DeepSeek has been praised by many US tech companies for its efficiency and is said to be on par with several established AI companies, costing only a fraction of the amount.
With the launch of ChatGPT in 2022, Chinese companies scrambled to develop an equivalent artificial intelligence (AI) chatbot model. When the first such model was developed and released by Baidu Beijing’s search engine giant, there was much disappointment due to the widespread gap between the AI capabilities of the US and China.
However, with the Hangzhou-based startup launching the DeepSeek AI models, the future of AI is gearing up for many swift changes. The company has been praised by many US tech companies for its efficiency and is said to be on par with several established AI companies, costing only a fraction of the amount.
According to a paper published by the company last month, training the DeepSeek-V3 model required less than $6 million worth of computing power from Nvidia H800 chips. This revelation triggered the global curiosity and excitement surrounding DeepSeek, which resulted in it becoming the top-rated free application on Apple’s App Store in the US, overtaking ChatGPT, which had previously held this record.
The Chinese start-up released a statement that the DeepSeek-R1 model released last week and the DeepSeek-V3 model are just as efficient as OpenAI and Meta‘s most advanced AI models. Along with quality, the company also stated that, depending on the task, DeepSeek-R1 is 20-50 times cheaper than OpenAI o1 mode.
According to Reuters, this start-up’s controlling shareholder is Liang Wenfeng, co-founder of quantitative hedge fund High-Flyer. Liang had announced in 2023 that they were on a mission to allocate resources to an independent group to research and develop artificial general intelligence (AGI). AGI is an autonomous system which exceeds human capabilities in performing economically valuable tasks. Chinese corporate records also suggest that while it is unclear how much High-Flyer has invested in this start-up, the company has a patent related to chip clusters used to train AI models.
DeepSeek’s low cost of production has also raised eyebrows in the tech industry, with experts and analysts questioning the reasoning behind US tech companies’ decision to invest billions of dollars into AI development. This doubt caused the stock market value of AI giants like Nvidia to plummet, as the top-ranked company recorded a loss of $593 billion, which is the highest loss in a single day for any company on Wall Street.
Nvidia, which was crowned the largest company in the world in November 2024, reported a 17% decrease in market capitalisation. Subsequently, shares of Microsoft, which backs ChatGPT, dropped by 2.1%, and Google’s parent company, Alphabet shares, also dived to 4.2%. The race to outdo the AI capabilities of others has led to a massive inflow of capital into the industry among private companies. This has inflated the valuation of many Washington-based tech companies.
However, if this company can achieve similar results at a much lower cost, this could completely overturn not just the tech industry but also the markets, which have been driven by the potential of AI for the past 18 months. Economists are predicting that if the production model of this particular AI software is applied to all other AI companies, then the demand for chips, data centres and massive power generation units would also reduce considerably.
The newly elected US President Donald Trump said on Monday that DeepSeek was a ‘wake-up call’ and a positive step towards enhancing AI development. Trump had previously unveiled a private-sector plan through a joint venture known as Stargate that could bring in $500 billion in investments towards AI infrastructure. This was followed by SoftBank pledging $19 billion to finance the Stargate venture. This venture is also backed by OpenAI and Oracle.
However, with Beijing’s recent innovation, these proposals are bound to undergo some major changes. Investors not only sold off shares of these tech companies but also the shares of power utility companies. These companies had only recently witnessed a surge in demand when the need for power to fuel data centres also increased exponentially.
However, the immediate selloff of market share is a knee-jerk overreaction. While DeepSeek has the potential to become the next big thing in the tech sector, there is still room for its growth and upgradations. One of the drawbacks of this AI model is that it is more suited for mobile phones and computers and not large data centres. This generative AI model would, therefore, only compete with ChatGPT, Gemini and MetaPlatforms.
Such competition with other open-source AI companies can result in reduced prices from these US-based companies, which are providing similar services as DeepSeek. Given the low operational costs, global adoption of this AI model is certain. However, there are concerns surrounding data privacy and security, particularly in the wake of heightened tensions between Washington and Beijing and the upcoming tariff wars. DeepSeek has confirmed that its user information is stored in servers in China.
With its sudden skyrocketing popularity, DeepSeek announced on Monday that it was temporarily limiting registrations after a cyber-attack. The website was initially hit by outages as it became the most sought-after application in the US. These outages were the longest in 90 days for the company, which quickly resolved the issues.
With the advent of DeepSeek, the previously held views about Silicon Valley’s supremacy in AI have capsized. The company’s low production costs have raised doubts that the chips being used are less powerful than the most advanced Nvidia products. Despite the Biden administration’s export controls to target the advanced chip and AI capabilities of China, this Hangzhou company has managed to upend the global tech industry within a matter of weeks.
Since 2021, the scope of the bans has widened significantly in a bid to prevent chips from being exported to China and, therefore, curb the Chinese companies’ attempts to train AI models. With DeepSeek’s news, US tech executives have questioned the effectiveness of such export restrictions. The Chinese officials have retaliated to such bans by accusing the US of unilateral bullying techniques.
While several other Chinese companies have developed their own AI models, DeepSeek is the first one to garner praise from the US tech industry as a product which is on par with or even surpasses Washington’s most recent and upgraded models. Some investors believe that cheaper AI services will push up demand for chips, and therefore it could only be a matter of time before DeepSeek becomes a household name in the tech industry.
There are new advancements in AI technology being made every day, and consumers and investors alike will throng to effective, cost-efficient and user-friendly services, which DeepSeek is promising today. More collaboration and innovation must be made through well-facilitated public and private partnerships so that the fruits of such labour can be collectively savoured.