RHB Investment Bank is turning cautious on Malaysia’s data center (DC) as the US artificial intelligence diffusion (USAID) rules present a structural risk to the multi-year growth story.
“Broadly, we expect the policy overhang to cloud the prospects of the DC market in the short to medium term, potentially stifling investments,” the research house said in a note on Monday.
According to RHB, the proposals may have wider implications across the entire technology value chain.
It noted that concerns are on co-location providers where growth may be impeded in the medium to longer-term.
It highlighted the new rules favor US-based hyper scalers with AI computational capacity controlled via trusted allies (Tier-1 countries).
It sees the policy response will likely crimp the exports of graphics processing units (GPU) to Tier-2 markets, with non-verified end user (VEU) entities likely to be the worst hit via capacity caps and conditional quotas.
“In our view, wholesale co-location providers stand to lose more, specifically those with Chinese links and/or off takers,
“Co-location providers make up about 70 percent of DC inventories (including live supply, under construction and committed), by our estimates,” said the research house.
It is noted that the USAID policy tightens exports controls implemented in 2022 and 2023.
It plugs existing loopholes that allowed China-based companies to access US-AI technology by leasing access via other markets and the cloud.
Under the interim final rule announced on Jan 13, global licenses will be required for the importation of US AI chips with countries defined under three groups.
Close allies and partners (Tier-1) are exempted from regulatory controls on AI exports as they carry low risk of theft and diversion.
Countries of concern (Tier-3) – those where the US has an arms embargo (notably China and Russia) – will see a blanket ban enforced.
The majority of countries (including Malaysia) fall under the middle category (Tier-2) where significant AI investments are being made by US and non-US-based hyper scalers and co-location providers.
The key measures announced included regulatory exemption on the sale of AI chips to 18 key allies and partners.
Besides, entities with US-based parents and in Tier-1 markets are eligible to apply for universal verified end user (UVEU) status which would allow them to place up to 7 percent of their global AI computational capacity in Tier-2 markets.
UVEU/US-based UVEU must have 75 percent/50 percent of their computing capacity in Tier-1 countries/US.
Meanwhile, entities in Tier-1 and Tier- 2 markets can apply for a national validated end-user (NVEU) status which allows for AIDCs albeit subject to volume controls (up to 320,000 H100 GPU equivalents) from 2025-2027.
Closed models are also subject to export restrictions.
According to RHB, the proposals are subject to public comments and a final review by the Trump administration before it takes effect, effectively 120 days from Jan 13 (the day the executive order was made by the Biden administration).
“It is possible that the new administration would look to undercut the regulation with the president elect previously stating his intention to loosen AI rules,” it said.
RHB, however, believed the direct impact for the restriction on Malaysian technology firms is limited.
“While the indirect impact is difficult to ascertain, we note that local technology supply chains are insignificant in the global AI supply chain,” said the research house.
It remains optimistic of a stronger 2025 on the back of a sector recovery, fueled by firmer broad-based demand and the replacement cycle.
According to RHB, the new restrictions could affect supply chains that are in the ecosystem of GPU and central processing unit (CPU) servers.
Broadly, it believes the potential curbing of AI-related chip exports – a major growth driver for the current semiconductor upcycle – could lead to a sector-wide slowdown, affecting the entire supply chain.