Fast reading
- Chinese start-up DeepSeek launched a new open-source AI model that it claimed represented significant cost and efficiency breakthroughs.
- The news caused tech stocks to sink as investors worried about the implications for AI leaders such as US chipmaker Nvidia.
- However, lower AI costs have the potential to provide a boost to chip manufacturers around the world over the long term.
US tech stocks tumbled this week following the launch of a low-cost Chinese artificial intelligence (AI) application, which threatened Big Tech’s market dominance.
The AI investment boom of the past two years has, until now, been largely driven by a handful of US tech stocks – and led by the so-called ‘Magnificent Seven’.
The arrival of Chinese start-up DeepSeek’s new R1 model on Monday quickly threatened the established order and had an almost immediate impact on US markets. Upon launch, R1 claimed the number one spot in the Apple app store, outperforming Microsoft-backed OpenAI’s ChatGPT and Alphabet’s Gemini.
“With corporate optimism running at record levels, the news about the launch of the DeepSeek app sent prices for some of the biggest tech stocks tumbling,” says Louise Dudley, Portfolio Manager at Federated Hermes.
The launch of the new open-source model triggered a 3.1% drop in the tech-heavy Nasdaq index, while the S&P500 was down 1.5% in trading on Monday. The Magnificent Seven – Apple, Microsoft, Alphabet, Meta, Amazon, Tesla and Nvidia – fell 2.7% in response to growing awareness of the R1 model and the potential implications this has for the AI trade industry.1
DeepSeek’s announcement has been dubbed a 'Sputnik moment' and a 'wake-up call' for US AI superpowers
Chipmaker Nvidia was hit particularly hard and had plunged 17% by the time US markets closed on Monday, representing a market cap loss of US$600bn in a single day.2
Sputnik moment
DeepSeek claims to have trained the model on a budget of just US$5.6m, using lower-end Nvidia H800 GPUs, because more powerful H100 GPUs, such as those used to train Meta’s Llama-3 model, are unavailable in China due to US export controls.
“This would suggest significant cost and efficiency breakthroughs that have the potential to disrupt the AI value chain through greater competition. The R1 has similar performance levels to AI research organisation OpenAI’s o1 reasoning model with less than 5% of the inference cost3,” explains Tej Sthankiya Senior Equity Analyst at Federated Hermes.
DeepSeek’s announcement has been dubbed a ‘Sputnik moment’ and a ‘wake-up call’ for US AI superpowers, by venture capitalist Marc Andreessen and US President Donald Trump, respectively.
“If AI deployment can be done at a lower price, it reduces barriers to entry and adoption, and presents a threat to mega-cap dominance. The US tech stock tumble also highlights the risk of market reliance on mega-caps,” says Sthankiya.
However, the next round of earnings reports from the Magnificent Seven appeared to offer investors some reassurance regarding the stability of the tech sector.
Meta and Tesla shares rose following the release of their earnings reports in the days after the R1 launch. But Microsoft shares fell after sales in its cloud computing division were weaker than expected.
Nvidia is set to be the last of the seven to publish results, scheduled for late February.
Sthankiya adds that long-term demand for Nvidia’s premium chips is unlikely to be affected by low-cost AI: “Lower AI inference costs mean more consumer and enterprise applications across end markets become cost-effective, and thus long-term demand likely remains unaffected.
“There’s even the case to be made that there will be more demand for chips because now it is economical to have many more models trained and used in different ways. A recent analogy for this is the cloud computing platform shift.”
BD015289