Where and how to add AI-grade package substrate capacity for the 2028 demand window

Reconstructs a component manufacturer’s choice of how to expand FC-BGA substrate capacity for AI accelerators ahead of a forecast 2027 to 2028 shortage. It compares domestic, overseas, phased and glas...

Published content

1. node: It is assumed that the customer funding and volume guarantees are contractually binding and cover a material part of the capital cost and output, although neither the customers nor the terms have been disclosed.
2. node: It is assumed that the Southeast Asian site could absorb a second, larger block of high-end capacity in addition to the tranche already approved, in terms of land, utilities, skilled staff and engineering support.
3. node: It is assumed that customers would still place long-term business with a supplier offering phased, uncommitted capacity rather than a capacity block reserved for them.
4. node: It is assumed that glass-core substrates reach commercial readiness and customer adoption soon enough to replace the demand the company would forgo in organic FC-BGA over 2027 to 2029.
5. node: The company is a large South Korean electronic-components manufacturer and a later entrant to the market for high-end FC-BGA substrates, which connect AI accelerators, GPUs and server CPUs to the circuit board. It already produces package substrates at domestic sites and at a Southeast Asian subsidiary. It is deciding now because AI-infrastructure demand has run ahead of qualified substrate capacity, anchor customers are offering funding and volume commitments, and capacity approved today cannot produce before 2028.
6. node: How should the company add capacity for high-end flip-chip ball grid array (FC-BGA) substrates for AI accelerators and servers ahead of the forecast 2027 to 2028 shortage: a large customer-backed expansion at its existing domestic site, the same capacity placed at its Southeast Asian subsidiary, a smaller phased domestic expansion without customer commitments, or a shift of capital toward next-generation glass-core substrates?
7. node: The company's board of directors takes this decision, as a capital commitment large enough to require a regulatory disclosure, on the advice of the chief executive and the management of the package-substrate business.
8. node: The choice covers where and on what terms to add the next large block of high-end FC-BGA capacity, and whether to prioritise the current organic substrate technology or glass-core substrates. It does not cover the separate equipment tranche for the Southeast Asian subsidiary approved on the same day, the company's other product lines, or the pricing of customer contracts, which are settled separately. Public reports differ on whether the domestic project is a new building or an expansion within an existing plant; the log treats it as capacity added at an existing domestic site.
9. node: Phasing keeps the company's balance sheet lighter and lets it respond to evidence on competitor capacity and AI-server demand before committing later phases.
10. node: A July 2026 market-research report describes a planned glass-core joint venture with about KRW 480 billion in capital, a 66 percent stake for the company and full-scale operations from the second half of 2027. It reports measured gains in flatness, thermal expansion and stiffness for glass cores, but notes that key manufacturing hurdles remain and that commercial-scale adoption is expected after 2030.
11. node: Reporting on the regulatory disclosure states an investment of KRW 4.27 trillion in the domestic site from September 2026 to May 2028, with mass production from September 2028, described by the company as its largest-ever investment in a single product and reported as 43.6 percent of 2025 shareholders' equity.
12. node: The subsidiary has been expanded for FC-BGA production since at least April 2026, when trade reporting described a KRW 1.8 trillion investment there to serve AI-chip customers, so the company has recent experience of building and ramping capacity at that site.
13. node: The anchor customers, described publicly only as global technology companies, were consulted on the size, timing and funding of the capacity through the mid- to long-term cooperation agreements on which the investment relies.
14. node: Execute binding customer funding and volume agreements before placing the main equipment orders, because the risk case for the investment rests on those commitments.
15. node: New capacity only earns the shortage premium if it is qualified by customers and shipping in volume while the forecast supply gap is open, which on published forecasts means during 2027 and 2028.
16. node: The company states that customers will provide part of the funding for the new facilities and guarantee mid- to long-term volumes. Korean technology press reports that this is intended to offset capital-expenditure risk and secure plant utilisation.
17. node: On the same day as the domestic decision, the company disclosed a separate KRW 2.51 trillion equipment investment for its Southeast Asian subsidiary running to April 2028, bringing the combined announcement to KRW 6.78 trillion.
18. node: Declining customer co-funding leaves the full cost with the company, and a later entrant offering only phased capacity risks losing the design-in positions that anchor customers are allocating now, which are hard to regain in later product generations.
19. node: Holding back organic capacity during the widest part of the forecast shortage would cede share to incumbents that are expanding, and the glass-core roadmap already has its own funded joint venture, so this option adds little that the parallel programme does not already provide.
