PETALING JAYA: The domestic technology sector could be in for more uncertainties as Chinese startup DeepSeek launches a free, open-source artificial intelligence (AI) model that experts say rivals OpenAI’s ChatGPT.
This comes amid heightened tensions in the AI trade, which saw a sell-off in the technology sector earlier this month after the Biden administration announced new restrictions on the export of advanced semiconductors and AI technology, citing national security concerns.
Under the new rules, Malaysia was placed in a category allowed to procure only a fixed and limited amount of such advanced technology, potentially constraining the development of its AI capabilities.
Subsequently, the local construction sector was downgraded by at least two research houses, on the basis that many of these contractors had factored in work related to the construction of data centres in Malaysia.
Analysts are largely of the view that US president Donald Trump, who aims to establish the United States as the global leader in AI, is unlikely to alter the directive set by former president Joe Biden regarding the limitations on exporting AI chips from the United States.
However, Tradeview Capital Sdn Bhd portfolio manager Ng Tzyy Loon said DeepSeek’s AI chatbot may throw the effectiveness of the proposed AI export control order into question.
“The US’s strategy to limit the development of AI in other countries by controlling their access to top-tier computing power and technology may not achieve its intended goals, as proven by the creation of DeepSeek,” he told StarBiz.
Last December, DeepSeek launched its DeepSeek-V3 model, which was reportedly developed at a much lower cost of US$5.6mil. In contrast, the training of OpenAI’s ChatGPT-4 model had reportedly required an investment of around US$100mil.
On Jan 20, the startup released another AI model, the DeepSeek-R1, which is said to rival OpenAI’s o1 (designed to complement ChatGPT) reasoning capabilities, sparking concerns over US tech dominance and prompting a reevaluation of technology companies’ lofty valuations.
The Bursa Technology Index has slipped by 0.88% since Monday. Yesterday, local technology or data centre-related stocks like YTL Power International Bhd
, Inari Amertron Bhd, Nationgate Holdings Bhd and PIE Industrial Bhd fell by as much as 3%, 2%, 5% and 2%, respectively.
Major Wall Street indexes also tanked as the market digested the release of DeepSeek-R1.
The S&P 500 tumbled 1.5% while the tech-heavy Nasdaq composite sank 3.1%.
The blue-chip Dow Jones Industrial Average, which is less dependent on tech stocks, gained more than 0.6% with investors flocking to more defensive sectors.
Major tech counters like Nvidia Corp and ASML Holding NV slid as much as 17% and 6%, respectively.
Ng noted one notable aspect of DeepSeek’s AI models is their use of Nvidia’s H800 chips for training, which are not top-of-the-line chips like Nvidia’s H-100 of which the Biden administration’s latest export controls had planned to target.
“This shows that restricting access to top-tier chips may not prevent advancements in AI development, as companies can innovate around these limitations,” he said.
While this may be the case, it is important to note that the H-800 chip itself has been included in the US export restriction list since 2023.
Tradeview’s Ng also pointed out the cost and complexity of monitoring and tracking AI chip usage make enforcement highly challenging for the United States.
“While the US government can track where the AI chips are distributed, enforcing such restrictions is challenging, given the number of countries, such as Singapore, that are eager to advance their AI capabilities.
“Countries may also find ways to smuggle in AI chips like what China does, making it difficult to monitor effectively,” he said.
Ng is of the view that Trump may employ a more pragmatic approach in going about Biden’s proposed AI export control order.
“I think he may repeal the order or at the very least, adjust the rules to make the restrictions less stringent,” he said.
In a report yesterday, Kenanga Research said all eyes are now on Big Tech’s response to the AI capital expenditure ahead, with concerns surrounding the risks of a smaller addressable market for high-end chips.
“On the heels of big spending announcements of a whopping US$500bil under the joint-venture entity Stargate, the pledge to spend multiple billions by Big Tech will likely come under more scrutiny, as we expect them to carefully evaluate strategies given this AI development.
“Demand for state-of-the-art chips will still be intact in our view for firms that are pushing the envelope in developing frontier large language models, or put simply, the most advanced and cutting-edge models to understand and generate text,” the research house said.
As for the data centre play in Malaysia, Ng said it remains intact in the near term looking at the committed data centres here. However, there may be delays or uncertainties around new data centre projects.
“This is because the graphics processing unit (GPUs) already committed are well below the levels planned by major players like Nvidia and Amazon globally.
“For now, the impact should be manageable in the near to medium term, but beyond three years, further expansion could become challenging if the restrictive AI executive order really comes through,” he said.
BMI telecoms and technology industry analyst Niccolo Lombatti said it is important to note that not all Malaysian data centres rely on US-supplied chips.
“The decision of what chips to use is largely a function of its intended use case and therefore its design.
“On the one hand, some Malaysian data centres can utilise a lower number of US-supplied GPUs or chip alternatives from non-US vendors because they are looking to address demand from non-AI related use cases, or less intensive AI use cases, thus insulating them from the AI executive order’s effects,” he explained.
Nonetheless, Lombatti said the main risks arise for data centres targeting high-density AI applications, particularly those in Johor aiming to attract Singapore-based customers seeking rack densities up to or even in excess of 120 kilowatt
“Therefore, Malaysian data centres designed around high-density racks using the latest US-manufactured GPUs face greater risks over the next few years. Owners may need to slow development or scale back to lower-density designs, leading to significant capital expenditure inefficiencies,” he said.
Ng remained optimistic the country will be able to continue to attract data centre investments, underpinned by Malaysia’s cost competitiveness in terms of land, labour and electricity.
“Additionally, Malaysia’s proximity to Singapore is a key factor. The geographical location is crucial for data transfer and connectivity, and many global players already have data centres in Singapore,” he said.
On this note, Kenanga Research said contractors are more insulated in the AI race to roll out data centres given the emergence of DeepSeek could accentuate Malaysia’s position in being able to provide the brick and mortar for the data centre competitively.
As for YTL Power, the research outfit said the negatives are priced in with data centres fully discounted in its share price. At this juncture, firm takers for YTL Power’s AI data centre GPU as a service may still be needed to re-rate the stock.
YTL Power International Bhd, Inari Amertron, Nationgate Holdings and PIE Industrial closed at RM3.11, RM2.52, RM1.79 and RM4.57, respectively, yesterday.
By ELIM POON
Related posts:
