Why did Chinese AI firm DeepSeek jolt the markets?

Since ChatGPT burst onto the scene in late 2022, Large Language Models (LLMs) have transformed AI technology, enabling machines to understand and generate human-like text. However, LLMs are costly to build and operate due to high computational and energy demands. This cost has been a major obstacle for companies deploying AI.

DeepSeek has changed the game with significant advances in LLM AI algorithms. Their R1 LLM matches a similar model by OpenAI (o1) in terms of performance but at a cost that is more than 90% lower. Cheaper LLMs make advanced AI accessible to more businesses and individuals, driving innovation, levelling the playing field, and enabling wider adoption of transformative technology.

Despite the export bans from the US of more advanced chips, DeepSeek’s team managed to make the most out of their limited hardware. Although some questions remain regarding underlying costs, such as prior model training expenses and R&D, as well as the extent of reliance on previous open-source models, the DeepSeek R1 stands as a breakthrough in both innovation and cost efficiency.

​​​This is not to say that local factors have not been pertinent. The step-up in the UK’s spending commitments following the October Budget means that we’re anticipating an influx of new government borrowing in the near term. This increased supply in UK gilts will be a headwind for bond prices.

Furthermore, UK core inflation remains elevated, curtailing the extent to which the Bank of England can cut rates. This has also been a significant factor in UK gilt performance.

Impact on the market and the outlook for AI?

AI infrastructure shares have fallen the most over the last few days. Nvidia lost $650 billion of its market capitalisation on Monday. The market’s main concern is that companies may reduce spending on chips and data centres.

While it is an easy conclusion to make, it is not a fait accompli. The lower cost may enable vastly more businesses and individuals to utilise AI and thus actually increase the computation and energy requirements of AI over the long term.

Furthermore, it remains to be seen whether DeepSeek’s technological breakthroughs, combined with increased computational power, will unlock even more advanced AI models, which would keep up strong demand for AI infrastructure. It is perhaps because of this more nuanced view we saw a sharp rebound in Nvidia’s shares on Tuesday.

Performance among other big technology names has been more varied. Lower AI costs could reduce the significant capital expenditure and operating costs seen previously. ​​​​Cloud providers, such as, Microsoft (Azure) and Amazon (AWS) are likely to benefit from higher cloud usage on broader AI adoption.

The ease and speed of software development will likely increase overall demand, rather than decrease it. According to the Jevons paradox, when something becomes cheaper and easier to access, consumption tends to increase dramatically instead of decreasing or levelling off.

There are still risks though. Moats for the big players in this sector may be smaller than initially perceived, as DeepSeek has demonstrated that small companies are able to make significant technological advancements with far fewer resources. Generative AI technology is still in its infancy, and it remains to be seen whether the winners of the past decade will be the winners of the next.

Navigating risks and opportunities

While this breakthrough marks an exciting step forward, it is still early days. As AI evolves there will be further risks and opportunities, and the market sometimes takes time to digest and understand the implications. As always, our team looks at the nuance along with the data to ensure we take a considered approach when we review our investment strategy.