20. node: Start customer qualification of the target products in parallel with construction, so that the September 2028 start of mass production is not followed by a long qualification gap.
21. node: Any single commitment must remain fundable alongside the company's wider capital programme without weakening its balance sheet to the point where a demand downturn would force asset sales or cuts elsewhere.
22. node: Spending is committed under 2026 market conditions while output arrives into unknown 2028 conditions. Analyst commentary notes the roughly 24-month gap reflects cleanroom construction, tool procurement and qualification cycles, leaving little room for mid-course correction.
23. node: Placing both new tranches at one overseas site concentrates operational, logistics and geopolitical exposure precisely where customers are asking for supply-chain stability.
24. node: Coordinate the domestic build with the Southeast Asian equipment tranche and the glass-core joint venture, so that product allocation between sites is deliberate and the domestic facility can later host next-generation substrate processes.
25. node: AI-chip customers select substrate suppliers on assured availability as well as technical performance, so any option must be credible to customers as a stable source of supply over several product generations.
26. node: It is inferred that lower labour and operating costs at the Southeast Asian site would improve unit margins over the plant's life; the company's disclosures do not quantify the difference.
27. node: Capacity arriving in September 2028 lands late in the forecast shortage window, when competitors' announced expansions are also coming on stream; if the gap closes faster than forecast, prices and utilisation fall and the company carries a very large fixed-cost base.
28. node: It is assumed that demand for large, high-layer-count substrates for AI accelerators and servers remains strong at least through 2028 to 2030. The company has publicly stated that it expects AI-infrastructure investment to continue for another two to three years.
29. node: If updated market forecasts show the 2028 supply gap narrowing to single digits as competitors' expansions come on stream, reassess the pace of equipment installation, because output arriving in late 2028 would then face falling prices.
30. node: Published market forecasts summarised by an industry analysis site put the gap between demand and supply of advanced build-up-film substrates at 20 to 40 percent, with one scenario of about 10 percent in the second half of 2026, about 21 percent in 2027 and up to 42 percent in 2028. The same analysis stresses that these are forward-looking institutional estimates, not audited counts.
31. node: If an anchor customer reduces its committed volumes or moves its next accelerator generation to glass-core substrates before 2030, reassess the allocation of the domestic capacity, because the demand certainty behind the choice would no longer hold.
32. node: The same analysis states that the scarce resource is not raw build-up film but capacity that has passed material, process, packaging and customer qualification. New lines go through repeated yield ramps and reliability tests, and the customer-qualification clock can restart. It cites a substrate project in China that was impaired because mass-production progress and returns fell short of expectations.
33. node: If customer qualification of the new lines slips beyond mid-2029, reassess the choice and the product mix, because the capacity would then miss most of the forecast shortage window.
34. node: In April 2026 the company's chief executive stated that customer demand exceeded its current production capacity by more than 50 percent. Trade reporting at the time put the combined share of the leading Japanese and Taiwanese substrate makers above 70 percent of the global market.
35. node: Competitors are also expanding: a leading Taiwanese substrate maker targets a 30 to 40 percent increase in capacity by 2027 with new plants from late 2027 and 2028, and a leading Japanese maker targets 2.5 times capacity at one plant by 2028, according to a September 2026 market report.
36. node: The company's chief executive said on announcing the investment that high-value semiconductor substrates are becoming key components that determine both chip performance and supply-chain stability as the AI and high-performance computing markets grow.
37. node: Customer-backed large expansion at the existing domestic site
38. node: Equivalent expansion at the Southeast Asian subsidiary
39. node: Phased smaller domestic expansion without customer commitments
40. node: Capital shifted toward glass-core substrates
41. node: Adopt the customer-backed large expansion at the existing domestic site, because it is the only option that combines full capacity in the 2028 shortage window, established technology, and customer funding and volume guarantees that reduce the risk of a commitment worth about 44 percent of equity, while giving customers a second production base. Expansion at the Southeast Asian subsidiary is not preferred because it would concentrate both new tranches at one overseas site, and whether anchor customers would back it on the same terms is not known. A phased uncommitted expansion is not preferred because it forgoes customer co-funding and risks losing design-in positions being allocated now. Shifting capital toward glass cores is not preferred because that technology is not expected at commercial scale before 2030 and already has its own joint venture.