Technology · Semiconductors
The Memory Supercycle: the most profitable companies in the world in 2027
Contents
- Executive Summary
- 1. Why This Cycle is Structurally Different
- 1.1 HBM as an Architectural Bottleneck
- 1.2 Long-Term Supply Agreements (LTA)
- 1.3 New Capacity Won't Arrive Until 2028
- 2. Q1 2026 Results — The Acceleration in Numbers
- 2.1 Samsung Electronics (KRX: 005930) — Reported April 30, 2026
- 2.2 SK Hynix (KRX: 000660) — Reported April 22, 2026
- 3. 2027 Projections — World Earnings Ranking
- 4. Price Dynamics: TrendForce and SemiAnalysis
- 5. The Valuation Anomaly
- 6. Investment Vehicles
- 6.1 Roundhill Memory ETF (DRAM) — NYSE
- 6.2 Micron Technology (NASDAQ: MU) — Direct Access
- 7. Main Risks
- 8. Conclusion
- Disclaimer
Miguel Braun · June 2026 · Special Investment Report
Samsung Electronics · SK Hynix · Micron Technology
Executive Summary
Three semiconductor companies —Samsung Electronics, SK Hynix and Micron Technology— are projected to break into the select group of the 10 most profitable companies in the world in 2027, driven by a structural shortage of high-bandwidth memory (HBM) that is the fundamental bottleneck of global Artificial Intelligence infrastructure.
The fundamental data is compelling:
- Samsung (KRX: 005930): ~$350B projected net profit 2027E · forward P/E ~5-6x · No company in the world will make more money than Samsung in 2027. Nvidia would take second place.
- SK Hynix (KRX: 000660): ~$270B net profit 2027E · forward P/E ~5-7x · only Samsung and Nvidia will earn more.
- Micron Technology (NASDAQ: MU): ~$133-145B net profit 2027E · forward P/E ~10-13x · 6 companies will earn more.
To put it in perspective: Apple, Microsoft, Alphabet, Meta and Amazon all trade at forward P/E of 17x to 30x with margins of 27-38%. The three memory companies project margins of 35-77% and trade at a third of that multiple.
1. Why This Cycle is Structurally Different
Memory chip companies have historically been one of the most cyclical sectors of the economy. The key question is: what makes this time different?
1.1 HBM as an Architectural Bottleneck
High-bandwidth memory (HBM) is not a fungible commodity — it is a critical and irreplaceable component in every AI chip from Nvidia, AMD and the custom ASICs of Google, Meta and Amazon. Without HBM, AI chips can't function.
- Each Nvidia Blackwell GPU requires up to 192GB of HBM3E. The successor Rubin will require even more.
- HBM production consumes 23% of total DRAM wafers — capacity that is no longer available for conventional memory.
- The three HBM suppliers are sold out until the end of 2027 and are formalizing long-term sales agreements.
- Michael Dell forecasts that memory usage from 2022 to 2028 will increase 600x. Elon Musk says that if he bought all the chips in the world, it would still only cover 2% of the amount he needs. Jensen Huang says the "AI infrastructure buildout" is just starting and has a solid 7/8 years ahead.
1.2 Long-Term Supply Agreements (LTA)
An unprecedented structural element: the hyperscalers (Google, Microsoft, Amazon, Meta) are signing multi-year contracts with guaranteed minimum-price (floor price) clauses with the memory manufacturers. This turns the manufacturers into strategic suppliers with multi-year earnings visibility and eliminates the cyclicality problem of memory chips.
1.3 New Capacity Won't Arrive Until 2028
Expanding manufacturing capacity takes between 4 and 5 years. Fabs can grow at a maximum pace of 20-30% a year.
- SK Group Chairman Chey Tae-won: "The global shortage will persist until 2030. HBM demand structurally exceeds capacity by more than 20%."
- Samsung EVP Jaejune Kim, Q1 2026: "Our demand-fulfillment rate is at an all-time low. Customers are pulling forward 2027 orders for fear of shortage."
- Dylan Patel, SemiAnalysis: "DRAM prices will double or triple from current levels — the real incremental supply doesn't arrive until 2028."
