With a maximum artificial intelligence (AI) chip install base of up to 49,900 H100-equivalent graphics processing units (GPUs) per Tier-2 country from 2025 to 2027 under US recent AI chip curbs, the growth of data center (DC) projects in Malaysia could face constraints, TA Securities said last Friday.
The research house said in a note that this limitation may hinder the expansion plans of existing DC players and delay the rollout of new DC projects in Malaysia.
Furthermore, it said the 7 percent cap on AI computational power per country and the 25 percent aggregate cap for all Tier-2 countries impose additional restrictions on the maximum size of DC capacity in Malaysia, depending on the hyper scalers’ total global DC capacity.
That said, research by SemiAnalysis suggested that global AI-driven DC capacity to reach approximately 20GW by 2025 and to triple by 2027.
This significant growth is fueled by the urgent global demand for hyper scalers to scale up their DC capacity to accommodate large-scale cloud storage needs, particularly driven by the rise of AI-intensive mobile and desktop applications and software, it added.
“Given these projections, we believe the impact from this new policy could be minimal and remain optimistic about the robustness of DC projects in the pipeline,” said the research house.
According to TA, the 7 percent cap of the projected 60GW global DC capacity by 2027 contributed by mainly US companies translates to an allocation of 4.2GW per Tier-2 country.
In this context, Malaysia has substantial room for growth, considering its total DC capacity stood at 504.8MW as of the end of 2024, according to the Malaysian Investment Development Authority (MIDA).
Overall, TA believes the headwinds arising from the U.S.’s new restrictions are likely to be temporary.
With the new rules set to take full effect in April 2025, it noted the upcoming administration under President Trump has time to gather feedback from industry experts and stakeholders.
It opined that this could lead to potential adjustments aimed at alleviating the restrictions and ensuring sustainable growth in the DC sector.
Key downside risks, however, include a narrower quota than the initial 7 percent of total computational power per country in each Tier-2 country; delays in the rollout of new DC projects by hyper scalers, and more stringent regulations and capping imposed by the U.S., including the potential suspension of AI chip exports to Malaysia.
On January 13, the Biden-Harris administration introduced an Interim Final Rule Titled “Export Control Framework for Artificial Intelligence Diffusion” through the Bureau of Industry and Security.
The policy aims to prioritize the application and development of AI chips within the U.S. while streamlining the export process for trusted partners, including key U.S. hyper scalers such as Microsoft, Google, and Amazon, as well as select approved international providers.
The new regulation, which introduces licensing requirements for DC semiconductors, and has initiated a 120-day comment period – starting from January 13 2025 – to allow revisions and public feedback before implementation.
This framework categorizes countries into three tiers:
– Tier 1: includes the U.S. and 18 allied nations with unrestricted access to advanced AI chips;
– Tier 3: encompasses countries under U.S. arms embargoes, including China, Iran, and North Korea, which are barred from acquiring such technology;
– Tier 2: covers over 100 other nations with varying export restrictions and caps including a maximum AI chip install base of up to 49,900 H100 equivalent GPUs from 2025 to 2027.
A validated end user (VEU) framework further ensures that only specific trusted entities can import these chips within each tier, with limited exceptions for non-VEU organizations.
All companies seeking to deploy GPUs must submit a VEU authorization application, which involves meeting 19 separate certification and policy requirements to qualify as either a Universal VEU (UVEU) or National VEU (NVEU).