DOMINO RESEARCH · CHAIN MAP

The Oxygen Supply for AI Is Running Low — and One Company Controls the Tank

Samsung's memory chip supercycle is repricing the entire semiconductor food chain, and most investors are still looking at the wrong stocks.

April 13, 20261,608 words7 min read

What to know

Every time you ask ChatGPT a question, summarize a PDF, or generate an image, a server somewhere burns through memory chips the way a car burns through gasoline. Not the processor — the memory. The fast, expensive kind that feeds data to AI chips so they don't sit idle.

Right now, the world's biggest companies are racing to build AI data centers. And they're all fighting over the same scarce resource: high-bandwidth memory chips. The company that makes more of those chips than anyone else on the planet? Samsung Electronics.

Samsung just signaled it expects record profits this quarter. Its stock has surged nearly 50% in three months. And somehow, it's still one of the cheapest large-cap tech stocks you can buy.

That combination — record earnings, massive rally, dirt-cheap valuation — doesn't happen often. When it does, the ripple effects tend to show up in places most people aren't watching.

48%stock rally in 3 months
7xP/E multiple (vs. cycle strength)
Record Q1profit guidance from Samsung

What just happened

Samsung Electronics is in the middle of a memory chip supercycle — a stretch where demand outpaces supply so much that prices keep climbing — a stretch where demand outpaces supply so much that prices keep climbing quarter after quarter. The company is targeting record Q1 2026 profit as this wave hits full stride.

The numbers tell the story. As of early April 2026, Samsung's stock had returned roughly 48% over the prior three months, making it one of the best-performing mega-caps in the world over that stretch.

Even after that rally, the stock still looks cheap by almost any measure. Memory chip shortages are stoking price hikes across the entire semiconductor industry and sparking a global chip rally.

First domino: Samsung's profit machine is running on fixed costs

Think of a memory chip factory like a movie theater. It costs roughly the same to run whether 10 people or 500 people show up. But every additional ticket sold is almost pure profit. That's how semiconductor manufacturing works — and it's why rising chip prices hit Samsung's bottom line like a multiplier.

Samsung posted record Q4 2025 profit as the AI-driven memory boom lifted earnings, then guided for another record in Q1 2026. Back-to-back record quarters signal that demand is accelerating, not just holding steady.

When a company's costs are mostly fixed and its product prices keep rising, every incremental dollar of revenue flows almost straight to profit. That's the dynamic Samsung is living right now. At roughly 7x expected earnings, the stock is priced like this cycle has an expiration date. In our view, that discount underestimates how long geopolitical constraints could keep supply tight.

Memory Chip Economics: Fixed-Cost Leverage

MetricLow Demand ScenarioSupercycle Scenario
Factory capacity utilization60%95%+
Revenue per wafer$500M$850M+
Operating margin flow-through30%70%+
Stock valuation (P/E)15x7x

Second domino: Trade restrictions are accelerating Samsung's AI chip qualification — and seeding the next bust

Imagine you run one of three pizza shops in town. The city shuts down one competitor. Your pricing power doubles overnight. Now here's the twist: you use those extra profits to build two more locations — which could flood the market later.

U.S. Moves to bar Chinese chips are widely understood as a boon to Samsung and SK Hynix. But the non-obvious effect is what those restrictions are doing to Samsung's product roadmap. With Huawei-linked memory makers sidelined, Nvidia and AMD have fewer suppliers they can approve for high-bandwidth memory (high-bandwidth memory (HBM — ultra-fast RAM stacked for AI chips) — ultra-fast RAM stacked for AI chips). That's speeding up Samsung's HBM approval timelines. Samsung is moving from backup supplier to primary source faster than the market expected.

The risk hiding inside that tailwind: Samsung is plowing supercycle profits into aggressive capacity expansion. Its Taylor, Texas fab and expanded Korean HBM lines are scheduled to ramp over the next 12–18 months. If SK Hynix and Micron follow suit — and they are — the industry could be building the oversupply that breaks this cycle, just as it did in 2018.

Geopolitics gave Samsung a wider moat. The question is whether Samsung uses that moat to over-invest, turning today's shortage into tomorrow's glut.

Editorial illustration

Third domino: The AI boom is repricing 'boring' memory chips too

When people think about the AI chip boom, they picture the flashy stuff — Nvidia GPUs, cutting-edge processors. But AI servers also need vast amounts of plain old memory, the kind that goes into everything from laptops to cars. And factories can only make so much at once.

When factories shift production toward high-margin AI memory like HBM, regular DRAM and NAND chips get scarcer. That pushes prices up across the whole industry. The repricing isn't limited to AI. It touches everything from car infotainment systems to gaming consoles.

The spread between DRAM spot prices and contract prices is where the non-obvious insight lives. Spot prices move first when supply tightens; contract prices follow with a lag. When that spread widens, it signals that the next round of contract negotiations will lock in higher prices for quarters to come — a leading indicator that most generalist investors miss.

This is the hidden tax of the AI buildout. It doesn't just affect AI companies. It reprices the entire memory supply chain.

DRAM & NAND Price Repricing Across Product Categories

High-bandwidth memory (HBM)
45%
Standard DRAM
28%
NAND flash storage
22%
Gaming console memory
18%

Repricing spreads across the entire memory supply chain as AI demand siphons capacity from consumer chips.

Fourth domino: TSMC's packaging bottleneck is Samsung's ceiling

Samsung makes the memory. TSMC makes the processors that use that memory. If you want to know whether AI chip demand is real, don't just listen to one company — check whether its neighbors are also capacity-constrained.

TSMC reported strong Q1 2026 revenue growth, and demand for its most advanced chip-making technology continues to outstrip production capacity. The company controls roughly 70% of the advanced semiconductor manufacturing market.

