Just as the AI market seemed destined to grow infinitely, news has emerged that has caused a major shock.
Just as the Chinese hedge fund company QuantConnect's AI research company Deepseek released an open-source language model called Deepseek-V3 at the end of 2024, it has sparked a significant reaction by showing results comparable to existing LLM services.
However, with many AI services already emerging that are on par with Chat-GPT.. why is there suddenly such a commotion?
As I write this, Nvidia's stock price, considered a leading stock in the AI sector, has plummeted by as much as 14.68%.
As well as Nvidia, Western stocks that were profiting from AI have all fallen in unison, causing the Nasdaq to drop nearly 3% at the current moment, truly delivering the Deepseek shock to the world.
Why is Chinese AI Deepseek impacting the US tech market?
So, why is this happening?
According to Deepseek's claims, they have revealed that they train and serve their model at a cost of only 1/10 that of Western AI-related companies. If this is true, it seems that the high-performance-centric AI development led by Nvidia, Meta, MS, and others in recent years, which argued that 'better AI requires better hardware performance!!', could be negated, and this reaction may also be affecting stock prices.
Source : https://zdnet.co.kr/view/?no=20250127133123
Of course, since this content is a claim from Deepseek, we need to look into the actual process used to create this AI model and the actual cost, but... it seems to be a significant shock to the US and its tech companies, who have been viewing AI technology as a national strategic project and trying to monopolize related technology by not selling high-performance GPUs to China for the past few years.
Is an extremely expensive GPU absolutely necessary for better artificial intelligence?
Those interested in the development of artificial intelligence, or those who have been interested for quite some time, will likely be greatly surprised by the explosive growth of the AI market recently, while also wondering if high-end chip development, as Nvidia claims, is truly the main driver of AI advancement.
Actually, when I was preparing to give a lecture on 'The Fourth Industrial Revolution and AI' at university two years ago, I looked up various scholars' opinions and related content from that time, and I was able to find quite interesting content...
Domestically, one could find Jim Keller, known as the 'Midas hand of chip development,' discussing the direction of development for TensorTorrent, which he represents. He mentioned that while Nvidia's incredible GPU technology is truly amazing and leading the era, it is unclear whether AI-related chips and technologies that are so expensive and resource-intensive are truly necessary for AI development.
Of course, that presentation was a bit different from the recent Deepseek story and was about the approach to hardware, but it gave me quite a lot of insights on various levels. Unfortunately, when I talked about the related content with people around me who are involved in AI or interested in future technology, it was dismissed as a very fringe opinion.
(After all, recent years of Nvidia's dominance and stock price surge were too explosive for that opinion to gain acceptance.. )
However, cautiously predicting that this recent Deepseek-related issue may lead to the rediscovery of approaches and reactions that were previously dismissed as a minority opinion in the AI field.
AI, 이제는 증명을 할때, 그리고 우리에게는 기회가 올 수도?
Undeniably, the development of AI is excellent. Over the past few years, AI-related technologies and services centered around various tech companies like MS, Meta, Nvidia, and OpenAI have developed dazzlingly, and I myself have been widely using and utilizing AI when organizing articles, creating Photoshop images, or working on game-related tasks.
However, this current development is actually a result created by looking at future value amidst a tremendous amount of investment and loss.
OpenAI, which is highly regarded as leading the AI technology, recently announced that it expects losses of over 7 trillion won this year alone, and announced that cumulative losses will reach 64 trillion won by 2028.
Source : https://www.etnews.com/20241230000196
Even forecasts have emerged that Microsoft will be completely acquired due to financial difficulties within three years..
Throughout the second half of last year, the 'AI bubble theory' was consistently raised, eventually leading to forecasts that 2025 would be the year to prove AI-related results. Consequently, many forecasts emerged that starting in 2025, AI companies would focus on delivering performance based on the infrastructure invested so far, rather than making indiscriminate investments.
However, as the Trump administration takes office with the perspective that AI development is just beginning, it was reported that the 'StarGate' project, which announced a massive investment of nearly 500 billion dollars (about 710 trillion won) in US AI infrastructure five days ago, is underway. It is evaluated as a move for the US to secure the hegemony of the upcoming 'Great AI Era' by OpenAI, Oracle, and Japan's SoftBank forming a joint venture to build data centers for AI within the US.
Of course, this is a project without substance, and there was also Musk's cold water issue that SoftBank or OpenAI didn't even have that kind of financial power..
Amidst such concerns, the sudden appearance of China's Deepseek seems to be pouring cold water on the AI-related initiatives of the Trump administration as well.
이를 반증하듯 엔비디아 말고도 스타게이트 프로젝트의 주역이 될 것으로 이름을 올렸던 소프트뱅크와 고성능의 반도체를 제작하기 위해서 필수적으로 갖춰야할 장비를 독점 공급하고있는 네덜란드의 ASML 의 주가도 장중 8% 가까이 폭락중이다.
Of course, today's crash could be just a fluke, and Deepseek's claims could be nothing more than empty talk, but looking back at the story Jim Keller told a few years ago... I can't help but wonder if, once again, Jim Keller was right all along.
虽然这是一个相当众所周知的故事，但由 Jim Keller 担任负责人的 TensorTorrent 是一家位于加拿大的硬件初创公司，有趣的是，其主要投资者是国内的巨头企业，而非西方公司。
Image source : https://www.kipost.net/news/articleView.html?idxno=324186
So, relatively, there are quite a few stories coming out domestically, and it seems there are also stories about being highly regarded as a bit more capable.
Regardless, following Jim Keller's footsteps, it is true that he has always participated in the innovative development of various chipsets, and if the big exclamation point and question mark Deepseek has thrown at the world this time is correct, that the leading AI-related technology with expensive solutions like Nvidia, which he has consistently advocated for over the past few years, was wrong, will this not give us, who have not been able to participate as a key member in the past few years amidst the dazzling development and growth of AI, a once-in-a-lifetime opportunity?
I don't know if the answer to that opportunity is Deepseek, or low-cost hardware like TensorTorrent, but... thinking that this might be the last chance for us, who do not hold an active position in the AI hegemony, regardless of what it is...
I hope that the numerous AI experts, institutions, and professors based in our country will carefully observe the situation and create good strategies and opportunities.