--- title: Samsung’s 2028 line is a lead-time statement, not a demand forecast url: https://www.consultingai.dev/blog/jukan-tight-supply-2028/ original: https://www.consultingai.dev/blog/jukan-tight-supply-2028/ --- # Samsung’s 2028 line is a lead-time statement, not a demand forecast By [Shreenivas Sangoram](/) · 29 Oct 2026 Short answer Jukan is right to lift this line, and it reads better when you put it next to the capex number from the same call. Samsung is guiding 2026 capital spending to KRW 84.4 trillion and still says 2028 is tight, because the money buys cleanroom shells and tools on 12-to-18 month lead times before it buys output. The sentence is not a demand forecast. It is a statement about how long it takes to build a fab in 2026 — and about HBM4E and HBM5 dies that are bigger, with more metal layers, so each wafer yields fewer bits. SAMSUNG ELECTRONICS SAID, “WE EXPECT THESE TIGHT MEMORY SUPPLY CONDITIONS TO PERSIST INTO 2028 AS WELL.” [@jukan05 on X](https://x.com/jukan05/status/2115622336397443263) ## The capex that does not produce bits Samsung’s own framing, from the live call, is unusually mechanical: asked about 2028, it said AI infrastructure investment and personal AI agents sustain demand, that capex rises sharply in 2027, but that higher fab construction costs and long lead times limit its contribution to 2028 bit production; rising equipment costs and longer delivery times further restrict capacity growth, while larger dies and more metal layers in HBM4E and HBM5 constrain physical DRAM supply ([@antfeedapp](https://x.com/antfeedapp/status/2115621094749352263)). That is a company explaining, in its own words, why its capital budget is not a supply curve. The number underneath it: KRW 84.4 trillion for 2026, about $63.2 billion and up KRW 31.7 trillion year on year, with KRW 56.9 trillion spent through Q3 and KRW 52.8 trillion of that in semiconductors ([@antfeedapp](https://x.com/antfeedapp/status/2115620305838489861)). Spending at that rate while telling customers the shortage lasts two more years is only contradictory if you assume fabs arrive when the money does. ## The quarter behind the line Samsung printed KRW 107.4 trillion of operating profit — its first quarter above KRW 100 trillion — with the DS division earning the overwhelming majority of it at a margin near 72%, on DRAM pricing and HBM4 for Vera Rubin. DRAM ASP rose about 20% quarter on quarter and NAND rose by a high-teens percentage ([@NEWS2082680](https://x.com/NEWS2082680/status/2115664701573181950); [@antfeedapp](https://x.com/antfeedapp/status/2115616506080112805)). The memory shortage paying Samsung’s chip unit is the same cost squeezing its own handset business, which posted a second straight quarterly loss and expects component costs to stay a headwind into Q4. That split is worth holding on to. The shortage is not a macro abstraction: it is a transfer inside the same company, from devices to silicon, and Samsung is not going to solve it by discounting to its own phone division. ## The supply side is not standing still SK hynix reported the day before: revenue of about KRW 104.6 trillion, operating profit around KRW 83.3 trillion at roughly an 80% margin, DRAM and HBM effectively sold out through 2029, HBM requests at more than twice what it can supply, and a structural shortfall through 2030 ([@JSCC2020Lee](https://x.com/JSCC2020Lee/status/2115657597777047636); the company release is at news.skhynix.com). Nomura kept Buy ratings on both Korean names and expects the shortage to run through 2028, tracking 22.3 GW of new data-centre capacity this year and 31.5 GW next year, and modelling the combined net cash of nine semiconductor firms rising from $282B in Q2 to about $2 trillion by end-2028. Then the physical calendar, which is the only part that cannot be talked around: Micron’s ID1 starts wafer output in mid-2027, with a second Idaho fab, Hiroshima expansion and a Singapore NAND fab following in 2H28; SK hynix opens Yongin Phase 1 cleanrooms in early 2027 and says 2027 capex rises significantly; Samsung says P5 runs at full scale from 2028; CXMT is reported to be targeting 500k wafers a month by end-2028, concentrated in commodity DRAM ([@Schulz_Research](https://x.com/Schulz_Research/status/2115500137976074556)). Add equipment lead times to construction lead times and you get exactly what