2. Q1 2026 Results — The Acceleration in Numbers
2.1 Samsung Electronics (KRX: 005930) — Reported April 30, 2026
Samsung broke all the company's historical records in Q1 2026:
| Metric | Q1 2026 | YoY change |
|---|---|---|
| Total revenue | KRW 134T (~$91B) | +69% YoY |
| Operating profit | KRW 57.2T (~$39B) | +756% YoY |
| Net profit | KRW 47.2T (~$32B) | x2.4 vs Q4 2025 |
| Operating margin | 43% | vs 21% in Q4 2025 |
| EPS (common share) | KRW 7,123 | All-time record |
| Memory revenue | KRW 74.8T (~$51B) | 2nd consecutive record |
Q1 2026 operating profit exceeds Samsung's entire fiscal-year 2025 result. Q2 2026 is reported on July 23.
2.2 SK Hynix (KRX: 000660) — Reported April 22, 2026
SK Hynix beat analyst estimates with a +49.7% positive surprise in EPS:
| Metric | Q1 2026 | YoY change |
|---|---|---|
| Revenue | KRW 52.6T (~$35.5B) | +198% YoY |
| Operating profit | KRW 37.6T (~$25.5B) | +405% YoY |
| Net profit | KRW 40.3T (~$27.4B) | All-time record |
| Operating margin | 72% | All-time high |
| Net margin | 77% | Highest in the global top 10 |
| EPS surprise | +49.7% vs consensus | +6.1% revenue surprise |
Q2 2026 guidance: DRAM shipments +high single digits QoQ, persistent favorable pricing environment. Q2 reports on July 29.
3. 2027 Projections — World Earnings Ranking
Consensus estimates Bloomberg / Goldman Sachs / KB Securities / UBS — June 2026:
| # | Company | Net profit 2027E | Forward P/E | Net margin | Earn more |
|---|---|---|---|---|---|
| 1 | Samsung Electronics | ~$350B | ~5–6x | ~35% | 0 companies |
| 2 | Nvidia | ~$300–340B | ~25–32x | ~56% | 1 company |
| 3 | SK Hynix | ~$270B | ~5–7x | ~77% | 2 companies |
| 4 | Alphabet | ~$190–210B | ~17–20x | ~31% | 3 companies |
| 5 | Apple | ~$160–180B | ~24–28x | ~27% | 4 companies |
| 6 | Microsoft | ~$155–175B | ~25–30x | ~38% | 5 companies |
| 7 | Micron Technology | ~$130–145B | ~10–13x | ~58% | 6 companies |
| 8 | Meta Platforms | ~$115–130B | ~18–22x | ~33% | 7 companies |
| 9 | Amazon | ~$110–125B | ~24–30x | ~17% | 8 companies |
| 10 | Saudi Aramco | ~$80–100B | ~11–14x | ~22% | 9 companies |
In gold: the three memory companies analyzed. Sources: Bloomberg, Goldman Sachs, KB Securities, UBS.
4. Price Dynamics: TrendForce and SemiAnalysis
| Product | Q1 2026 QoQ | Q2 2026 QoQ | Outlook |
|---|---|---|---|
| Conventional DRAM | +90–95% (RECORD) | +58–63% | Peak Q3–Q4 · new capacity not before late 2027 |
| PC DRAM (DDR5) | >+100% (ABS. RECORD) | +58–63% | Seller-dominated market; OEMs with declining inventory |
| HBM (AI) | Sold out 100% 2026 | LTAs with floor price | 2027 ASP keeps rising. Server DDR5 surpasses HBM profitability for the first time in history |
| NAND / Enterprise SSD | +55–60% | +70–75% (accelerating) | CSPs stockpiling enterprise SSD for inference workloads |
| LPDDR5 | +3x since Q1 2025 | >$10/GB spot | SemiAnalysis: 2027 repricing still NOT reflected in Wall St. estimates |
"DRAM prices are going to double or triple from here still, because that's how much capacity is needed. The real incremental supply doesn't arrive until 2028." — Dylan Patel, SemiAnalysis (Invest Like the Best EP.468)
Source: TrendForce Market Bulletin May 2026; SemiAnalysis April 2026.
