The announcement wiped over US$1 trillion off the US tech sector, including chipmaker Nvidia’s $600 billion one-day fall – the largest in US history. Now that the dust has settled, let’s examine what DeepSeek’s advent could mean for the wider AI sector and investors.

What is DeepSeek?

DeepSeek is an advanced AI model that requires significantly less processing power and fewer advanced chips compared to models like ChatGPT. It reportedly cost only $6 million to develop, making it substantially cheaper than its established rivals, which have spent billions. Additionally, DeepSeek is open-sourced, meaning its code is freely available for anyone to use and modify.

How does it work?

DeepSeek's AI model training approach is unique because it doesn’t rely on the most advanced Nvidia graphic processing units (GPUs) or a brute-force method. Instead, it uses a different type of optimisation, focusing on reasoning and time as key variables. This method, known as test-time scaling, replaces the need for extensive computational resources (‘compute’) and pre-training, allowing DeepSeek to achieve extraordinary results with more efficient hardware and training techniques.

OpenAI's approach has driven capital expenditure from 'hyperscalers’ – large-scale data centres providing massive computing resources via cloud platforms. Pioneers like Microsoft, Google, and Amazon have sought to build a competitive edge in this space. The impact of the DeepSeek fallout on these companies is still uncertain. However, they will need to reevaluate the justifications for future spending and reconsider how best to compete.

Products, not potential

AI tools remain solutions in search of a problem, focusing on increasing the number of GPUs in advanced clusters to train the most powerful models. DeepSeek, however, requires far less initial processing . This could enable greater competition by allowing more participants to develop custom AI models. As a result, we may see an acceleration in the commoditisation of AI, with market participants competing through product differentiation and specialisation.

Demand shifted, not diminished

Despite the initial derating of many tech stocks across the AI semiconductor supply chain, demand should remain robust. We therefore remain relatively bullish on the sector’s prospects. The increased democratisation of AI can drive greater uptake among end users, sustaining the demand for increased processing power in the operational phase of the AI model lifecycle. This continued demand for processing power should also drive players in the semiconductor, data centre, and utility spaces.

What is fact?

It could be argued that the US’s export ban on advanced chips to China led to DeepSeek. As a result, developers were forced to optimise their models in more efficient ways due to limited access to cutting-edge hardware. The classic example of the law of unintended consequences in action.

What is fiction?

We question DeepSeek’s headline-grabbing $6 million figure. The company itself stated that the figure relates to the final training run. Either way, DeepSeek’s distillation and text-time computing training are significantly cheaper than OpenAI’s approach.

View from our analysts

Nvidia still holds significant advantages with its software coding language, CUDA, and the flexibility of its GPUs. However, China has shown that innovation can stem from optimisation rather than brute-force scaling.

This doesn’t mean that ‘compute’ isn't still useful, and we suspect hyperscale capex numbers will continue to reflect its importance. Additionally, falling costs could lead to a material increase in AI monetisation and the availability of AI applications at the edge.

What does it mean for emerging market (EM) and Asia investors?

We believe DeepSeek's ability to overcome sanctions is positive for the overall sector. More broadly, a lower competitive advantage and diminished US exceptionalism could be constructive for tech in Asia and EM. That said, we still expect the shift from hardware to be gradual.

At the time of writing, China tech giant Alibaba had released its AI model, the catchily named ‘Qwen 2.5-Max’. If its claims about the AI agent’s capabilities are true, this would further strengthen the narrative of innovating to overcome sanctions. Watch this space.

Implications elsewhere

DeepSeek should be a positive for ‘Rest of World’ equities compared to ‘US exceptionalism. It's not about EPS (earnings per share) trajectories or differences in terms of return on invested capital, but about democratising AI costs. This could benefit AI laggards and older economies by making the technology more affordable.

Interestingly, 70% of the S&P 500 was up on the same day that Nvidia and other tech stocks tumbled some 15%. This might suggest improving market breadth after years of high concentration (Nvidia alone accounted for 23% of S&P 500 returns last year).

Interestingly, 70% of the S&P 500 was up on the same day that Nvidia and other tech stocks tumbled some 15%. This might suggest improving market breadth after years of high concentration (Nvidia alone accounted for 23% of S&P 500 returns last year).

Final thoughts…

The launch of DeepSeek has upended the AI sector. Assumptions around training methods and raw computing power are no longer sacrosanct. This will have far-reaching implications for the sector, particularly for capital expenditure. Of course, this doesn't mean hardware's days are over – far from it. Instead, the world of AI, like the technology itself, has again proven to be innovative, surprising, and rapidly evolving. The democratisation of AI will undoubtedly open new frontiers. We will closely watch where these exciting developments take us and the investment opportunities they create.