while its training cost was 20 times lower.
R1 performed on par with OpenAI’s o1 and outperformed o1-mini, an earlier model.
It seems that “cheap” does not always mean “low quality.”
Misunderstood Topics
The training in the final stage of DeepSeek V3 cost $5.6 million; however, this figure does not include the experiments conducted before reaching the final result. So the perception that it was "developed for $5.6 million" is incorrect. The training cost of R1 was likely higher, and how much it was remains unknown.
Over a period of 20 months, even if the average gross salary for 200 people is $1400, it would still amount to a cost of $5.6 million. Considering that most of the employees are PhD holders, this is not realistic at all. Similarly, the cost of training OpenAI's O1 model is also unknown. The only thing we know is that the training cost of GPT-4 is over $100 million.
DeepSeek may have used Nvidia’s high-performance H100 chips, which raises questions. If true, they might have found a way around U.S. export restrictions, which could explain why they have not publicly addressed this issue. Regardless, achieving this level of AI development in less than two years is a remarkable feat.
Panic in the Markets: Is It Real or Temporary?
In the United States, corporate value losses have reached $1 trillion in a single day. Companies specializing in AI language models, like OpenAI, are inevitably affected by this growing competition. However, the race has just begun. In the near future, new versions could emerge, potentially leaving DeepSeek behind as well.
I believe that much of the value loss in the tech sector is driven by emotional reactions. For many companies, this downturn is likely to be temporary. For instance, Nvidia’s only customer is not OpenAI. Even if DeepSeek doesn’t require extreme computing power, it still relies on Nvidia’s lower-capacity H800 chips.
The fact that DeepSeek has managed to develop such a cost-effective and high-performance AI model in just two years suggests that similar startups may emerge in other countries and companies. As the number of firms offering language models grows, demand for Nvidia and AMD’s chips is expected to increase, potentially boosting their long-term value.
Let’s assume I’m wrong. If companies like Nvidia, ASML, TSMC, and AMD start facing lower demand, my counterargument is this: these companies were previously selling their products at inflated prices due to high demand. A drop in demand could bring product prices down, making electronic equipment more affordable. As a result, firms could maintain profitability by selling more units at sustainable margins, ultimately increasing their long-term value.
No matter how you look at it, I believe the sharp declines in these companies’ stock values are driven by panic rather than fundamentals.
My Personal Evaluations
It is pleasing that this type of technological breakthrough comes from a different country. Because I believe that it is not healthy for the world for a single country to have absolute dominance over technology. The United States, which claims to be a leader in artificial intelligence, has recently acted with aggressive statements such as buying Greenland, viewing Canada as a state, and regaining control over the Panama Canal. Even to its own allies, it has acted with extreme statements and actions, such as limiting high-tech sales. I hope this development will calm them down a bit.
The restriction on high-tech exports to China is pushing China to develop its own products and infrastructure. Working with limited resources, China is overcoming these problems one by one with its own innovations, as seen in the examples of Huawei and BYD. Competition and a bipolar world will encourage the provision of higher quality but affordable products to end-users.
The U.S. has rendered domestic technology firms sluggish by putting them into hibernation with all the restrictions it imposes to be the sole leader in technology using political power. Methods that are far from efficiency and low cost have been prioritized. This created an ecosystem far from innovation. However, the situation in China is developing in the opposite direction. Companies are being pushed towards innovation and creativity due to all these restrictions.
The high processing power required by artificial intelligence language models has left many countries behind in this race. Even if technological adequacy is provided, the lack of energy infrastructure is pushing many countries back in this race, or rather, ending the race before it begins. However, this development has shown that these initiatives can be undertaken without requiring incredibly powerful energy infrastructures demonstrated.
It appears that language models that can operate with lower processing power will require less energy and cooling capacity. This also means lower costs. With the decrease in costs, the usage of artificial intelligence technologies will become widespread, which will lead to an increase in overall energy demand.
In conclusion, artificial intelligence has been promoted in a way that will increase the productivity of end-users. It is clear that this definition is correct, but incomplete. In fact, artificial intelligence will force not only the end-users but also the companies developing AI to increase their productivity.
The success of DeepSeek has been a sign that the world of artificial intelligence will not become monopolized. The upcoming months—perhaps even days—will determine how this competition will shape up. But one thing is clear: Artificial intelligence is no longer under the control of just a few major players, and this heralds significant changes in the technology world.