Samsung said: tight in 2028. ## What would falsify it Four things, all observable: equipment spending running ahead of construction in 2027; spot DDR4 and DDR5 falling while contract prices hold; contracts reopened or deposits handed back early; and a new CXMT fab appearing in a filing. The first crack, if there is one, shows in commodity DRAM and NAND — TrendForce already has NAND loosening in 2H27 — not in HBM, where 8-Hi and 12-Hi stacks are booked out and the packaging queue is longer than the memory queue. The asymmetry worth naming: memory suppliers can be short through 2028 and still have a great 2027, because the pricing is set by contracts signed now. Buyers cannot hedge the same way. For a lab or a neocloud, the practical conclusion is that contracted volume through 2028 is the only reliable route to hardware, and that waiting for spot to soften is a bet against a construction schedule that cannot be accelerated by wishing. ## What we could not verify The 2028 capex figure Samsung will actually spend has not been guided; the KRW 84.4 trillion is 2026. Nomura’s data-centre gigawatt numbers and net-cash projection are a bank’s model, not a supply calendar. The CXMT wafer target is press reporting, not a company disclosure. ## Sources - [@jukan05, 29 Oct 2026 01:50 UTC — Samsung: tight memory supply into 2028](https://x.com/jukan05/status/2115622336397443263) - [@antfeedapp — live transcript on the 2028 outlook, construction and equipment lead times](https://x.com/antfeedapp/status/2115621094749352263) - [@antfeedapp — 2026 capex KRW 84.4T, KRW 56.9T spent through Q3](https://x.com/antfeedapp/status/2115620305838489861) - [@antfeedapp — DRAM ASP +20% QoQ, NAND high-teens](https://x.com/antfeedapp/status/2115616506080112805) - [@NEWS2082680 — KRW 107.4T operating profit, DS margin ~72%, 10 LTA clients](https://x.com/NEWS2082680/status/2115664701573181950) - [@JSCC2020Lee — SK hynix 3Q26 numbers, sold out through 2029, 14 LTA customers](https://x.com/JSCC2020Lee/status/2115657597777047636) - [@Schulz_Research — the dated bear case: ID1 mid-2027, three sites in 2H28, Yongin P1, P5, CXMT](https://x.com/Schulz_Research/status/2115500137976074556) - [Samsung Electronics — Q3 2026 results](https://news.samsung.com/global/samsung-electronics-announces-third-quarter-2026-results) - [SK hynix — Q3 2026 business results](https://news.skhynix.com/en/q3-2026-business-results/) - [TrendForce — Samsung Q3: HBM4 sales triple, DS margin 72%, 10 LTA clients](https://www.trendforce.com/news/2026/10/29/news-samsung-says-hbm-share-matches-dram-standing-as-hbm4-sales-triple-ds-margin-hits-72-10-lta-clients-signed/) ## FAQ ### Does “tight into 2028” mean prices keep rising until then? Not necessarily. Contract prices can plateau while volumes stay allocated. Tightness is about availability first, and about price second. ### Is capex guidance a signal of confidence or of desperation? Neither. At these margins it is a signal that depreciation is cheap relative to the price of a gigabit. The question is timing, not willingness. ### Who loses from a two-year shortage? Consumer device makers, price-sensitive cloud buyers and anyone whose product was designed with a 2025 memory bill in mind. Suppliers and their tool vendors capture the difference. ### When does the picture change? Watch 2H28 for the capacity additions and 2H27 for NAND. HBM is the last thing to loosen, not the first. Share Post on X LinkedIn Building something like this? I help teams ship agents, RAG and model evaluations that hold up in production. Work with me → More in AI news AI news · 29 Oct 2026 Between two Samsung lines, the eyes post: the shortage is now a payout story Jukan's wordless 👀👀 post sat between Samsung's 2027 shortage line and its buyback remarks: what repriced that morning was the KRW 55–75 trillion return plan. AI news · 29 Oct 2026 Samsung’s HBM4 sold at “target price” is the memory bill for Rubin Mid-to-high $4 per gigabit is about $36–40 per GB, or $9,000–10,500 of HBM4 in one 288 GB Rubin GPU. Sold out at target price makes the shortage contractual. AI news · 29 Oct 2026 Powertech’s $400M panel fab is where Broadcom’s ASIC volume goes when CoWoS is full Taiwan cleared Powertech's US$400M Singapore panel-level packaging fab on 28 October: the second queue for AI packages when TSMC CoWoS is full, with Broadcom ASIC volume.