Samsung and SK Hynix plan 590 billion dollar chip expansion
South Korea backs new fabs and packaging as AI demand lifts memory prices, scarce high-bandwidth memory turns consumer electronics into a downstream cost centre
Images
Samsung and SK Hynix plan $590 billion chip investment as AI demand sends memory prices soaring
the-decoder.com
South Korea’s two biggest chipmakers are planning an investment program worth $590 billion as demand from AI data centres pushes memory prices higher, according to The Decoder citing South Korea’s Ministry of Trade, Industry and Energy. Samsung and SK Hynix would expand production capacity and related facilities in South Korea, with a large share of the spending earmarked for new factories and a separate allocation for packaging and next-generation chip development.
The numbers land in the middle of a global scramble for high-bandwidth memory, the specialised chips that sit next to AI accelerators and feed them data fast enough to keep expensive compute from idling. The Decoder reports that Samsung and SK Hynix together control nearly 80 percent of the global market for high-bandwidth memory chips, which makes their domestic buildout less a normal capex cycle than a supply-chain decision that ripples through every company renting or building AI capacity.
Price signals are already doing much of the policy work. Jefferies Equity Research projects memory chip prices could rise sharply through late 2026 and into 2027, and The Decoder links those increases to higher costs in consumer electronics. When components that used to be a rounding error become a constraint, the bargaining power shifts from device brands and cloud customers toward the handful of suppliers that can deliver at scale.
The plan also underlines how “AI” spending keeps turning into construction and geography. The Decoder says hundreds of trillions of won are allocated for building four new factories in the country’s southwest region, with additional spending for a packaging centre and longer-term funding for next-generation chips. Governments can announce national AI strategies, but the operational bottlenecks still show up as permits, grids, water, trained technicians, and a limited number of sites where fabs can be expanded without breaking logistics.
For the rest of the world, the risk is not just higher prices but tighter coupling: a larger share of critical AI hardware capacity concentrated in one country, supported by state policy, and timed around when new production can realistically come online. The Decoder notes that new capacity is expected later in the decade, implying that the near-term market remains defined by scarcity.
In the meantime, the market is pricing memory like a strategic asset while the factories that could relieve it are still drawings, land, and budget lines.