The Economic Daily News report on August 18 about HBM exports to Malaysia added new material to the discussion surrounding Intel's advanced packaging business. Citing SemiAnalysis Chipbook export data, the paper reported that approximately $1.3 billion worth of HBM was shipped to Malaysia, while shipments to Taiwan fell below $3 billion. HBM is not a finished AI accelerator but rather memory that gets integrated into the same package as a logic die. Changes in export destinations can serve as a clue to where back-end processes are beginning to move.
However, trade flows alone cannot reveal the receiving companies, packaging methods, or the brands of the final products. The published Economic Daily News article does not specify the aggregation period or the HS codes covered for the approximately $1.3 billion and sub-$3 billion figures. Beyond the comparison baseline, the shipping and receiving companies also remain unknown. These dollar figures cannot be translated into Intel's foundry revenue or market share. Even so, the industry sources cited by the paper, Intel's publicly announced Penang capacity expansion, and the commercialization stage of EMIB-T all point in the same direction—suggesting that back-end processing load may be shifting toward Intel Malaysia.
How to Read the $1.3 Billion in HBM
Economic Daily News named Intel as a Malaysian facility capable of integrating large volumes of HBM with logic, reporting that Intel Malaysia is supporting a joint major customer of TSMC and Intel in back-end processes. What can be confirmed here is the paper's own exclusive reporting alongside a logistics-based signal derived from export data. Neither the customer nor the product has been identified, and neither the contract value nor the volume has been disclosed.
Malaysia also hosts a broad semiconductor assembly and testing industry beyond Intel. Therefore, it cannot be assumed that all HBM shipped to Malaysia is destined for Intel. Export value represents the sales value of HBM, not a figure indicating packaging processing capacity or foundry revenue. The contrast between roughly $1.3 billion and Taiwan's sub-$3 billion figure is a striking comparison, but without a defined period or denominator, it cannot serve as grounds for discussing changes in monthly processing volume or regional share.
It's also important to separate dollar value from physical volume. Because HBM pricing varies by generation, capacity, and contract terms, export value alone cannot determine how many stacks were shipped. Public information also cannot confirm whether Malaysia is the final integration site or merely a logistics hub before shipment elsewhere. What this figure indicates is the existence of a large material flow—not, on its own, a measure of advanced packaging production volume.
At the same time, if HBM is moving between regions before integration with logic, the location of back-end processes—including assembly and testing—becomes more significant than before. While the Economic Daily News information does not confirm a single contract, it does offer a starting point for tracking where advanced integration capacity is being established.
Penang: Capacity Expansion and New Facility Plans in Parallel
In its Q1 2026 earnings release on April 23, Intel announced that it had expanded assembly and test capacity in Penang to support customer products in response to rising global packaging demand. While the release did not disclose technology names, customer names, or shipment volumes, it represents an existing investment that geographically overlaps with the HBM inflow reported in August.
Separately, on March 17, Prime Minister Anwar Ibrahim stated that the first phase of Intel's new advanced packaging complex would begin assembly and testing operations, with the plan calling for operations to start sometime during 2026. Expanding existing Penang capacity and bringing a new complex to full operation are not the same thing. Export data alone cannot confirm whether the new facility's launch has been completed or whether any specific plan has reached full-scale operation.
This distinction matters for reading supply capacity accurately. Even though both the demand absorbed by expanded existing lines and the capacity handled by Phase 1 of the new facility are slated to begin operations within the same year of 2026, the actual scale of mass production and the timing of customer qualification could differ significantly between the two.
Advanced packaging cannot move directly to commercial shipment simply because buildings and equipment are in place. Each customer product requires process ramp-up, yield and reliability verification, and qualification of mass-production conditions. Intel's explicit statement as of Q1 that it expanded capacity to support customer products indicates that the existing site already plays a role in meeting demand. Regarding the new complex, however, what can be confirmed is only the plan for Phase 1 to begin operations within the year—conflating the two risks overestimating the actual timing of supply availability.
EMIB-T Backlog Grows, But Customer Ramp-Up Targeted for 2027
During its Q2 earnings call on July 23, Intel stated that customer interest in EMIB-T is very strong and that the backlog is growing. The company said yield and reliability have reached internal targets, and that focus is now shifting toward transitioning to a high-volume, high-quality production system to support customer ramp-up in 2027. What the company disclosed here are the stages of backlog growth and future customer ramp-up. Neither specific yield figures nor backlog dollar amounts were disclosed. Customers and products were not identified, nor was any 2026 mass-production shipment volume indicated.
EMIB is a 2.5D packaging technology that connects logic-to-logic or logic-to-HBM. Intel describes it as a method of locally embedding a silicon bridge into an organic substrate, with EMIB-T adding TSVs to that bridge. EMIB in general—using both Intel-made and externally manufactured silicon—has been in mass production since 2017, but this track record should not be treated as evidence that ultra-large EMIB-T products for external customers are already shipping in mass production.
In other words, the reports of Penang's capacity expansion and HBM inflow do not confirm mass production in 2026 of any specific product using EMIB-T. Based on the timeline Intel itself has presented, building the systems to support mass production and the 2027 customer ramp-up remain the primary confirmable milestones for now.
Growing Back-End Orders and TSMC's Front-End Business Are Separate Matters
During TSMC's earnings call on July 16, CEO C.C. Wei stated that packaging capacity has become so tight that it is constraining customer growth. He went on to welcome additional market options like EMIB-T, explaining that if competitors take on a share of the back-end processing load, it would actually help TSMC's core front-end wafer business.
In the same call, TSMC treated its front-end wafer business and back-end processes as separate businesses, stating that the gap between supply and demand is particularly large in back-end processes. In a scenario where Intel takes on back-end foundry work and packages TSMC-made logic dies using EMIB-T, Intel would gain back-end revenue and customer relationships, while TSMC could maintain its front-end wafer revenue. Winning back-end business is not equivalent to taking away TSMC's front-end share.
Under this structure, the two companies may compete over portions of the same deals while still complementing each other across the overall supply chain. If AI accelerator demand is bottlenecked at the back end rather than the front end, then having more pathways to deliver its own wafers to finished products actually benefits TSMC as well. CEO Wei's comments should be read not as an acknowledgment of Intel's technological superiority, but as a welcome for additional routes that help resolve capacity shortages.
TSMC has also indicated that, out of its 2026 capital expenditure budget of $60–64 billion, 10–20% will be allocated toward advanced packaging, testing, mask manufacturing, and similar areas. While the exact breakdown by use has not been disclosed, it is clear that the back-end shortage cannot be explained by the technological superiority or inferiority of any single company alone.
Indicators to Watch for Future Disclosure
To bring this logistics signal closer to business reality, the receiving company and the monthly HBM processing volume need to be confirmed first. Beyond that, key indicators would include which packaging method customers have qualified, what level mass-production yields have reached, and when finished products actually shipped. None of these have been publicly disclosed at this time.
Intel's official information in particular links the growing EMIB-T backlog to customer ramp-up in 2027. Rather than reinterpreting the August export figures as confirmed 2026 mass production of a specific product, it would be more productive to track whether Penang's expanded capacity actually translates into customer qualification and finished product shipments—this is what will reveal how far Intel's back-end business truly extends.
For now, the continuity of export statistics also offers a useful clue. If inflows to Malaysia continue under the same definitions and time periods, and if the ratio compared to Taiwan proves not to be a temporary fluctuation, then the shift in logistics can be evaluated with greater confidence. Only once Intel's processing capacity, customer qualifications, and shipment results align with this data will it become possible to connect this $1.3 billion figure to an actual shift in business share.