Samsung and SK Hynix promise to increase memory‑production spending, but that won’t solve the short‑term shortage of memory.

Brief on the Memory and AI Situation

A technology giant, TSMC, believes that its limited chip‑fabrication capacity is slowing the growth of the artificial intelligence sector. However, one must not overlook the critical shortage of memory—an essential component for AI applications. Currently, the major players in the memory market, Samsung and SK Hynix, have announced plans to increase capital expenditures, but their efforts will only materialize several years from now.

1. Key Facts

- TSMC points to chip‑production limits as the main bottleneck for AI development.
- Samsung & SK Hynix both report a memory shortage and an inability to rapidly scale production.
- By 2025, SK Hynix has surpassed Samsung in operating profit overall.
- SK Hynix remains the largest supplier of high‑performance memory (HBM) and intends to maintain that position.
- Memory prices are rising outside the AI segment, boosting manufacturers’ revenues.

2. Samsung’s Plans

- Kim Je‑jun, Executive Vice‑President for Memory, announced an intention to “significantly increase” capital spending on memory production this year.
- Budget: Last year’s spend was $33 million. In 2026 a hike is planned, but the exact figure has not yet been disclosed.
- Implementation Timeline: The investments will translate into operational capacity only in a few years. In the short term Samsung will focus on optimizing existing sites.

3. SK Hynix’s Plans

- Focus on HBM: The company maintains leadership in high‑performance memory and does not intend to cede ground to competitors.
- Capital Expenditure: Aims to keep the capex-to-revenue ratio around 35 %. As revenue grows, expenses will rise proportionally.
- Demand vs. Supply: President Son Hyun‑jeong noted that demand is surging sharply, but expanding production takes time, so the shortage fuels price inflation.

4. Why Memory Producers Are Cautious

1. Market Cyclicality – When supply exceeds demand, prices fall and companies may operate at a loss for several quarters.
2. Scaling Time – Building new fabs takes years; investors expect returns only after a long period.
3. “One‑Year Order” Cycle – Citigroup observes that customers now sign contracts at least one year before delivery, increasing pressure on manufacturers.

Conclusion

TSMC believes its own manufacturing constraints are hindering AI progress, but the memory shortage from Samsung and SK Hynix is also a serious barrier. Both firms have announced plans to raise capital expenditures, yet real new capacity will appear only in several years. In the meantime they will optimize existing sites, while rising memory prices continue to stoke demand and exacerbate shortages.