China’s DeepSeek has shaken the global AI industry with claims of developing a highly advanced model at a fraction of the cost incurred by US tech giants. However, skeptics are questioning its legitimacy.

The AI industry is witnessing a significant upheaval as DeepSeek, a Hangzhou-based startup, asserts that it has developed an advanced AI model at a fraction of the cost spent by US tech leaders like OpenAI and Google. This claim, if true, challenges the dominance of American AI companies, which have historically commanded billion-dollar investments to build cutting-edge models.
The revelation that DeepSeek trained its AI model R1 using just $5.6 million and 2,000 Nvidia H800 GPUs—a chip designed to comply with US export restrictions—has raised eyebrows across the industry. Comparatively, OpenAI’s GPT-4 is estimated to have cost over $100 million to train, using 25,000 high-end H100 GPUs.
The AI community is now engaged in intense scrutiny of DeepSeek’s claims, with some experts dismissing them as exaggerated or misleading. Meanwhile, Nvidia’s stock price plummeted 17%, erasing nearly $593 billion in market value, before partially recovering the next day.
DeepSeek’s Announcement and Its Claims
- DeepSeek was founded in late 2023 by Liang Wenfeng, a serial entrepreneur in the AI industry.
- The company unveiled DeepSeek V3, a 671 billion-parameter model, and DeepSeek R1, an advanced reasoning model.
- DeepSeek claims it used 2,000 Nvidia H800 GPUs and spent $5.6 million to train R1’s foundational model, V3.
- This sharply contrasts with OpenAI’s GPT-4, which reportedly cost over $100 million to train.
- The startup’s success has raised concerns about the effectiveness of US chip export restrictions aimed at limiting China’s AI advancements.
Skepticism from AI Experts and Industry Leaders
- Pedro Domingos, a professor emeritus at the University of Washington, expressed doubt about DeepSeek’s claims:
- “It’s very much an open question whether DeepSeek’s claims can be taken at face value. The AI community will be digging into them and we’ll find out.”
- He acknowledged that training a model with $6 million is plausible but added that it could be merely fine-tuning an already expensive model.
- Palmer Luckey, founder of Oculus VR, outright dismissed DeepSeek’s claims, calling them “bogus”.
- He accused the startup of engaging in propaganda to mislead American investors, stating:
- “It is pushed by a Chinese hedge fund to slow investment in American AI startups, service their own shorts against American titans like Nvidia, and hide sanction evasion.”
- Alexandr Wang, CEO of Scale AI, suggested that DeepSeek had access to 50,000 more advanced H100 GPUs, but he did not provide evidence.
- Elon Musk, a close ally of US President Donald Trump, also voiced skepticism, responding “Obviously” on X to a post questioning DeepSeek’s claims.
Financial and Market Impact
- DeepSeek’s announcement caused a sharp decline in Nvidia’s stock price, which dropped 17% on Monday, wiping out nearly $593 billion in market value—equivalent to Sweden’s GDP.
- However, on Tuesday, Nvidia’s stock rebounded by nearly 9%, as initial concerns began to fade.
- The Nasdaq 100 index, which is heavily weighted toward tech stocks, also saw a 1.59% recovery after its steep decline.
China’s AI Progress and US Export Controls
- DeepSeek’s rapid progress in AI has cast doubt on the effectiveness of US chip export restrictions.
- Zihan Wang, a PhD candidate who worked on an earlier DeepSeek model, dismissed skepticism, stating:
- “Talk is cheap. If they’d spend more time working on the code and reproduce the DeepSeek idea themselves, it would be better than talking on the paper.”
- In a past interview, Liang Wenfeng revealed that DeepSeek had stockpiled 10,000 Nvidia A100 chips before US export bans took effect.
Limitations and Censorship in DeepSeek’s Model
- Despite its technological advancements, DeepSeek’s AI model R1 reportedly censors topics that are politically sensitive in China, including:
- The 1989 Tiananmen Square massacre.
- The status of Taiwan.
Experts’ Analysis on DeepSeek’s Cost Efficiency
- Tim Miller, an AI professor at the University of Queensland, noted the uncertainty around DeepSeek’s claims:
- “The model itself gives away a few details of how it works, but the costs of the main changes they claim don’t ‘show up’ in the model itself so much.”
- He pointed out that while breakthroughs do occur, this development seemed “too good to be true.”
- Lucas Hansen, co-founder of CivAI, provided insights into why DeepSeek’s model may have been cheaper to train:
- “There have been a lot of algorithmic and hardware improvements since 2022, driving down the cost of training a GPT-4 class model.”
- He added that the real breakthrough was not the base model but the fine-tuning process, which is comparatively inexpensive.
Source: Al Jazeera and NDTV