But here's what most coverage misses. TSMC's real bottleneck isn't making the wafers. It's the advanced packaging step — specifically its CoWoS (Chip-on-Wafer-on-Substrate) — a packaging process that bonds memory stacks onto AI processors (Chip-on-Wafer-on-Substrate) process, which bonds HBM stacks onto AI processors. Every Nvidia H100 and B200 chip needs CoWoS packaging to integrate Samsung's or SK Hynix's HBM. Until TSMC's packaging yield ramp catches up, Samsung's memory chips sit in a queue.

That makes TSMC's constraint Samsung's ceiling, not just its corroborating indicator. If you're watching Samsung's earnings, track TSMC's CoWoS capacity updates just as closely. They control how fast Samsung's highest-margin product can actually reach customers.

TSMC CoWoS Packaging Bottleneck & Capacity Milestones

Q1 2026CoWoS capacity fully allocated; Nvidia/AMD waitlists extend
H2 2026TSMC advanced packaging capacity expansion begins
Q4 2026–Q1 2027Competitive packaging capacity (Samsung, others) comes online
Editorial illustration

Fifth domino: Supercycle profits are funding Samsung's consumer AI moat

Most investors are focused on Samsung the chipmaker. But Samsung is also one of the world's largest smartphone sellers. And the chip supercycle is bankrolling a move that could trigger an upgrade wave among hundreds of millions of price-sensitive consumers.

Samsung has been moving AI features down from its premium Galaxy S and Z lines into its cheaper A-series phones. Think voice tools, image editing, productivity assistants, and hardware-based security. The company's Knox Vault privacy tools, previously reserved for flagships, are now reaching a wider global user base.

The link to the chip story is direct: record memory profits give Samsung the R&D budget to embed AI features into budget devices that competitors can't match at the same price point. Premium features in budget phones historically trigger upgrade cycles among price-sensitive buyers — and Samsung sells more A-series phones than flagships by a wide margin.

In our view, this double exposure — to the chip supercycle and a possible smartphone upgrade wave — could make Samsung's earnings steadier than companies that only sell memory chips. If chip prices eventually soften, a strong phone business provides a cushion that competitors like Micron and SK Hynix don't have.

The last time this happened

The closest parallel is the 2017–2018 memory supercycle. A surge in cloud computing and smartphone demand pushed DRAM prices up for six consecutive quarters. Samsung's operating profit more than doubled, and its stock rallied sharply.

That cycle ended with a clear trigger. In late 2018, new NAND factories came online — including Intel-Micron's Lehi, Utah fab and Samsung's own Pyeongtaek P2 expansion. Supply flooded the market right as smartphone demand softened. DRAM spot prices fell roughly 50% from peak to trough over the following year. Samsung's stock gave back a significant chunk of its gains.

The key difference this time: geopolitical constraints on Chinese memory makers may be structurally limiting new supply in a way that didn't exist in 2018. But the equivalent risk events are already on the calendar. Samsung's Taylor, Texas fab and Micron's Boise HBM expansion are both scheduled to contribute meaningful wafer output by late 2026 or early 2027. If those ramps coincide with any softening in AI capex, the 2018 playbook could repeat. As of early April 2026, Samsung's 52-week range — from roughly ₩53,700 to ₩223,000 — shows just how violently this stock can reprice when the cycle turns.

What could go wrong

Risk 1: The classic memory bust. Memory markets are cyclical by nature. High prices incentivize every producer to expand capacity. Samsung's Taylor fab, Micron's Boise ramp, and SK Hynix's Korean expansions are all scheduled to add meaningful output by late 2026 through early 2027. If those capacity additions arrive simultaneously, oversupply could crash prices — and Samsung's stock — just like in late 2018.

Risk 2: AI capex declines on a measurable timeline. The trigger to watch: if aggregate disclosed AI capital expenditure from Microsoft, Google, and Amazon declines more than 10% versus prior guidance over the next two quarterly earnings cycles (Q2 and Q3 2026 reports), memory demand assumptions for H2 2026 break. That's a calendar-specific risk with earnings dates already set.

Risk 3: Samsung is the thesis — and the risk. This article is deliberately Samsung-focused because Samsung sits at the intersection of memory supply, AI demand, and consumer devices. But that concentration cuts both ways. A Samsung-specific stumble could hurt the thesis without affecting the broader memory market. Think HBM yield problems, a delayed factory ramp-up, or losing smartphone share to Chinese phone makers. Micron and SK Hynix face the same cycle dynamics but different company-specific risks.

Invalidation trigger: If Samsung guides Q2 2026 operating profit flat or below Q1 levels — implying margin compression back toward mid-cycle levels — treat it as confirmation the cycle may be turning. Reassess the position within two weeks of that guidance release.

AI doesn't run on processors alone — it runs on memory, and the world's biggest memory maker is posting record profits while trading at a fraction of the valuation the market gives its peers.

Watchlist

TickerLevelStatusWhy
005930.KSMonitor — confirm current price at time of readingholdingSamsung is the epicenter of the memory supercycle. Targeting record Q1 2026 profit, still cheap at roughly 7x earnings. Watch Q2 guidance for cycle-turn signals.
TSMMonitorwatchingTSMC's CoWoS packaging capacity is the bottleneck gating HBM deployment. Controls ~70% of advanced chip manufacturing. Track packaging yield commentary on earnings calls.
MUMonitorwatchingMicron is the U.S.-listed pure-play on the same AI memory cycle. No consumer device cushion — more levered to the upside and the downside than Samsung.
000660.KSMonitorwatchingSK Hynix is Samsung's Korean rival and benefits from the same U.S. restrictions on Chinese competitors. Currently leads Samsung in HBM qualification with Nvidia.