The emergence of the Chinese artificial intelligence platform DeepSeek-R1 has raised questions on how it will change the AI value chain, from providers to consumers to real-world business applications of the technology.
According to a report by the BBC, researchers with Hangzhou DeepSeek Artificial Intelligence Basic Technology Research Co. Ltd., the company behind DeepSeek, claimed it cost just US$6 million to train the generative AI model, which is a fraction of the price of its rival OpenAI. The report noted the price reduction was largely due to DeepSeek using more cost-efficient but less sophisticated chips.
Read: Balancing the risks and opportunities of AI
DeepSeek uses significantly less energy in the process, reducing the cost exponentially, says Ruslan Goyenko, associate professor of finance at McGill University’s Desautels Faculty of Management and president and chief executive officer of FINAIX Corp. While much of the data still has to be verified, researchers agree the DeepSeek training model is plausible, he says, noting since it will be open-sourced, other players can now borrow from its groundbreaking engineering process.
This type of cost-cutting never decreases demand when it comes with an improvement, he adds. “What it shows [is that] smaller companies with more limited resources can also run with this [engineering method] and . . . compete with the big giants like OpenAI. Smaller companies, which before could not enter the AI space because the costs were too high, are now considering it and are now budgeting to invest into it.”
The short-term market reaction to DeepSeek’s release on Jan. 27 was swift and steep; in a matter of half a day, stocks in this sector lost roughly 25 per cent, he says. “Right now, the markets are very conservative [in a] wait-and-see position.”
Read: 2024 Risk Management Conference: How can AI play a bigger role in investment decisions?
One area poised to withstand the fallout is the production of semiconductor chips. To date, Taiwan has had the benefit of dominating in this arena, particularly Taiwan Semiconductor Manufacturing Co. Ltd. — one of the world’s main producers of semiconductor chips. While other countries, including the U.S. and Germany, have moved to begin domestic production of semiconductor chips, it will take some time due to the high capital costs.
By contrast, DeepSeek has had a residual effect on the future of the data centres asset sector. Once deemed a promising space due to a more bullish outlook on AI, these stocks took a hit and haven’t yet recovered fully, says Goyenko, noting many organizations are rethinking whether high-capacity data storage needs will still be required in light of DeepSeek’s lower-cost engineering model.
However, he believes over the long term, data centres will recover and even prove resilient. “The thing is, given that demand is going to grow [and] that more businesses are going to be adapting [generative AI models], that energy will be demanded and maybe even more so than they can offer right now, which is why there is also [interest in] investment in nuclear energy.”
Goyenko says some small businesses are reaching out to him to gauge whether they can use the DeepSeek model to predict consumer demand for their products. “So, there is an interest [in] investing in it, and . . . they will [need] to use data centres. . . . So I think [data centres are experiencing] a short-term negative sentiment . . . that will lead to higher [investments] over the long term.”
Read: Are AI’s energy demands impacting the appeal of data centres to institutional investors?