Companies poised for growth amid DeepSeek and other AI advancements

The launch of ChatGPT in November 2022 caused quite a sea change in the technology landscape due to its potential to disrupt large swaths of the global economy. As AI becomes more widely adopted, there will inevitably be both winners and losers. This recently became evident when Chinese AI company DeepSeek took the markets by surprise with the release of a Large Language Model (LLM) that performs on par with the market leading reasoning model, OpenAI’s o1.

DeepSeek’s LLM is particularly impressive considering the fastest Nvidia chips used to train these state-of-the-art models are barred from the Chinese market. Despite working on outdated hardware, DeepSeek not only built a model, but also managed to train it at a fraction of the cost incurred by leading AI labs. DeepSeek claims that the final training run cost the company only $6 million, while Meta’s latest Llama model cost $60 million. Moreover, DeepSeek is offering access to its models at approximately 1/25th of the price charged by OpenAI. These massive efficiencies have prompted investors to reevaluate the power, data center, and chip requirements for the development of large AI models.

As the market digested the news about DeepSeek, many companies that have ridden the AI wave in recent years saw their stock prices de-rate. Looking ahead, the cost of developing an AI model appears to be coming down –  which would be beneficial for consumers as well as AI infrastructure companies, as lower prices typically stimulate higher demand.

AI continues to rapidly evolve with new applications and practical use cases

At Vontobel Quality Growth, we believe in the power of connecting with people on the ground and gaining firsthand experience. That’s why our investigative analyst, Robert Berner, and portfolio manager, Ramiz Chelat, spent many hours walking the floor of the Consumer Electronics Show (CES) a few weeks ago.

A celebration of groundbreaking ideas and innovations, the CES has been kicking off the new year since 1978. Amid the 4,500 vendors showcasing their latest inventions across 2.5 million square feet, our Quality Growth team dove headfirst into the sea of electronics, on a mission to unearth deeper themes.

The growing dominance of AI was apparent in devices ranging from autonomous vehicle systems and smart glasses to robotics and baby-monitoring cribs all connecting to the cloud. This surge in innovation reflects a shift in the consumer mindset as AI becomes increasingly ingrained in everyday life. As Kevan Yalowitz, a managing director at IT consulting firm Accenture, pointed out, "Each week, 32% of internet users are now consulting AI-powered chatbots, like ChatGPT. And 5% of consumers are even willing to pay for more advanced chatbots."

In addition to increased development and utilization, there is a profound shift occurring in AI from training models to inference. In other words, AI is becoming involved in decision-making and executing actions independently, a concept known as agentic AI. Many view this as the next wave of AI, where, for example, AI agents assist software developers in writing code, an evolution that may render some functions, such as call-center jobs, obsolete within the next 10 years.

We expect that many of the devices showcased at CES will fail. AI smart glasses were a dime a dozen. However, AI had much less of a “bubble-like” atmosphere at this year’s show, mainly due to the rapidly increasing number of practical use cases for AI. AI has significantly cut the drug discovery cycle for pharmaceutical companies, reduced time to develop software code, and is aiding the design of more complex semiconductors.

Executives at chip software design house Synopsys and chipmakers Nvidia and Marvell claim that AI is changing long-standing assumptions around technological advances. For instance, Moore’s law states that chip complexity doubles every two years. This observation made in 1965 by Intel’s co-founder Gordon E. Moore held true for decades. “Now it is two times every three months,” said Ravi Subramanian, Synopsys’ chief product officer. It is also powering robotics, like Nevada-based Richtech Robotics’ wheeled robots that can pick up vehicle components in the parts departments of Mercedes dealerships and deliver them to the mechanics repairing cars. They work at Walmart, too.

CES also gave us insight into new areas. We will spend more time investigating the logistics and robotics space given our insight into Nvidia’s simulation platform, which replicates the physical environment in a warehouse, and allows customers to optimize the design and construction of an end-to-end goods handling system. Advancements in semiconductors and other IT to reduce data center power demand is another central theme.

Identifying reliable AI-related companies

Companies that have already been facilitating data collection and synthesis are poised to benefit tremendously from the integration of AI into their products. For example, Wolters Kluwer, a Dutch information services company, RELX, a British information and analytics company, and Experian, an Irish consumer credit reporting company, utilize large amounts of data to help their clients make better decisions. We believe these companies will be able to increase demand and expand margins through the use of AI. As AI becomes more affordable and better integrated into their operations, these companies will be able to offer their clients solutions that not only provide superior results but also save them time and money. In the software segment, SAP, which provides companies with CRM and ERP software, has already incorporated AI into its demand forecasting and supply chain management products.

The increasing computing power needed for AI applications is likely to lower the cyclicality of AI-related chips, such as GPUs, CPUs and ASICs for specific AI tasks. That will benefit chip-related holdings Taiwan Semiconductor Manufacturing Company and Japan’s Disco.

AI’s growth trajectory should translate into steady demand for new data centers through this decade, which in turn will drive energy demand and the need for secure sources of power. Data center power demand is expected to rise to 20% of US power needs by 2030. This should benefit our holding Schneider Electric, which produces power management and cooling systems for data centers, as well as grid infrastructure equipment to support the networks’ increasing energy needs.

The recent developments with DeepSeek, coupled with our observations at this year's CES, demonstrate the increasing emergence of AI products and use cases. Some of these applications may take longer to materialize, while others might appear to be overnight successes. There are risks that some investments could be pulled forward, which is why we maintain a solid position in AI infrastructure providers and steer clear of the more cyclical segments of the market. We balance this approach by keeping a close eye on the evolution of AI innovations and applications.