Vulnerabilities in the US technology sector have been revealed by the arrival of a cheaper Chinese AI model.
- Nvidia share price dropped almost 20% as Chinese group DeepSeek announced its new AI model
- There are still questions around the accuracy of the cost figures and how the model was trained
- It showed the vulnerability of parts of the technology sector to disruption
It has been a rollercoaster ride for investors in Nvidia this week, as Chinese group DeepSeek announced a new AI model that has been built at around a tenth of the cost of its US equivalents. The Nvidia share price dropped almost 20% on the news, as investors concluded that more efficient AI models would lead to reduced demand for chips. ASML was also hit.
There are also potential beneficiaries. For companies that use AI, it means costs will come down. That may encourage innovation, as technology becomes accessible to a broader range of customers. Software companies could benefit, as could the so-called hyperscalers (Microsoft, Meta, Google), which may be able to reduce their AI spending.
There are still plenty of unanswered questions. Are the cost figures accurate? ChatGPT has already retaliated by saying that it believes DeepSeek has used its proprietary models. Karen Kharmandarian, portfolio manager at Thematics Asset Management, says: “Some market participants suggest that the company may have a cluster of 50,000 Nvidia H100 GPUs, which would push the investment into billions of dollars. Without even considering the salaries of engineers working on the project, the $6 million figure seems highly questionable.”
Equally, it is not clear that it will lead to a reduced demand for chips. It is possible that it accelerates AI adoption and increases demand. Schroders Paddy Flood, a global technology sector specialist, points to a phenomenon called Jevon’s Paradox, which suggests that improvements in resource efficiency often drive greater consumption of that resource.
However, it does reveal a chink in the armour for Nvidia. It shows how developments can emerge from left-field and disrupt the ecosystem. If it hadn’t been DeepSeek, developments in quantum computing could have had a similarly disruptive effect – and still might. It should be a reminder that the current concentration in US markets is a significant risk. After two years of exceptionally strong returns from the US technology sector, investors may have a lot of ‘unintentional’ exposure and may not recognise the potential piffalls.
One encouraging factor in the recent rout is that the remainder of the market has been largely unaffected. There had been concerns that any wobble in the technology giants would spark a sell-off in the broader market, but that hasn’t been the case. This suggests a welcome resilience in global stock markets, but should also be an encouragement to explore other parts of the market.
