Micron vs SK hynix vs Samsung: Who is Winning The AI Memory Chip Race?

When you ask who is winning the AI memory race, you get three different answers: SK hynix sells half the world's high-bandwidth memory, Samsung leads the broader DRAM and NAND markets and Micron's stock is up about 590% in a year. 

The awkward question is whether the best memory business is also the best stock at the price an investor must actually pay.

Let's break down who leads in HBM, ordinary DRAM and storage, what each company actually earns and how its stock has performed. Then we'll test the valuations against a less generous memory cycle and decide which exposure makes the most sense for an investor today.

Micron vs SK Hynix vs Samsung: What Do They Actually Sell?

DRAM is working memory: it holds the information a processor needs right now and loses that information when power goes off. 

NAND flash keeps information when power is off and goes into phones and solid-state drives (SSDs). 

High-bandwidth memory (HBM) is a technically demanding form of DRAM stacked close to an AI processor so huge amounts of data can reach it quickly. 

Think of DRAM as a cook's worktop, NAND as the pantry and HBM as an exceptionally fast serving counter beside the stove. The same restaurant needs all three, but each earns money under different conditions.

The word storage needs care. NAND-based SSDs compete in one part of storage; Seagate and Western Digital also make hard-disk drives. HBM and ordinary DRAM are working memory, so calling all three companies interchangeable “storage stocks” hides the contest that matters most to AI investors.

Our way to judge the winner is to keep three scoreboards separate: 

• who owns the valuable product, 
• whose profits can survive a normal memory cycle, and 
• what the specific share or depositary receipt costs. 

A lead on the first scoreboard cannot settle the other two.

Who Leads HBM, DRAM And NAND Today?

Sources: TrendForce, September 7 and August 18, 2026; Counterpoint Research, September 2, 2026. Shares may not sum to 100% because of rounding and other suppliers. DRAM, HBM and NAND have different market definitions; do not add the rows together.

Here is the important twist. SK hynix still controlled 50% of HBM revenue in Q2, but that was down from 58% in Q1. Samsung rose from 21% to 33% over the same period; Micron moved from 21% to 18%. SK hynix's lead is real, yet a rising Samsung can capture a bigger share of the next generation even if the HBM market itself keeps growing. A falling percentage does not prove falling sales.

Samsung's 39.4% DRAM share tells a different story: it sells the most across the broader working-memory market. Micron's 23.3% share looks modest beside Samsung's, but TrendForce says its DRAM revenue grew 65.5% quarter on quarter, faster than Samsung's 63.4% and SK hynix's 37.9%, as it prioritised higher-priced server DRAM. The three together accounted for about 87.6% of DRAM revenue in that particular TrendForce quarter. This is a concentrated industry, but its shares are not permanent entitlements.

NAND flips the focus from AI accelerators to storing the mountains of data around them. Samsung led with nearly $23.06 billion of Q2 NAND revenue and a 29.3% share. SK hynix Group, including Solidigm, came second with more than $14.27 billion; Micron moved to third with $11.85 billion, up 99.2% from Q1. That helps explain why a company can lose ground in HBM share while making substantial gains elsewhere in memory. (Source: TrendForce, August 18, 2026.)

Micron vs SK Hynix vs Samsung: What Is Each Company's Competitive Advantage?

SK hynix: the best HBM customer position

HBM is difficult to make reliably: multiple memory layers must be stacked, connected and tested, then qualified for a customer's processor design. A supplier cannot instantly replace another by cutting its quoted price. SK hynix has turned this manufacturing and qualification head start into half of global HBM revenue. It says HBM4 mass shipments began in Q2 2026 and it has long-term agreements with around 10 customers. Its Solidigm operation adds a second leg in high-capacity enterprise SSDs.

The limitation is just as important as the advantage. Qualification is sticky for a product generation, not an eternal licence to charge any price. Samsung's HBM share gain is evidence that large customers are increasingly willing and able to qualify another supplier. Our view: SK hynix owns the strongest existing HBM franchise, but investors should stop treating 50% market share as a permanent forecast.

Samsung: the ability to win on more than one chip

Samsung has the broadest manufacturing footprint within the three businesses being compared: leading positions in DRAM and NAND, a growing HBM business and a separate chip foundry that manufactures designs for outside customers. Its Q2 results say HBM4 sales scaled up and HBM4E samples were shipped to major customers. Samples show technical progress; they are not the same as large, qualified revenue. Samsung's scale gives it many ways to benefit when memory prices rise, but foundry and consumer electronics also demand capital and management attention.

The stock is sometimes described as a defensive way to own memory because it also sells Galaxy phones and TVs. That description fails this quarter's reality check: Samsung's semiconductor division earned KRW 89.2 trillion of the group's KRW 89.5 trillion operating profit in Q2, while its mobile and networks unit lost KRW 0.7 trillion. The other businesses may help across a full cycle; right now, investors buying Samsung are overwhelmingly exposed to memory profitability.

Micron: customer contracts and a broader server opportunity

Micron's position is bigger than its HBM rank suggests. It sells conventional server DRAM, low-power data-center memory, HBM and enterprise SSDs. Its fiscal Q3 data-center revenue exceeded $25 billion; by its June earnings call, it had shipped more than $1 billion of HBM4 products. It had signed 16 strategic customer agreements, 14 of which represented roughly $100 billion of minimum contracted revenue over their remaining terms.

