Malaysia’s semiconductor industry is benefiting from accelerating China+1 supply chain diversification, deeper localization by global semiconductor equipment makers and rising demand for optical and power semiconductors tied to artificial intelligence (AI)-driven data center expansion, while a recovery at Intel is also providing support to the sector, Hong Leong Investment Bank Research said Wednesday.

The research house said in note that Malaysia’s outsourced semiconductor assembly and test (OSATs) largely sat out the first leg of the AI upcycle in 2024-2025 (driven by Graphics Processing Unit [GPU]/High Bandwidth Memory [HBM] demand), tracking instead the protracted analog downcycle as post-pandemic inventory correction weighed on soft auto/industrial demand.

“We see 2026 as the inflection point. As AI DC buildout accelerates, high-performance power management chips are emerging as the new demand driver for analog suppliers, therefore offering a new growth vector for Malaysia OSATs as well,” it noted.

According to the research house, the reason is straightforward – AI server racks are becoming denser and more power-hungry (from 120-150kW to 600-1,000kW eventually), necessitating more power semi content across the entire power delivery chain.

Efficient power delivery is increasingly a binding constraint in AI DC deployment, and ongoing architectural shifts are structurally expanding addressable content.

Industry leader, Infineon estimates the TAM (total addressable market) for power semi in AI could expand about 3 times by 2030, reaching $9 to $14 billion from about $4 billion in 2026.

“For Malaysia’s OSATs, we see this as a meaningful re-rating catalyst from 2026 onwards, with revenue growth expected to track the accelerating order momentum across their analog customer base,” said the research house.

According to the research house, there are two major shifts in the industry. Firstly, NVIDIA is leading the transition toward 800VDC architecture (Kyber) for AI DC deployment starting in 2027.

Current 415VAC architecture is already hitting physical limits (rising copper density, lower efficiencies from conversion losses) and cannot support the step-change in rack power requirements (120–225kW for Blackwell/Rubin, scaling to 600–1,000kW on Rubin Ultra and Feynman), it noted.

Adding to this, wide-bandgap materials (SiC, GaN) will gain more usage given their superior efficiency in high-voltage workloads.

Secondly, wider adoption of vertical power delivery (VPD) — placing the VRM directly beneath the AI chips helps to shrink transmission distance and reduce thermal load, said Hong Leong.

It noted the shift from lateral to VPD is estimated to lift Voltage Regulator Module (VRM) bill-of-materials ~3x at the compute-board level, said the research house.

At the compute-board/VRM tier — where only a handful of analog players are qualified into hyperscaler designs — Unisem’s anchor exposure here is MPS, while MPI is aligned to Infineon.

Further up the chain, it noted the Power Supply Unit (PSU)/Intermediate Bus Converter (IBC) segment is more fragmented and competitive, with multi-sourcing the norm.
MPI carries the broader, well-diversified analog base (e.g. Infineon, STMicro, TI, Wolfspeed), while Unisem’s exposure is narrower (e.g. MPS, Power Integrations, Renesas).

On wide bandgap (SiC/GaN), MPI has the edge given that Infineon and Wolfspeed are among the leading players globally, it added.

“The thesis has played out faster than anticipated — supply tightness in analog semi is already emerging, with multiple Integrated Device Manufacturers (IDMs) announcing price increases over the past one to two months,” it said, adding that an additional tailwind to this is the accelerating China+1 diversification, further benefiting Malaysian players.

Similar to HBM, Central Processing Units (CPUs) and GPUs, Hong Leong highlighted the demand for power semi is starting to run into capacity limits.

Recent commentary from major power semi suppliers such as Renesas and Infineon suggests that their AI server-related growth is becoming more supply limited than demand-capped, with delivery increasingly dependent on how quickly both internal and partner capacity can scale.

Onsemi’s recent commentary also points toward lengthening lead times for AI DC-related power components.

“This is beginning to translate into pricing power, with major analog suppliers such as Infineon, Texas Instruments, Analog Devices, STMicro and NXP announcing price increases across Power Management Integrated Circuits (PMICs), MOSFETs (foundational components in electronic power systems) and industrial control chips to pass through higher input costs, said the research house,” it said.

Notably, it said the market is also beginning to price in a potential AI “halo effect”, where stronger demand for high-voltage MOSFETs, PMICs and other AI-related power devices could tighten availability for adjacent automotive and industrial applications.

Importantly, this tightening is occurring before the full ramp of next-generation 800VDC products, where a multi-fold increase in power semi content is expected to begin translating into deliveries in 2027.

“In our view, this should drive analog players to accelerate capacity expansion, with OSATs playing an increasingly important role in backend assembly, testing, and burn-in as analog IDMs prepare for the next wave of power semi demand,” said Hong Leong.

Malaysia boosts AI, semiconductor push for technological sovereignty amid global uncertainties