One of the primary objectives of the Trump administration is to make United States of America (US) the artificial intelligence (AI) capital of the world. This is supported by the fact that President Donald Trump signed an executive order ordering the revocation of existing AI policies and directives implemented by the Biden administration to remove barriers to AI innovation. Just one day after being sworn in, Trump announced the ambitious Stargate Project with the intention of infusing US$ 500 billion over the next four years in a new company called ‘Star Gate’ that will aim to build AI infrastructure in US. This will be achieved with the help of OpenAI, SoftBank, Oracle, Nvidia, Microsoft and others. This indicates the US’s sentiment that investing billions of dollars’ worth of capital into AI is the only way to win big. However, on January 27, 2025, this sentiment was completely blown out of the water when the US stock market crashed, with Nvidia losing about 17% or close to US$ 593 billion in market value, which is a record for highest one day loss for any US company. Keeping aside the fact that AI related stocks have been deviously bullish since the launch of ChatGPT in November 2022, it still does not explain how Nvidia, which has been one of the highest gainers in the AI revolution, lost so much so quickly.
The reason for the historic sell-off was the launch of DeepSeek’s open-source language model ‘DeepSeek R1’. DeepSeek is a Chinese artificial intelligence company led by Liang Wenfeng and is funded by a hedge fund founded by engineers from Zhenjiang University. The shocking part is that the model developed by DeepSeek is free and open-source and yet Liang Wenfeng claims that DeepSeek is profitable, which is something OpenAI has not been able to achieve yet. What is even more unusual is that DeepSeek had already announced before release of their model that their model was developed at an abysmally low cost and also claimed that their final training run for the model cost DeepSeek US$ 5.6 million which was approximately 5% of what OpenAI spent on their o1 model. Initially, these tall claims were met with scepticism and scoff by industry experts and investment community given that it was a Chinese company making such claims. However, when R1 released on 20th January 2025, which was the same day Trump got sworn in (coincidence?), the industry experts and public at large finally got the chance to evaluate the model in real time. Just as the evaluations started confirming that R1 is indeed competitive, the market was quick to react and the sell-off began. Nvidia was not the only loser in this situation, as collective US$ 1 trillion was wiped out from the shares of companies exposed to AI. One might ask, was this truly a market correction coerced by a Chinese AI model? Or was the AI ‘bubble’ prone to such correction, given that it was evident that the AI bandwagon in the investment community was fuelled by sheer ignorance of the investors who have absolutely no idea why they are blindly investing into AI stocks in the first place? Although the stocks are slowly regularising, there have been early indications and hints that companies like Nvidia have been overvalued and that there was a need for reassessment of the competitive AI landscape.
DeepSeek had convincingly upended the notion that you need significant amount of capital for training models. However, the grass is always greener on the other side, as there have been claims by AI experts that DeepSeek is not being entirely truthful about their claims. While initially, CEO of OpenAI Sam Altman posted on X that DeepSeek was an impressive model, and that it is invigorating to have a competitor, the sentiments have quickly shifted with many experts claiming, including David Sacks who is Trump’s AI czar that DeepSeek has committed IP theft by distilling data out of OpenAI’s models. Even OpenAI publicly stated that they have found evidence that DeepSeek was using distillation which is a technique used by developers to extract data from larger more powerful models. Apparently, OpenAI got a taste of its own medicine, since OpenAI has been notoriously involved in a series of lawsuits globally wherein they are accused of scrapping data including copyrighted material off the internet to train their models. However, this time, OpenAI seems to be the victim as they are the ones who allegedly got their data scrapped by DeepSeek.
One might ask, what does this have anything to do with Nvidia and why is Nvidia the one who lost the most in this situation? The answer is simple, Nvidia’s chips are quintessential for compute in development of AI models. The inflation in Nvidia’s market value was attributable to the investment theory that whoever comes out on top in the AI revolution will inevitably require Nvidia’s chips for scalability purposes. As these companies start scaling their models, they require more compute, and therefore more powerful and expensive chips. However, DeepSeek has shown the world that the job can be done in a cost-efficient manner by using the chips in an effective manner, which completely shattered the theory dictating investments in Nvidia stocks, and therefore its prices dropped. Notwithstanding the aforesaid, it is still a fact that Nvidia’s chips are ‘crucial’ as if almost indispensable, so much so that the Biden Administration had to put global sanctions restricting the access to AI chips to other countries, with the maximum restrictions on China and Russia. In relation to sanctions and ongoing controversies relating to DeepSeek, Scale AI’s CEO Alexdr Wang speculates that DeepSeek is hiding 50,000 Nvidia H100 chips stockpile which DeepSeek will not/cannot talk about, due to strict US export controls. Elon Musk seems to share the same sentiment as Alexdr Wang; however, these are just conjectures and there has been no statements from DeepSeek disproving these claims.
Regardless of whether DeepSeek’s claims, or the AI experts’ speculations are true, the biggest takeaway in this situation is that the AI revolution is no longer a one-sided hop skip and jump for companies like OpenAI, who are now facing competition and pressure from Chinese companies who are notoriously resourceful and ostensibly innovative in the face of US government sanctioned restrictions. In this AI war, Nvidia’s role is pivotal, and it will be interesting to see how they address the demand for their chips while trying to remain relevant in a fairly competitive AI ecosystem.
Author: Amartya Mody