(This is part 2 of a two-part article. Read part 1: Southeast Asia’s chance to become a strategic hub for the AI global supply chain.)
AFTER more than two decades of sourcing shifting to China, Southeast Asia has once again taken the limelight in the global supply chain, this time in the competition over AI hardware. With foreign direct investment flowing in and large capital going into manufacturing infrastructure, the question is how the region will evolve over the next decade.
Southeast Asia’s participation is concentrated in the downstream half of the AI hardware value chain, and that is where the most consequential changes of the coming decade will unfold.
Four developments stand out
The first is the migration of advanced packaging into the region (stage five). Assembly, test and packaging was long the commodity end of chipmaking, but the shift towards chiplets and three-dimensional integration has made it a strategic bottleneck in its own right. Intel's advanced packaging complex in Penang, its first such facility outside the United States, begins operations in 2026, while ASE, Amkor, and Samsung are all expanding packaging and test capacity across Penang, Bac Ninh, and Thai Nguyen. A consortium of Malaysian firms is targeting seven percent of the global advanced packaging market by 2035.
The second development is the deepening of the component base (stage six). AI has revived demand for the region's traditional strengths rather than displacing them. Roughly 87 percent of the data held in large cloud data centers is stored on hard drives, and Thailand, which produces about 80 percent of the world's supply, is attracting renewed expansion investment from Western Digital and Seagate. Thailand has simultaneously become Asean's largest printed circuit board manufacturing base, drawing more than 200 billion baht in investment applications since 2022 as Taiwanese and Chinese producers relocate capacity. Infineon's silicon carbide fab in Kulim positions Malaysia in the power semiconductors that data centers and electric vehicles need.
Third, system and device assembly is consolidating the China Plus One shift (stage seven). Server and device assembly, historically concentrated in China, is spreading across Vietnam and Thailand as contract manufacturers follow their customers' diversification mandates. The constraint is depth rather than breadth: localisation rates in Vietnamese electronics remain around five to ten percent, with most components still imported from China and Northeast Asia.
Fourth, the data center layer is regionalising around a small number of corridors (stage eight). Singapore's land and power constraints are pushing capacity into Johor, which has a pipeline of roughly five gigawatts increasingly positioned for AI training workloads, and into Batam, which serves compliance-driven and Singapore-overflow demand. Malaysia and Indonesia are both projected to overtake Singapore's share of regional data center capacity by 2028, and data sovereignty rules across Indonesia, Vietnam, Thailand, and Malaysia are reinforcing in-country buildout at the same time. By the early 2030s the likely configuration is a Singapore-Johor-Batam core complemented by national hubs around Jakarta, Bangkok, and Hanoi.
Taken together, the center of gravity (Figure 1, stages five to eight) is shifting towards Southeast Asia, and the region’s role is thickening from a collection of discrete manufacturing sites into interconnected clusters spanning adjacent stages of the value chain.
What are the key bottlenecks and challenges for supply chain strategy?
The developments above rest on inputs that Southeast Asia does not control. The upper half of Figure 1 remains concentrated elsewhere: Taiwan fabricates roughly 90 percent of leading-edge logic chips, two Korean firms control approximately 80 percent of high-bandwidth memory, and critical equipment and materials are sourced from the Netherlands, the United States, and Japan. High-bandwidth memory capacity is effectively sold out through 2026, and semiconductor lead times reached 40 weeks in early 2026. A packaging plant in Penang or a data center in Johor is therefore only as reliable as its allocation of wafers, memory, and specialty gases, a dependence the Strait of Hormuz helium shortage made visible.
The second set of constraints is physical. Regional electricity demand from data centers is projected to rise steeply, and the corridors absorbing it are where energy and water are already contested: Johor has seen the region's first residents' protests against data centers, and experts are urging that new facilities be sited in more water-rich areas.
Third, talent may bind before land or capital do. Penang alone has been estimated to face a shortfall of around 50,000 engineers relative to semiconductor industry demand, and the skills gap in Vietnam's electronics sector is widening as the industry shifts towards semiconductors and AI hardware.
Fourth, because the region's participation is concentrated in the lower-margin stages (Figure 1), capacity may grow faster than capability, leaving Southeast Asia executing processes that are designed, tooled, and priced elsewhere.
What are the policy implications for governments?
Three implications follow for governments.
First, talent pipelines are the binding constraint most within policy control: vocational and university programs tied to anchor investors, of the kind pursued under Malaysia's National Semiconductor Strategy and Vietnam's semiconductor workforce initiatives, are likely to matter more than further rounds of tax competition.
Second, infrastructure policy needs to treat power, water, and grid capacity as part of industrial strategy, siting data centers where resources allow and accelerating renewable supply, rather than approving capacity that outruns utilities.
Third, because the cluster logic is regional, the payoff to coordination across borders, visible in the Johor-Singapore Special Economic Zone and the Singapore-Johor-Batam corridor, is likely to exceed the payoff to national subsidy races that duplicate the same value chain stages in every country.
Alongside all three, compliance with widening export controls is becoming a condition of participation, so trusted-node status with the major chip-producing economies is itself a policy asset. Each country would also benefit from a supply chain foresight strategy that builds on existing competencies to advance up the value chain.
What are the strategic implications for multinationals?
The region is best treated as a portfolio of complementary specialisations rather than a single location: front-end fabrication and design in Singapore, packaging in Penang and northern Vietnam, storage and boards in Thailand, assembly across Vietnam, and computing capacity in the Singapore-Johor-Batam corridor.
Resilience requires securing long-term upstream allocations of wafers, memory, and advanced packaging slots contractually, since downstream capacity without upstream supply is idle capital.
Co-location decisions should be paired with deliberate hedges against the risks that co-location creates, including diversified energy contracts, cyber hardening, and geographic redundancy within the region.
Finally, supplier development is strategy rather than philanthropy: raising local content reduces exposure to the China-routed intermediate components that tariff and rules-of-origin scrutiny increasingly target.
At this juncture, Southeast Asia's claim to be a strategic hub for the AI supply chain is real but partial. The region's established strengths sit in packaging, components, assembly, and increasingly in computing infrastructure (see the lower half of Figure 1), while the chokepoints that define the industry — namely leading-edge fabrication, high-bandwidth memory, equipment, and materials — remain concentrated in Northeast Asia, the United States, Europe, and Japan.
The coming decade will determine whether downstream scale converts into upstream capability. The tests are already observable: whether Intel's Penang complex and Malaysia's advanced packaging consortium seed a genuine advanced packaging ecosystem, and whether Vietnam's back-end boom deepens into component and design capability.
If these succeed, Southeast Asia will have moved from being where AI hardware is assembled to being one of the places on which the AI supply chain depends.
Dr Neale O’ Connor is associate professor at LaTrobe University, Australia and Dr Andrei OJ Kwok is associate professor at Monash University Malaysia