The numbers are worth watching because Intel’s large Penang advanced-packaging expansion, codenamed Project Pelican, is the most visible candidate for handling a major new HBM flow in Malaysia. But “candidate” is doing substantial work here. Country-level import and export data can show destination, value and product category; it cannot by itself identify the consignee, the final package house, the design customer, or whether a shipment is intended for immediate assembly rather than inventory, redistribution or another supplier’s operation.
What the figures do show is a potentially significant shift in where the AI chip supply chain is physically being assembled. What they do not show is that Intel has already won a meaningful share of packaging business away from TSMC.
The shipment shift is real enough to examine, but the attribution is not proven
SemiAnalysis markets ChipBook as a subscription semiconductor tracker built around trade and supply-chain data, and it has separately noted growing Korean memory exports to Malaysia. That is consistent with the broad direction described by TechPowerUp: Malaysia is receiving more high-value memory as AI hardware supply chains expand beyond their traditional Taiwanese concentration.
HBM is not a retail memory product that simply arrives at a regional warehouse. It is stacked DRAM intended to sit beside powerful processors, GPUs or custom AI accelerators, usually through advanced 2.5D or 3D packaging. A sharp rise in shipments to Malaysia therefore points to a real increase in local demand for sophisticated packaging, assembly and test work.
But the original report makes a further leap: because TSMC does not operate a comparable advanced-packaging plant in Malaysia, Intel must be the logical destination. Intel is certainly a logical possibility. It is not the only explanation that the trade data can support.
Malaysia already has a deep semiconductor assembly-and-test industry, including multinational outsourced semiconductor assembly and test providers, component suppliers and electronics manufacturing operations. Some HBM could be moving through an existing supply chain rather than into Intel’s cleanrooms; some could be imported by an intermediary; and some could be tied to a customer using Malaysia for board or system integration. The report does not identify the importer, the package type, the HBM generation, or the chips that the memory will accompany.
Until those details emerge, the appropriate conclusion is narrower: Malaysia is becoming a more important destination for Korean high-bandwidth memory, and Intel is positioned to benefit if Project Pelican has entered volume operation.
Project Pelican is a major facility, but its ramp remains the missing fact
Intel announced its Malaysian expansion in late 2021 as part of a $7 billion investment program, and the Penang project was designed to become its largest advanced packaging facility for 3D packaging. The company had originally expected the facility to reach operational readiness in 2024, but Intel later said it would complete construction while aligning the startup with demand and utilization at its existing factories.
That delay matters when interpreting the HBM shipment story. A factory that is structurally complete is not necessarily a factory producing high volumes of advanced packages. Tool installation, process qualification, customer validation and yield ramping can separate a building’s completion from commercial output by months or longer.
In December 2025, Malaysia’s InvestPenang said Intel’s advanced-packaging plant had reached 99% completion, with about RM12 billion in infrastructure capital expenditure. In March 2026, Malaysian Prime Minister Anwar Ibrahim said Intel planned to begin the first phase of operations later in 2026, after a briefing with Intel chief executive Lip-Bu Tan and Intel Foundry leader Naga Chandrasekaran.
Those public milestones conflict with the looser implication in TechPowerUp’s article that Pelican had already completed its full buildout late last year and was ready to absorb a large share of global HBM traffic. Construction may be essentially finished, but the publicly described schedule still put the first operational phase in the second half of 2026. Intel has not publicly disclosed Pelican’s present production status, package output, customer count, or process yields.
That absence is central. If the facility is only beginning phased operations, a $1.3 billion HBM import number would need to be reconciled with its actual qualified capacity. If it is already well into a rapid ramp, the shipment pattern becomes much more suggestive. Neither Intel nor SemiAnalysis has publicly supplied the operational data needed to distinguish those cases.
EMIB and Foveros give Intel a credible HBM route
Intel does have packaging technologies suited to exactly the type of products that consume large amounts of HBM. Its Embedded Multi-Die Interconnect Bridge, or EMIB, links dies through a small silicon bridge embedded in the package substrate. Foveros stacks dies vertically. Both approaches are intended for heterogeneous packages that combine logic, I/O and memory chips from potentially different manufacturing sources.
Intel’s foundry materials explicitly position EMIB for logic-to-HBM integration, while its newer EMIB-T approach adds through-silicon vias to improve power delivery and routing in larger packages. The company has also emphasized packages beyond the normal reticle-size limit, a capability relevant to AI accelerators that connect multiple compute dies to multiple HBM stacks.
That technology does not make Intel a direct substitute for every TSMC CoWoS customer. CoWoS is an established packaging family with a broad production base and an enormous installed customer pipeline, particularly around leading AI GPUs. EMIB, Foveros and CoWoS also differ in their interconnect architectures, design rules and supplier qualification requirements. Switching a high-end accelerator from one packaging flow to another is an engineering project, not a logistics decision prompted by an HBM shipment.
The more plausible near-term reading is that growing HBM imports could reflect new demand seeking additional packaging capacity, rather than a clean transfer of existing TSMC business to Intel. AI vendors have strong incentives to qualify alternative packaging routes because HBM, substrates, advanced assembly capacity and test capacity can all constrain system shipments. An additional Intel line in Malaysia would be strategically valuable even if it handled incremental work rather than displacing an established CoWoS program.
Amkor’s EMIB deal supports the capacity story, not the Malaysia claim
Intel’s effort to broaden EMIB production is not confined to Malaysia. In April 2025, Amkor Technology announced a strategic partnership with Intel to implement EMIB assembly processes in South Korea, Portugal and Amkor’s planned Arizona site. Amkor said the arrangement was intended to expand the EMIB ecosystem for high-performance computing, AI and other data-intensive applications.
That partnership independently supports the idea that Intel sees outside assembly capacity as part of its advanced-packaging strategy. It also undercuts any assumption that every EMIB-related HBM shipment must lead to Pelican. Intel’s packaging network spans its own sites and outsourced partners, and the company can qualify more than one geography for a customer program.
Amkor’s 2026 investor materials describe several advanced-packaging engagements already in production, but they do not identify Intel customers, EMIB volumes or HBM programs. In its first-quarter 2026 earnings call, Amkor characterized the EMIB collaboration as continuing but still early in its development cycle. That is a considerably more measured picture than a confirmed surge in large AI-customer orders.
The same caution applies to claims that Pelican has doubled the capacity of Intel’s New Mexico packaging operation. Intel’s public materials describe New Mexico as a flagship advanced-packaging location and confirm its long-running EMIB and Foveros work. However, the public record reviewed for this article does not provide a comparable unit-capacity figure for Pelican, nor a package-per-month comparison with New Mexico.
What would turn the signal into evidence
The Malaysian HBM surge will become much more meaningful if it is followed by any of three disclosures:
- Intel could identify Pelican as operational and name the packaging technologies or customer programs entering volume production there.
- Memory suppliers such as SK hynix, Samsung Electronics or Micron could disclose an HBM supply relationship tied to an Intel Malaysia packaging flow.
- Customs or supply-chain data could identify a consignee, importer or product pattern that directly connects HBM deliveries to Intel’s Penang operations.
For now, the data is a strong signal that Malaysia’s role in AI hardware assembly is expanding. It is not proof that Intel has diverted billions of dollars in HBM demand from TSMC, nor proof that Project Pelican is already running at the scale implied by those shipments.
Intel’s Penang facility remains strategically important precisely because it could give foundry customers another route for large HBM-equipped packages. The next hard evidence will not be another destination-country total; it will be Intel disclosing what Pelican is actually building, for whom, and at what volume.
