Technology shares have fallen around the world amid fears that the emergence of a low-cost Chinese artificial intelligence model could threaten the dominance of AI leaders including Nvidia.
The stock market value of Nvidia fell almost $600 billion.
Chinese startup DeepSeek last week launched a free AI assistant that it says uses less data at a fraction of the cost of other services.
By Monday, DeepSeek’s assistant had overtaken US rival ChatGPT in downloads from Apple’s app store.
On Monday, the Nasdaq fell 3.1%, Nvidia shares fell almost 17%, chipmaker Broadcom tumbled 17.4% and Google parent Alphabet fell 4.2%.
The Philadelphia semiconductor index fell 9.2%. Marvell Technology declined 19.2%.
ASML fell 7% and Japan’s SoftBank Group slumped 8.3%.
Nvidia shares have fallen 11.8% in 2025 after soaring around 171% in 2024 and 239% in 2023.
Vertiv Holdings, which builds data center infrastructure, declined 29.9%.
Investors also sold shares of power utilities, which had risen sharply on expectations for a massive demand surge from data centers required for AI.
Constellation Energy shares fell 21%, Vistra shares fell 28.3% and NRG Energy fell 13%.
REACTION:
Jamie Mills O’Brien, Investment Director, Abrdn: “Recent jitters in the tech market reinforce our view that investors should not be confining themselves only to the so-called MAG7 when looking for growth.
“The US stock market has been highly concentrated in a small number of tech companies, including Apple, Amazon and Nvidia, for several years. DeepSeek – and we are working with our regional teams to assess reliability of claims that the cost to build the material is materially lower – represents another version of this risk. Chipmaker Nvidia represented more than 20% of the returns of the S&P500 last year, and, with such a run, the stock, and broader AI ecosystem, entered this year with more vulnerability around specific pieces of news flow.
“The gap in earnings growth between the largest stocks in the US market – the so-called ‘MAG7’ – and the rest of the market is the narrowest it has been in a long time. So although these large stocks are expected to outgrow the other 493 in 2025, the difference is much smaller than it has been in a long time. Specifically, the gap in earnings growth between these large stocks was c.28% in 2023, c.35% in 2024 – and this is expected to shrink to c.4-5% in 2025.
“We think this narrowing growth premium for the larger cap technology stocks likely also brings with it a far less narrow US equity market – by which I mean a US market that is much less concentrated in terms of performance.
“This is likely material for fund performance too – given these seven stocks now represent c.32% of the S&P500, the relative over/underweight to this small number of stocks has also been a key driving factor in relative performance over the last few years.
“We are also starting to see signs of a market behaving slightly differently, with earnings growth – rather than multiple expansion – becoming a more important arbiter of equity returns. Historically when you hit the peak of a hiking cycle, market focus turns to earnings growth.
“And that is what we are seeing again – earnings, not multiples, are driving relative performance in 2024 than they have done over the last couple of years. This is even more relevant in a late-ish cycle backdrop where there is potentially some growth vulnerability on the horizon (and the incoming administration’s tariff policy potentially exacerbates that).
“This is actually a very positive environment for those global equity managers – like us – which are focused on identifying that select and small group of high quality companies globally which are able to deliver earnings and cash flow growth over and above market on a consistent basis. And with a slightly longer term hat on, the next 10 years are going to be lower growth than the last 10 – driven by lower fiscal headroom globally, demographics and deglobalization – and that again means that there will be a much higher premium on those rare companies which can grow independently of the economic cycle.
“Part of our thinking here is because we have seen this before. Over the very long run, you do get these patterns that are remarkably similar of US outperformance followed by outperformance of markets outside the US. The period 2000-2010 was very much a lost decade for US stocks – and what followed has been a very strong run for US stocks. Similarly, the earnings and cash flows of US vs non-US stocks go through cycles. Over the last 10 years, we have seen enormous valuation compression between US and non-US stocks.
“US stocks used to be at a 14-15% premium. Today that number is greater than 50%. That isn’t to say we are negative on US equities – because we are not – rather we see potential for a rotation towards a less narrow market structure, regionally and sectorally, in 2025 and beyond.
“Emerging markets and China are set to deliver earnings growth over and above the US market. We think that might be significant, in particular given the valuation discrepancy we are seeing.
“With a longer term hat on, the next 10 years are going to be likely lower growth than the last 10 – driven by lower fiscal headroom globally, demographics and deglobalization – and that again means that there will be a much higher premium on stocks that can grow independently of the economic cycle.”
Kathleen Brooks, research director at XTB: “The US dominates global stock markets due to the mega cap tech rally in recent year. However, the assumption that US tech will lead the future in the AI revolution will be tested this week.
“The reason is Chinese AI company Deepseek, which released its new R1 product, a large language model, last week. It also released details on how it created its model, and this is causing an existential crisis in the US stock market … Deepseek is a clear reminder that US AI dominance cannot be taken for granted, as China plays catch up and Chinese tech stocks have surged on Monday.
“Deepseek is causing excitement in financial markets for a few reasons. Firstly, it is an open-source model, compared to OpenAI, which is a closed source model, which means anyone can help develop it, and helps to keep costs low. Deepseek’s R1 model is also free to use with a limit of 50 messages a day, suggesting that it could be a real competitor to OpenAI and other US-based AI products.
“Deepseek also shows that China can produce AI tech that is cheap to use for the consumer, which is leading to questions about the high level of capex that US tech firms are spending on their AI capabilities. For example, Meta has pledged to spend $60-$65bn on boosting its AI capabilities this year, but does it need to spend this money if Deepseek can produce a large language model for a fraction of this price? Meta could see its shares get targeted on the back of this development, ahead of its earnings report later this week.
“The other interesting development is that Deepseek seem to be able to build advanced AI tech even though there are sanctions in place that are blocking the sale of high-tech US chip exports to China. This calls into question whether Nvidia’s latest high tech and expensive chips are really necessary …
“Deepseek may have resource challenges, such as the ban on exports of high-tech chip technology from the US, but the nature of Deepseek means that it can pool collective expertise and knowledge, which boosts innovation and has allowed it to develop a cutting-edge technology that can compete with its US counterparts.
“This is the first main threat to US-based AI. Having a non-American rival could trigger fears that AI could go the way of the EV market: China moves from importer to exporter and floods the market with its product. However, it is worth noting that Deepseek is still reliant on Nvidia chips, albeit those that are less advanced than the Blackwell chip …”
Solita Marcelli, chief investment officer Americas, UBS Global Wealth Management: “Low-cost models are positive for AI adoption: It is not necessarily a zero-sum game — the overall market can grow, with potentially lower costs accelerating AI adoption.
“Continued capex is required to produce innovative AI models: Even after the release of DeepSeek’s model, Meta said it plans to spend as much as $65bn this year on AI infrastructure. AI spending plans also include models linked to audio and video, which are not currently in scope for DeepSeek’s application.
“Where value in the AI value chain accrues will shift over time: We have long thought that value over time would eventually shift from those developing the infrastructure (enabling layer) toward those using it (application layer and beneficiaries). Still, we caution against jumping to conclusions at this stage. Lower costs may also even increase demand for the enabling layer as lower compute intensity is more than offset by wider adoption.
“Bottom line: AI is here to stay, and if anything, DeepSeek reinforces that. But the latest developments are also reminder to avoid an investment approach that is too concentrated or overly passive. We prefer an active and diversified approach to gain exposure to AI. Further clarity on DeepSeek’s impact, big tech’s capex plans, and AI monetization should emerge in the fourth-quarter earnings season in the coming weeks.”