Technology sector volatility: Investors spooked by China’s AI startup

Following a period of extraordinary growth, the technology sector recently felt the brunt of market volatility on the back of a new, potentially game-changing innovation.

Technology shares fell sharply at the end of January after Chinese start-up, DeepSeek, launched a rival to the hugely popular chatbot ChatGPT, which spurred on investors to question the broad assumption that the U.S. dominates the artificial intelligence (AI) sector.¹

DeepSeek’s large language model (LLM) – technology which learns from vast data sets to carry out a huge range of tasks – can reportedly match ChatGPT’s performance but allegedly at a much cheaper cost and with the need for fewer expensive microchips.²  The emergence of potentially cheaper and more efficient AI models has questioned the need for huge capital spending on AI infrastructure – namely computing power and electricity.

Mixed volatility

If capital expenditure does not need to be as high as previously thought, revenues for AI infrastructure providers would be impacted. Shares in Nvidia, which has been the leading maker of powerful semiconductors required to run AI applications, alongside other chip makers and firms geared towards generative AI, fell sharply on the news of DeepSeek’s arrival.³  Power and infrastructure-related companies were also impacted, as they have benefited from the expected increase in electricity demand to power AI ambitions.

In Europe, the impact was less severe as semiconductors represent a smaller share of indices in the region – less than 4% of MSCI Europe versus more than 10% of the S&P 500.⁴  Semiconductor equipment provider, ASML, announced strong results on 29 January, suggesting that demand for high power chips will continue to be a major theme for equity investors.⁵  The main impact in Europe was among power/electricity infrastructure related names such as Schneider Electric and Prysmian

who notably rallied earlier in January due to the expected increase in electricity demand to power global AI ambitions.⁶

Following the initial market correction, technology shares have regained some ground. There have been suggestions that DeepSeek’s model development utilised existing models, such as Meta’s Llama 3. Therefore, the real development cost could be significantly higher than some press reports suggested.

Tech outlook

The technology bull market of recent years, driven primarily by the so-called “Magnificent Seven” stocks — Alphabet, Amazon, Apple, Meta Platforms, Microsoft, Nvidia and Tesla – was never going to rise in a straight line.

At this point, we need to ask what is DeepSeek challenging? It is training LLMs more efficiently than U.S. incumbents and therefore needs less computing power i.e., fewer chips, less power, and less capital expenditure? DeepSeek can also run the inference stage of the process more cheaply, but this is part and parcel of the rapid innovation taking place in the industry, without which AI would not successfully scale. For example, Google’s Gemini 2.0 Flash Thinking model, which is currently in the experimental phase, could potentially deliver better performance at significantly lower cost than DeepSeek.

Ahead of Nvidia’s results to be announced on 26 February, it is too soon to claim that we have seen peak capital spending on the AI infrastructure ecosystem. Indeed, technology companies are generally saying that they are still building capacity to run ever more complex applications.

Notably, both Microsoft and Meta posted robust calendar fourth quarter (Q4) results, with each reporting a rise in earnings per share and revenues – beating Wall Street expectations – and pledged to invest billions of dollars into AI infrastructure.⁷

On their respective results calls, they directly referenced DeepSeek but highlighted that such efficiency improvements in AI modeling and implementations are somewhat the normal course of business. Microsoft chairman and chief executive, Satya Nadella, said the business has “typically seen more than 2X price-performance gain for every hardware generation, and more than 10X for every model generation due to software optimizations.”⁸

A new frontier

DeepSeek could be a classic example of how the cycle works in new technologies – continual improvements in efficiency and reducing the cost base. This could fundamentally mean greater take up of AI applications among non-technology enterprises and organizations if the developments being made by DeepSeek and others continue to improve the AI investment returns. However, for players like Nvidia, any structural reduction in the computing needs from the development of open-source

technologies like DeepSeek’s application could potentially hit future revenues. So far, the capital investment outlooks from the biggest buyers of Nvidia graphics processing units remain firm - with Meta, for example, announcing its capital expenditure guide for 2025 at $60bn-$65bn, well ahead of consensus expectations.⁹

Overall, in our view, it remains hard to bet against the U.S. The Q4 earnings season has got off to a strong start with results beating expectations overall and growth looking to be strong. Sector performance has been more balanced since the beginning of 2025 with industrials and energy the top performers.¹⁰  It is not surprising to see rapid advancements in technology in a strong growth area like AI and there is no reason why companies in countries other than the US cannot develop efficient LLM models and AI applications.

We strongly believe that AI remains a key theme and will provide investment opportunities in software, cloud computing, and applications, as well as the infrastructure of semiconductors, data centers, and power generation in the short term while likely providing broader opportunities over the longer term.

What’s next?

From a macroeconomic perspective, recent stock volatility could have little discernible impact. However, a broader correction might be healthy given the technology sector was central to 2024’s performance where the Nasdaq and S&P 500 respectively achieved total returns of 30% and 25%.¹¹

Ultimately, AI innovation and leadership may come from a broader group of players than initially expected, and it remains a fast and structural growth theme which could support the global economy over the next decade. And while there will always be disruptors, there is still much to be positive about in terms of the U.S., in our view. The U.S. consumer continues to benefit from a healthy jobs market, growing real wages, and a substantial wealth effect.

On the corporate side, the new U.S. administration is very pro-business – President Donald Trump’s support for AI, and technology more generally, as well as the oil and gas sectors (and for deregulation in finance) are the nexus of hope for continued stock market performance. However, any prolonged set-back in the stock prices of big technology companies could have a more discernible impact and potentially contribute to softer U.S. economic activity, in part through its impact on consumer spending.