5. The Valuation Anomaly
Full multiples comparison. Annualized P/E = price ÷ (last-quarter GAAP EPS × 4). Meta and Alphabet normalized excluding non-recurring items.
| Company | P/E ann.* | P/E fwd | P/B | ROE | Net margin |
|---|---|---|---|---|---|
| Samsung [Mem · Q1 2026] | ~11x | ~5–6x | ~2.5x | ~19% | ~35% (47T÷134T KRW) |
| SK Hynix [Mem · Q1 2026] | ~10x | ~5–7x | ~8x | ~61% | ~77% (40T÷52.6T KRW) |
| Micron [Mem · FQ2 2026] | ~20x | ~10–13x | ~6x | ~40% | ~58% ($13.8B÷$23.9B) |
| Nvidia [FY2026] | ~40x | ~25–32x | ~35x | ~120% | ~56% ($120B÷$216B) |
| Alphabet [norm. Q1 2026] | ~31x | ~17–20x | ~7x | ~32% | ~31% ($33.9B÷$109.9B) |
| Apple [FQ2 2026] | ~34x | ~24–28x | ~50x | ~165% | ~27% ($29.6B÷$111.2B) |
| Microsoft [Q3 FY2026] | ~25x | ~25–30x | ~12x | ~40% | ~38% ($31.8B÷$82.9B) |
| Meta [norm. Q1 2026] | ~23x | ~18–22x | ~8x | ~38% | ~33% ($18.7B÷$56.3B) |
| Amazon [Q1 2026] | ~41x | ~24–30x | ~10x | ~28% | ~17% ($30.8B÷$181.5B) |
| Saudi Aramco [FY2024] | ~14x | ~11–14x | ~4x | ~22% | ~22% ($106B÷$480B) |
* Annualized P/E = price ÷ (last-Q GAAP EPS × 4). Sources: Seeking Alpha, StockAnalysis, official reports, Investing.com.
6. Investment Vehicles
6.1 Roundhill Memory ETF (DRAM) — NYSE
- Launched: April 2, 2026 — the world's first pure memory ETF
- Current AUM: ~$13.4B (gathered in less than 2 months)
- Performance since inception: +96% as of June 1, 2026
- Expense ratio: 0.65% annual · options available
- Holdings: SK Hynix ~24% · Samsung ~25% · Micron ~24% (~73% concentrated in three positions)
- Next catalyst: SK Hynix ADR June-July 2026 · ~$10B · underwriters: Goldman Sachs, Citi, JPMorgan
6.2 Micron Technology (NASDAQ: MU) — Direct Access
- Price: ~$971 (5/29/2026) · Market cap: ~$1.1T
- FQ3 2026 guidance: revenue $33.5B ± $750M · EPS ~$19.15 ± $0.40
- UBS price target: $1,625 (revised up 3x on 5/26/2026)
- HBM capacity 100% sold out for 2026 under long-term contracts
7. Main Risks
- Historical cyclicality: memory had 80-90% collapses in previous cycles (2018, 2022). LTAs mitigate this but don't eliminate it.
- Geopolitical risk: Samsung and SK Hynix are in South Korea. Micron has sales restrictions in China.
- Post-2027 margin normalization: when new capacity arrives in 2028-2029, the 60-77% margins will compress.
- DRAM ETF concentration: 73% in three stocks — a profit warning from any of them hits hard.
- SK Hynix ADR: issuance of new shares (~2-3% dilution). If delayed, it creates short-term uncertainty.
- Timing: Samsung +189% YoY, SK Hynix +789% YoY, Micron +909% YoY — part of the supercycle is already in the price.
8. Conclusion
- The three cheapest companies in the global top 10 by 2027 earnings are also the ones with the highest margins.
- Long-term contracts with floor pricing reduce the downside risk that destroyed investments in previous cycles.
- HBM demand for AI is structural — each chip generation requires more memory, not less. AI agents are about to multiply the amount of memory needed.
- Supply can't grow more than 20-30% a year. New capacity isn't available until 2028.
- The 2027 repricing isn't yet in Wall Street estimates — there is additional uncaptured earnings upside.
Disclaimer
This report is informational and prepared exclusively for advised clients of Fox Capital. It does not constitute personalized financial advice, nor an offer or recommendation to buy or sell securities. The projections mentioned are third-party analyst consensus estimates and do not guarantee future results. Investing in capital markets involves risks, including the possible loss of the invested capital. Investors should consider their individual risk profile before making investment decisions. Past returns do not guarantee future returns.
Sources: TrendForce Market Bulletin (May 2026) · SemiAnalysis — Dylan Patel (Feb-Apr 2026) · Bloomberg consensus · Goldman Sachs Research · KB Securities · UBS Research · official reports Samsung Q1 2026 · SK Hynix Q1 2026 · Micron FQ2 2026 · Seeking Alpha · StockAnalysis.com.
Financial Advisor · Author · Columnist