Five minutes. The number that mattered, where vendor share stands, what the HBM supply squeeze actually means, what to watch this month. All sourced. Nothing fabricated.
The memory market had its largest quarter on record in 4Q25 — $53.58 billion in DRAM revenue alone, up +29.4% quarter-over-quarter. To put that in context: two years earlier, in 1Q24, the full industry did $18.35B. The market has nearly tripled in eight quarters.
The driver is not mysterious. Conventional DRAM contract prices spiked roughly +90% quarter-over-quarter in 1Q26 (per TrendForce's February 2026 press release, marked as a record across PC, server, and mobile DRAM). That spike alone — stacked on the HBM AI premium — is why the revenue numbers look like they do.
Source: TrendForce free press releases (4Q25 quarterly revenue, Feb 2026; 1Q26 price direction, Feb 2026). Price figures are directional forecasts/estimates, not audited actuals. The exact $/Gb DXI index is licensed and not reproduced here.
The headline from 4Q25: Samsung is back at #1, ending a three-quarter run by SK hynix that began in 1Q25 — the first time in recent history that SK hynix had held the top DRAM revenue rank.
| Quarter | Industry total | Samsung | SK hynix | Micron |
|---|---|---|---|---|
| 1Q 2025 | $27.01B | $9.10B / 33.7% | $9.72B / 36.0% | $6.58B / 24.4% |
| 2Q 2025 | $31.63B | $10.35B / 32.7% | $12.23B / 38.7% | $6.95B / 22.0% |
| 3Q 2025 | $41.4B | $13.50B / 32.6% | $13.75B / 33.2% | $10.65B / 25.7% |
| 4Q 2025 ↑ | $53.58B | $19.30B / 36.0% | $17.22B / 32.1% | $11.98B / 22.4% |
Source: TrendForce free press releases. ↑ marks the rank-flip quarter. Full 8-quarter table (1Q24–4Q25): dram-market-share.html
Three things worth noting in this table. First, Samsung's 4Q25 comeback was powered partly by conventional DRAM — the company's own earnings call (paraphrased in English coverage) noted that conventional DRAM was delivering higher profitability than HBM that quarter, because conventional DRAM prices are reset quarterly while HBM is locked on annual contracts. A price spike of this magnitude favors whoever has the most conventional exposure. Second, Micron's 3Q25 high watermark of 25.7% share was its best in the dataset — a genuine HBM-driven gain, not a fluke. Third, 1Q26 quarterly share data is not yet available in TrendForce's free press releases as of this writing; the next PR will be the one to watch.
The "HBM is sold out" narrative is real but requires some unpacking for investors.
What is confirmed: SK hynix has stated publicly that HBM customer demand over the next three years exceeds production capacity. Micron said on its FQ2’26 call (March 18, 2026) that it can fulfill only 50–67% of its key customers’ near-term HBM demand. Micron's HBM calendar 2026 supply is entirely committed. These are real, sourced supply constraints.
What is more nuanced: HBM ASPs are locked on annual contracts, not reset quarterly. So when conventional DRAM prices spiked +90% in 1Q26, HBM holders actually underperformed on a near-term margin basis relative to conventional DRAM — a data point Samsung's management flagged explicitly. This does not mean HBM is unattractive; it means HBM pricing is smoother (less volatile, more predictable) than the spot-like conventional market.
Supply roadmap: SK hynix currently supplies more than half of NVIDIA’s 2026 HBM (per industry reporting). Samsung began HBM4 mass production in February 2026 and is targeting full-scale supply by approximately June 2026. Micron began HBM4 volume shipments (36GB 12H, over 11 Gbps) in calendar 1Q26, designed for NVIDIA’s Vera Rubin platform. HBM3E remains the volume-shipping generation into 2026.
Capacity absorption context: One HBM bit requires roughly 3× the wafer capacity of a DDR5 bit. HBM absorbed approximately 19% of total DRAM wafer output in 2025 and is projected to reach ~23% in 2026 (TrendForce, November 2025). This is the structural reason conventional DRAM has tightened: HBM is eating the fab.
Sources: SK hynix 1Q26 earnings (Apr 23, 2026) via Korea Herald/CNBC [rendered, not original-language verbatim]; Micron FQ2’26 prepared remarks and Q&A (Mar 18, 2026) [VERBATIM, official IR PDF + Motley Fool]; TrendForce Nov 13, 2025 free PR (HBM wafer share); Samsung 1Q26 call paraphrase via Wccftech/TrendForce.
"AI hasn’t just increased demand for memory — it has fundamentally recast memory as a defining strategic asset in the AI era." — Sanjay Mehrotra, Chairman, President & CEO, Micron — FQ2’26 earnings call, March 18, 2026 [VERBATIM, official IR PDF]
This line is worth sitting with. Mehrotra is making a structural argument, not a cyclical one. The conventional way to think about memory is as a commodity that swings with capex cycles — oversupply, price crash, constrained investment, recovery. What he’s arguing is that AI has altered that dynamic: memory capacity and bandwidth are now a first-order competitive constraint for AI infrastructure, which gives memory makers pricing power and strategic leverage they have not historically had. The data so far — +90% DRAM contract price spike in 1Q26, HBM sold out years forward, Micron capex raised from ~$20B to >$25B in one quarter — supports that framing.
The bear case, worth keeping in mind: AI capex buildouts have historically run in waves, and memory is famously the first thing cut when hyperscalers pull back. The structural argument is credible; the question is whether this cycle is longer and deeper than past ones. That remains an open bet.
On May 27, 2026, CXMT (ChangXin Memory Technologies) cleared the Shanghai STAR Market IPO review committee — the largest mainland China IPO since 2022. The offering targets at least RMB 29.5 billion (~$4.3B, potentially topping $5B), with proceeds earmarked for wafer capacity expansion, DRAM technology development, and HBM R&D.
The market context: CXMT held approximately 3.97% of global DRAM revenue share in 2Q25 and grew to approximately 7.67% by 4Q25 (TrendForce estimates — directional). It is now the world’s fourth-largest DRAM maker. In 1Q26, CXMT reported net profit growth of +1,688% year-over-year.
Why it matters for a memory investor: this IPO, if it closes, gives CXMT the capital to accelerate capacity at a moment when the rest of the industry is deliberately constraining supply. Whether that additional Chinese capacity hits global markets or stays within China’s domestic technology ecosystem is the key variable. If it bleeds into the global conventional DRAM market, it is a price-suppression risk. If it stays domestic (or is absorbed by China’s AI buildout), the global tightness persists. Watch this space — the allocation question will take months to clarify.
Source: Reuters/Caixin/Bloomberg coverage of the CXMT STAR Market review, May 27, 2026; TrendForce free press releases for CXMT share estimates (directional). See also: What is CXMT?