Maybank Investment Bank has estimated ASEAN data center growth at 20 percent revenue compound annual growth rate (CAGR) by 2028.
The research house said in a note on Tuesday that of this, it estimated about 14 percent is driven by ASEAN’s own needs and 6 percent to be driven by global demand.
This is because it sees ASEAN could emerge as a data center hub on the back of its favorable cost economics and power/water infrastructure.
Assuming half of the 6 percent global demand comes from China and assuming China is totally banned from hosting data or running large language models (LLMs) in ASEAN, it noted overall demand growth would reduce to 17.2 percent.
“As such demand doesn’t fall off the cliff. While China company led investments in ASEAN may face increased regulatory hurdles, we think this will be small at about 10 percent of total investments,” it added.
Following stringent new US export controls to limit access to advanced artificial intelligence (AI) technologies, Maybank chooses not to be overly negative and rather see some pockets of potential optimism.
While AI provides long-term growth optionality within ASEAN, the research house sees medium-term growth being driven by ASEAN catching up on traditional data centers for cloud computing and storage.
It estimated ASEAN is 55 percent to 70 percent behind evolved markets on data center infrastructure on a per capita or adjusted for gross domestic product (GDP) basis.
“Large AI-led data center announcements in ASEAN are US hyper scaler-led, which we see as relatively better placed to navigate Tier-2 market sanctions,
“Additionally, we see potential for greater demand for GPU as a service (GPUaaS) in ASEAN if access to advanced graphics processing unit (GPUs) in China is curtailed,” it said.
According to Maybank, under new US restrictions, companies headquartered in Tier-1 countries can deploy no more than 7 percent of their total computing power in any Tier 2 country.
“Large investments in ASEAN are announced by US based hyper scalers and we estimate those investments are in the 2 percent to 3 percent range per market of their expected AI-linked investments globally,” it said.
As an example, it noted Microsoft has pledged $2.2 billion cloud and AI investment in Malaysia whereas its planned FY25 global investment is $80 billion.
“Tier-2 headquartered companies, through national validated end user (VEU), can deploy 690,000-equivalent H100 GPUs over 2025-2027,
“We estimate this will entail total investment of $17 billion whereas we estimate the largest deployment in ASEAN is $4.3 billion by YTL power,” it noted.
Given this, it thinks the US restrictions are accommodative and announced data-center builds are unlikely to be impacted even in Malaysia which has attracted the bulk of the investments.
It is noted that the US framework introduces controls on exporting AI model weights, focusing on extremely advanced models that require over 1026 FLOP for training (costing at least USD70m to develop).
BIS estimates that fewer than five models globally currently exceed this threshold.
Open-weight models are excluded from these controls.
“Based on our understanding, model controls are to restrict advanced AI models from training and inferencing in Tier-2 markets,” said Maybank.
While access controls are placed on model training/access, however, with a bigger restriction on AI GPU exports to China, Maybank thinks Chinese companies may still use ASEAN
infrastructure focusing on less advanced AI systems or leverage public/open-weight models that fall outside of US controls.
PublicInvest: US AI chip export restriction has muted impacts on Malaysia technology sector