Date first published: 12/02/2025
Key sectors: all
Key risks: business risks; business disruptions; economic risks
Risk development
On 20 January, Chinese AI company DeepSeek shocked markets when it released DeepSeek-R1 and a paper detailing that training DeepSeek-V3 required less than US$6m worth of computing power from low-powered Nvidia H800 chips. The emergence of the low-cost AI model raised concerns among global investors, leading to a mass sell-off of tech stocks. This resulted in a US$593bln loss in market value for US tech firm Nvidia, marking the largest single-day loss for any Wall Street company, and a 3.1 per cent decline in the tech-heavy Nasdaq. DeepSeek’s release challenged the belief that AI’s future rested on massive energy consumption, surging electricity demand and reliance on nuclear power, which had driven a series of multi-million-dollar investments by US firms into complex and expensive AI models and massive AI-related infrastructure.
Why it matters
DeepSeek has entirely changed the narrative of AI competitiveness, positioning itself on par with the most advanced models alongside OpenAI and Meta, combining cost efficiency and high quality, with DeepSeek-R1 being reportedly 20 to 50 times cheaper to use than OpenAI’s o1 model, depending on the task. This innovation has paved the way for AI models which – despite being less powerful – can be offered to users at a lower cost, and rebalanced market power by tilting in favour of companies deploying AI models internally rather than solely benefiting AI developers and suppliers.
US President Donald Trump characterised the launch of DeepSeek as a “wake-up call” for the US tech industry, emphasizing his goal to make the US the “world capital of artificial intelligence”. Additionally, DeepSeek reportedly also possesses less powerful H20 chips, which are still legally shipped to China. While the Trump administration is considering tightening restrictions on these chips, they remain in the very early stages.
Background
The launch of OpenAI’s ChatGPT in November 2022 triggered a race, especially among Chinese tech firms, to successfully develop their own AI chatbots. However, when Chinese tech giant Baidu introduced its equivalent of ChatGPT in November 2023, it was met with widespread disappointment, highlighting the gap between Chinese and US AI development capabilities.
In the US, the excitement around AI fuelled a huge inflow of capital into equities over the past 18 months, inflating valuations and lifting stock markets to new highs. However, the release of DeepSeek quickly changed the landscape, as it only took a few days to overtake US rival ChatGPT in downloads on the Apple Store, triggering a global sell-off of shares. Both Microsoft and Amazon have since begun offering DeepSeek models, although concerns around data privacy persist in various countries.
While not much is known about the Hangzhou-based start-up behind DeepSeek, and its controlling shareholder Liang Wengfeng – also the co-founder of hedge fund Highflyer – Beijing welcomes its success elevating Liang to the status of a pop culture celebrity.
Risk outlook
Following a rapid sell-off fuelled by concerns that this disruption could lead to lower demand for chips, less need for power usage and less reliance on large-scale data centres, tech stocks quickly rebounded as investors weighed the long-term implications of DeepSeek’s release for AI development. While DeepSeek’s emergence illustrated that less can be more, its development using an open-source model suggests that competitors are able to build on it. By increasing efficiency and decreasing costs, the use of AI and therefore demand for AI chips such as Nvidia’s is set to increase – a phenomenon known as the “Jevons paradox”.
Meanwhile, the US Commerce Department continues to investigate the acquisition of the 50,000 banned Nvidia H100 chips, while other countries – including Australia, South Korea and Taiwan – have banned the use of DeepSeek on government devices due to data privacy concerns. On 20 January when DeepSeek-R1 was publicly released, Liang attended a closed-door symposium for businessmen and experts hosted by Chinese Prime Minister Li Qiang, signalling that DeepSeek’s success could be aligned with Beijing’s policy objectives of countering Washington’s export controls and achieving self-sufficiency in strategic sectors such as AI.