Inside the Deepseek stock-market selloff

Last week, Deepseek, a Chinese AI startup, launched its R1 ‘reasoning’ model – an AI system that is free to use and reportedly cheap to train. This triggered a stock-market selloff, the effects of which were felt from Wall Street to Silicon Valley. 

The biggest loser was Nvidia, the US chipmaker, which shed nearly $600bn (£484bn) in market capitalisation. The tech-heavy Nasdaq index dropped more than 3% after the launch.

Two other chipmakers, Broadcom and Taiwan Semiconductor Manufacturing Company, also saw their share values slide: 17% and 13% respectively. 

The market has since bounced back, but the wipeout represents the biggest single-day loss in the history of the US markets. What’s more, the entrance of a Chinese competitor to the US-dominated AI market has significant geopolitical implications. 

Meet Deepseek – China’s open-source GenAI alternative 

According to reports, the cost of training Deepseek’s models was just $5.6m (£4.5m) – far less than the training costs of Silicon Valley’s large language models. Investors fear that firms such as OpenAI and Google, which rely on power-guzzling processes to train their AI systems, could be undermined by the Chinese upstart.

In December 2024, Deepseek released its V3 AI model – a language model that was trained on just 2,048 Nvidia chips, according to the company. The model unveiled last week, R1, is an advanced-reasoning model, competitive with OpenAI’s o1 model, which was made available to users less than two months ago.

The training data for Deepseek has not been made public, but the system’s code base and its ‘weights’ – the numerical parameters that show connections between nodes in a model – are open source. This means anyone with the hardware and technical knowhow can audit or host some of Deepseek. This is not possible with OpenAI.

Reactions in the tech industry have ranged from bullish to mystified. A spokesperson for Nvidia described Deepseek as an “excellent AI advancement”. Deepseek’s systems were trained on older Nvidia graphics-processing units (GPUs), so the American chip giant might welcome the company’s entrance into the AI sphere. 

However, the CEO of Anthropic, Dario Amodei, said in a statement that the market panic is “baffling” and the resulting frenzy is “greatly overstated”.

Amodei also voiced suspicions about Deepseek’s efficiency. Some analysts claim the startup has access to more than 50,000 top-of-the-line Nvidia Hopper GPUs. According to Amodei, such an arsenal would cost close to $1bn (£804m) – far more than the millions that were reported around the time of the selloff. If his estimate is correct, this would put Deepseek’s total spend quite close to that of US AI labs.

Deepseek and the US-China AI arms race

Perhaps more significant than the cost of training Deepseek’s models, suggested Amodei, are the geopolitical implications of new, competitive models emerging from China. 

The timing of Deepseek’s announcement is curious. It follows the announcement of Stargate, a $500bn (£402bn) joint infrastructure venture between the US government, OpenAI and Oracle. This project, too, is perceived as an escalation in the AI arms race between the US and China.

Deepseek’s timing “might be strategic, but their technology is real,” wrote Lennart Heim, a tech-policy analyst at Rand. “R1’s release during President Trump’s inauguration last week might be intended to rattle the public’s confidence in the US’ AI leadership.” 

According to Heim, this move mirrors the launch of a new mobile phone model by Huawei in 2023, which coincided with, and overshadowed, a goodwill visit to China by the former US commerce secretary, Gina Raimondo. But, Heim wrote, the PR timing shouldn’t obscure Deepseek’s technical progress – or the structural challenges it faces from export controls.

While there may be disagreement about the soundness of the market logic, technologists broadly agree that Deepseek is a significant, disruptive entrant to the AI space. 

“It’s clear that competition is heating up,” says Simon Baxter, principal analyst at TechMarketView. “Other than Deepseek, a number of Chinese suppliers are making significant AI progress – Alibaba Cloud also released more than 100 new open-source AI models. This will continue to put more pressure on Western AI suppliers to find new, innovative ways to reduce costs.”

What caused the selloff? 

The volatility displayed on Monday is a sharp reminder of the risks that come with extreme market concentration. The so-called magnificent seven – Apple, Amazon, Alphabet, Meta, Microsoft, Nvidia and Tesla – account for a whopping 31% of the S&P 500’s value. Any wobble among these firms will significantly impact on investors, especially those that have targeted the tech sector or failed to trim positions in these stocks as they rose.

“When investors lean too heavily on a handful of stocks, what once looked like unstoppable momentum can quickly become a glaring vulnerability,” says Sanjay Rijhsinghani, chief investment officer at LGT Wealth Management. 

This situation is not necessarily novel, he explains. In the late 1980s, for instance, Japanese banks dominated the Nikkei 225, fuelling incredible gains – until the bubble burst. “What followed was a painful correction and a lost decade for investors.” 

Post-Deepseek: who will be the winners? 

The Deepseek tremor emphasises the importance of stock selection and the need for active management. “Rather than reacting to short-term swings, investors should take this moment as a reminder to build resilience,” Rijhsinghani says. “Markets shift, cycles change – diversification remains one of the best ways to navigate whatever comes next.”

Market shocks are fairly common in the aftermath of major technological breakthroughs. “Entrepreneurs latch on to the breakthrough and find ways to make it cheaper and more useful in the race to commercialise,” says Will Hobbs, head of multi-asset wealth at Barclays UK Wealth Management. 

Global competition and information-sharing are very helpful in this process, he argues. Nvidia and the hyperscalers are the first-phase winners, but they may not be the ultimate winners in this new paradigm.

Perhaps, Hobbs continues, the concentration of financial clout in a few corporate titans is a necessary part of this revolution. “The AI infrastructure being laid down by these ‘mega caps’ may end up being a public good.”

While it’s tempting to see this market correction as AI’s dotcom-bubble moment, there are other dynamics at play. So says Shailesh Shukla, former vice president and general manager at Google Cloud, now CEO of Aryaka, a networking security company. “The emergence of Deepseek challenges the assumption that AI requires heavy capital investment. This could stimulate the market in the long run,” he says.

Greater competition and innovation in the sector could drive down costs further, leading to more applications and deployments and expanding the total market opportunity. Some analysts predict that, by making AI more accessible, Deepseek will increase demand for Nvidia chips.

“The impact of Deepseek’s innovative approach is yet to be fully understood, so this may not be the moment where the wheels come off the AI-investment bandwagon,” says Rob Morgan, chief investment analyst at Charles Stanley, an investment firm. “It could be beneficial in accelerating the adoption of AI technology through lower costs.”

The Deepseek saga shows how vulnerable financial markets are to innovation, which by its nature is unpredictable. For investors, it underscores the importance of building resilient portfolios. Markets are based on competition. At the end of the day, stocks that deserve to recover will recover.

Last week, Deepseek, a Chinese AI startup, launched its R1 ‘reasoning’ model – an AI system that is free to use and reportedly cheap to train. This triggered a stock-market selloff, the effects of which were felt from Wall Street to Silicon Valley.

The biggest loser was Nvidia, the US chipmaker, which shed nearly $600bn (£484bn) in market capitalisation. The tech-heavy Nasdaq index dropped more than 3% after the launch.

Two other chipmakers, Broadcom and Taiwan Semiconductor Manufacturing Company, also saw their share values slide: 17% and 13% respectively.