deepseek ai vs chatgpt - what it means for markets

What is Deepseek?
The new Chinese AI tool is the most downloaded app in the US and claims to be able to operate quicker and more cheaply than Chat GPT.  By requiring fewer chips and smaller server farms, DeepSeek can deliver its AI generated results, which incorporates self-corrective learning, at lower cost.  The arrival of DeepSeek in the AI field creates a challenge to Chat GPT and also means more could potentially be done with fewer advanced chips. The DeepSeek app is free, and its code is open source, which means companies in the US can easily create similar copycat versions of it. Its AI model is called R1 which has some 670 billion parameters, making it the largest open-source LLM (large language model) yet. The R1 is unlike traditional models that immediately generate responses. The R1 are trained to think extensively before answering. Its approach is similar to how a human carefully considers a complex problem before answering it. The method is known as test-time compute and requires the model to spend up to minutes working through its chain of thought. The company estimates it has cost $6 million to train the model compared to other AI models which cost over a $100 million each. The app challenges the notion that the only way of building better AI models is by spending large amounts of money buying high powered processing chips.

What happened to NVidia stock?
Lower potential demand for chips has led to a rapid sell-off in NVidia shares, which declined -17% in 1 day on 27th January 2025.  Because of NVidia - and the broader technology sector's high exposure within the S&P 500 this led to a US equity market sell off, dragging down market-cap weighted world equity indices too. Other stocks which suffered similar down moves were Oracle (-13.79%), Cisco (-5.06%), Broadcom (-17.40%). These companies along with Nvidia provide important components in building data centers and digital infrastructure to support training AI models. NVidia released a statement saying "DeepSeek’s work illustrates how new models can be created using that technique, leveraging widely-available models and compute that is fully export control compliant." While President Trump said the breakthrough in technology is a positive for America and that it is a wake-up call for US tech firms to not rest on their laurels.

Elevated concentration risk in the US equity market is well-documented and one of the reasons we recommend advisers we work with to balance traditional S&P500 market-cap weighted exposure (which was down -1.46%) with S&P500 Equal-Weighted exposure (which was up +0.03%), and also for active selections within Sectors and Styles.

How much NVidia is in a portfolio?
For a balanced, well diversified portfolio built with funds, direct exposure to NVidia should not be more than 2% or so.  This reflects the importance of diversifying away from stock-specific "idiosyncratic" risk.

​Creative disruption requires diversification
As with any paradigm shift, it's hard to identify clear-cut winners in advance.  We welcome the AI revolution as a broader technological innovation theme to consider within portfolios.  But this episode emphasises the value of having a diversified approach.