MIDF Research has on Thursday maintained its positive outlook on Malaysia’s data center despite news reports have surfaced, alleging that Chinese firms were training their artificial intelligence (AI) models at a data center in Malaysia that are powered by Nvidia chips.
The research house said in a note that there is no evident slowdown of or delay in ongoing projects and contractors are still actively bidding for new data center construction
jobs.
It is noted that just last month, Gamuda sold 389 acres of land in Port Dickson to Google-linked Pearl Computing Malaysia Sdn Bhd and signed a MYR 1.01 billion ($240 million) external infrastructure contract for enabling works for data center development while SunCon secured a MYR 1.16 billion ($270 million) contract from a US tech giant to build two data centers.
Microsoft also recently reaffirmed its commitment to a MYR 10.5 billion ($2.47 billion) investment in cloud and AI infrastructure in Malaysia, including the development of hyperscale data centres in the Klang Valley.
MIDF also reiterated its opinion that not all data centers in Malaysia are AI data centers and while most of them are AI-ready, they may eventually be utilized for non-AI purposes.
For example, YTL Power International previously allocated 100MW for AI from its 500MW data center in Kulai, of which the first 20MW is expected to be launched in August.
The remaining 80MW had since been converted into a colocation data center due to a shift in market landscape that saw a rush for colocation space.
The Wall Street Journal reported recently that four engineers from Beijing flew to Malaysia in March, carrying suitcases containing 80TB of hard drives containing spreadsheets, images and video clips for training an AI model.
It further claimed that a total of 300 servers were rented at a Malaysian data center where the engineers fed the data into the servers and have returned to China recently with a few hundred GB of data.
The Chinese company set up a Malaysian entity with three Malaysian directors and an offshore holding company upon the advice of the data center when Nvidia and its vendors started to have stricter audits on the end users of the chips.
The Ministry of Investment, Trade and Industry (MITI) of Malaysia is still in the process of verifying the matter.
While servers using Nvidia and AI chips are not classified as controlled goods under the Malaysian Strategic Trade Act 2010, the ministry said Malaysia will cooperate with any government that required assistance in monitoring trade in sensitive goods under the export control of their respective countries.
MITI added that while data centres operating in Malaysia are free to make their own commercial decisions, provided they operate within the scope of Malaysian laws and regulations, the country stood firm against any individual or company that attempts to circumvent export controls or engage in illicit trade activities.
“As we await further clarity from MITI on the veracity of the news reports, we strongly believe the Nvidia chips involved were that of previous generations and not the latest GB200,
“As the allegations were said to happen in March, we believe this could be a move to ‘speed up the process’ before the eventually rescinded Framework of AI Diffusion that was expected to come into force on May 15,” said MIDF.
It is noted that the framework, also known as the AI Diffusion Rule, was introduced during the final days of the Biden-Harris administration, which essentially puts in place restrictions for the exports of US-manufactured AI chips and controls on AI model weights through a three-tier country system.
The rule was eventually rescinded by the Trump administration, just two days before it was supposed to be enforced.
The Bureau of Industry and Security (BIS) stated that the rule would have stifled American innovation and saddled companies with burdensome new regulatory requirements and undermine US’ diplomatic relations with dozens of countries by downgrading them to second-tier status.
The BIS also issued three guidance documents after rescinding the rule, on the risks of using China’s advanced computing integrated chips (ICs); consequences of allowing US AI chips to be used for training and inference of Chinese AI models and how to protect supply chains against diversion tactics.
“As the Trump administration refashions the AI Diffusion Rule, among speculations include replacing the three-tiered country system with a licensing regime based on G2G
agreements,
“Regardless of the changes, we believe the essence of it remains, which is to contain China’s AI advancement,” said the research house.
As far as Malaysia is concerned, it expects negotiations to revolve around the regulation and enforcement on the movement of US-made AI chips, the monitoring of its end-users and its seriousness in dealing with those that violate the control measures.
MIDF noted that the scrapping of the rules under the Framework of AI Diffusion is a positive development as far as the data center value chain is concerned.
“For one, we believe there will be looser restrictions on the allocation of AI computing power by tech giants as they may no longer be tied down to the 7 percent restriction of AI computing power to any country outside Tier 1, which allows for more AI capacities to be planned in other countries such as Malaysia,” it noted.