AnswersPublished21 sources
Why the AI Memory-Chip Shortage Could Last Through 2030
Meritz Securities reportedly expects memory makers to fulfill only about 80% of 2027 orders; SK hynix says demand could outstrip supply through 2030. Samsung is increasing investment, while Samsung and SK hynix report tight or sold out HBM allocations.
Published byEdited with GPT-6 LunaImages generated with GPT Image 2
Research answer
Key takeaways
- Meritz Securities reportedly expects memory makers to fulfill only about 80% of 2027 orders; SK hynix says demand could outstrip supply through 2030.
- Samsung is increasing investment, while Samsung and SK hynix report tight or sold out HBM allocations.
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: How is AI-driven demand creating a prolonged memory-chip shortage, and how are Samsung, SK hynix, OpenAI, and AMD responding, considering Me. Article summary: AI demand is creating a multiyear memory bottleneck: buyers are seeking capacity faster than manufacturers can add it, making supply allocation—not just accelerator performance—a constraint on AI expansion. The figures y. Topic tags: general, general web, user generated. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts with fa
AI companies are competing for memory as well as processors. Meritz Securities reportedly expects memory makers to fulfill about 80% of 2027 orders, while SK hynix has forecast that demand will continue to exceed supply through 2030. These outlooks point to a prolonged squeeze, but they are forecasts—not a promise that every buyer will receive the same share of its order.9
60
Key takeaways
- Meritz’s reported fulfillment estimate falls from 80%–85% in 2026 to about 80% in 2027, while orders for 2027–2028 are said to exceed production capacity.
9
- Building capacity takes time, and SK hynix cites both facility construction and the growing complexity of advanced memory production as reasons the supply-demand gap may persist.
60
- Samsung is raising investment; OpenAI has met with Samsung and SK leaders amid efforts to secure memory; AMD says it is working with customers as memory and component costs rise.
18
20
35
What the 80% order-fulfillment estimate means
Meritz’s reported estimate compares available supply with order volume. If suppliers fulfill about 80% of orders in 2027, that indicates demand is running ahead of the capacity available to meet it. It does not mean every customer will receive exactly 80% of what it requested: allocations can differ, and the estimate is a forecast rather than a reported final production result.9
The same reporting says memory orders for 2027–2028 already exceed production capacity, and that possible delays to some U.S. data-center projects would not necessarily change the broader imbalance.9 Separately, SK hynix has said that demand exceeding supply could continue through 2030, pointing to facility lead times and the complexity of advanced processes.
60 These are company and analyst outlooks, not a certain timeline for when supply will normalize.
Why AI demand is difficult to meet quickly
High-bandwidth memory, or HBM, is used alongside AI accelerators. As demand for AI infrastructure grows, competition for this specialized memory adds pressure to manufacturers’ supply plans.54 Expanding output takes more than announcing a new facility: construction and production ramp-up take time, while advanced processes add technical complexity, according to SK hynix’s explanation of the supply outlook.
60
Price forecasts reflect the same pressure. TrendForce forecasts that HBM’s blended average selling price will rise 121% in 2027 compared with 2026. The reported forecast also reflects a greater share of higher-priced HBM4 products, so it should not be read as a like-for-like price increase for every HBM product.49
How Samsung and SK hynix are responding
Samsung plans record facilities spending of 84.4 trillion won, including about 77 trillion won for its semiconductor business, according to reporting on its investment plans.20 That scale of spending signals an effort to expand capacity, but new investment does not immediately become usable chip supply: SK hynix has pointed to the time required to build facilities and the complexity of advanced production.
60
The two suppliers also describe tight future allocations. Reporting says Samsung’s 2027 HBM supply is sold out, while SK hynix’s production through 2029 is effectively booked. SK hynix has also said customer requests for HBM exceed twice its available supply.60
62 These reports show strong competition for planned production, but “sold out” or “booked” does not by itself explain contract terms, product mix or the amount ultimately delivered.
OpenAI is seeking supply, but meetings are not contracts
OpenAI CEO Sam Altman met about 50 Samsung executives in Yongin and separately met SK Group Chair Chey Tae-won, with SK hynix executives also reported to be present.18 Reports describe the discussions against a backdrop of tight HBM supply and OpenAI’s interest in future memory products.
9
The public reporting provided here does not establish that those meetings produced a new supply agreement. Samsung did not announce one following the meeting, according to one report.52 The meetings show how important supplier relationships have become for AI companies; they do not, on their own, add production capacity or guarantee an allocation.
AMD is growing while managing higher component costs
AMD’s third-quarter 2026 filing reports revenue of $13.4 billion, up 45% year over year, and data-center revenue of $8.546 billion.33 The results illustrate the growth of its data-center business, but they do not remove the memory constraint facing the broader supply chain.
AMD has said it has long-term agreements with some suppliers and customers and is working with customers as memory and other component prices rise.35
44 That approach may help manage higher costs across commercial relationships; it does not resolve the underlying shortage.
What to watch next
The core issue is timing: AI-related orders are arriving faster than new memory capacity can be brought online. Meritz’s roughly 80% fulfillment estimate for 2027, Samsung’s investment plans and SK hynix’s forecast of tight supply through 2030 all point to continued competition for allocations—but none guarantees a uniform shortage through that date.9
20
60
For buyers, the practical questions are whether suppliers expand qualified production, when additional output becomes available, and how prices and delivery schedules change. For now, supplier bookings and investment plans are useful signals of pressure, not proof of when the market will rebalance.
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "Why the AI Memory-Chip Shortage Could Last Through 2030"?
Meritz Securities reportedly expects memory makers to fulfill only about 80% of 2027 orders; SK hynix says demand could outstrip supply through 2030.
What are the key points to validate first?
Meritz Securities reportedly expects memory makers to fulfill only about 80% of 2027 orders; SK hynix says demand could outstrip supply through 2030. Samsung is increasing investment, while Samsung and SK hynix report tight or sold out HBM allocations.
Sources