Those agreements reduce some uncertainty over future volumes and pricing. They do not mean $100 billion of profit or cash already in the bank: the revenue extends over several years, performance obligations depend on delivery and the proposed customer deposits are eventually returned or credited. Micron expects about $27 billion of capital spending in fiscal 2026 and has warned that quarterly capital spending in fiscal 2027 will rise above its projected fiscal Q4 level. Its moat is improving commercial visibility as well as technology; the test is whether it can preserve returns on all that new capacity when the shortage eases.

How Do Their Latest Earnings Actually Compare?

Sources: Micron fiscal Q3 2026 results; SK hynix Q2 2026 results; Samsung Electronics Q2 2026 earnings presentation. Currencies, quarter ends and business mixes differ. Samsung does not separately report memory-only operating profit in this comparison, and Micron's free cash flow figure is adjusted.

The margins are astonishing. They also make the comparison dangerous. These businesses have benefited from scarce supply and higher selling prices; a 76% margin is not a sensible perpetual assumption for a memory producer. Samsung's group margin is lower partly because the group includes phones, displays and appliances, so do not infer that its memory fabs are inherently less profitable from the group figure alone.

SK hynix's KRW 93.92 trillion of Q2 net income is a particularly poor figure to annualise: it exceeds the company's KRW 79.32 trillion of revenue and was boosted by gains below operating profit. That is why our valuation exercise uses operating profit as a second lens. Micron, meanwhile, has guided to $50 billion ± $1 billion of revenue and roughly 86% gross margin for fiscal Q4, but those are management projections for a September 30 release, not results already earned.

How Have Micron, SK hynix And Samsung Shares Performed?

Source: Google Finance

Micron wins the stock-price race over one year; that is an observation about investors' returns in dollars, not proof that its memory technology is best. SK hynix's US-listed SKHY receipts began regular trading only in July 2026, so a one-year SKHY return cannot honestly be placed beside MU's. The table uses the long-established Seoul ordinary share instead.

September 28 was also a reminder of how quickly sentiment changes: the Korean ordinary shares finished down about 5% for SK hynix and 5.4% for Samsung in one session. A day's selloff does not by itself tell us whether customers ordered fewer chips. The next useful evidence will come from realised selling prices, HBM qualifications and the next earnings reports.

Micron vs SK Hynix vs Samsung Valuation: Which Stock Looks Cheapest?

This table gives Micron credit for what it has achieved without pretending the stocks cost the same. Its shares trade for roughly twice the price per dollar or won of recent operating profit of either Korean company. Micron might merit a premium for US-market access, product execution and its new contracts. But a premium of that size requires a stronger belief that exceptional earnings will remain exceptional. Samsung's lower price/book leaves more balance-sheet value behind each unit of market value, although book value alone cannot tell us whether future fabs will earn attractive returns.

Here is our simple memory-cycle test. Hold each share price fixed and reduce the last 12 months' operating profit by 25% or 50%. These are deliberately hypothetical cuts, not earnings forecasts or price targets.

The arithmetic is just current equity value ÷ assumed annual operating profit. It shows why a low P/E after a price surge is not automatically a bargain: the denominator may be near a cycle high. A 50% decline is one sensitivity, not our prediction. Long-term contracts may cushion it and new capacity or weaker chip pricing may worsen it. SK hynix's reported P/E is especially flattering because the Q2 non-operating gains inflated trailing net income.

Capital returns belong on this scoreboard too. SK hynix's board approved a KRW 40 trillion buyback and cancellation programme, worth about 3.1% of its September 28 quoted equity value. Samsung outlined KRW 90-110 trillion of planned 2026 shareholder returns; roughly KRW 15 trillion of its planned repurchases are earmarked for employee compensation. These are plans, not an extra source of operating profit. Check how much cash actually reaches holders and how many shares actually disappear.

The extra valuation trap: SKHY is not the Seoul share at the same price

SK hynix also trades on Nasdaq as SKHY. One ADR represents one-tenth of a Korean ordinary share. On September 23, the Korean share closed at KRW 1,862,000, implying KRW 186,200 per ADR. At roughly KRW 1,366 per US dollar, parity was about $136.31; SKHY closed near $189.28 in New York. That is an indicative 39% premium to parity on matched-date prices. (Sources: SK hynix's US depositary-share disclosure; Korea share and Nasdaq ADR closing data; USD/KRW historical rate, September 23, 2026.)

That gap changes the investment question. A reader looking at SK hynix's ~7.8× Seoul P/E cannot assume the same earnings are available at ~7.8× through SKHY. Depositary-share supply, conversion frictions, settlement timing, FX moves and demand can all affect the gap, and it need not vanish on an investor's timetable. Recalculate it with live, matched prices before acting; the September 28 Seoul close and September 25 US close should not be presented as if they occurred at the same instant. Our earlier SKHY ADR analysis explains this listing issue in more detail.

Who Else Competes With Micron, SK Hynix and Samsung in Memory Chips?

The three do not own the entire storage world. Kioxia and SanDisk are important NAND rivals, while China's YMTC is another NAND competitor. China's CXMT and Taiwan's Nanya and Winbond matter in DRAM, especially as customers and manufacturers revisit older product lines. Hard-drive makers Seagate and Western Digital compete for some data-storage budgets without being HBM manufacturers. None currently changes the fact that SK hynix, Samsung and Micron dominate the high-end DRAM contest; their ability to defend that position must keep being tested, particularly against Chinese supply in conventional memory.

For investors who want the theme without choosing a single supplier, the Roundhill Memory ETF (DRAM) holds global memory businesses including these three. Roundhill reports a 0.65% annual expense ratio and says holdings and weights can change. It spreads company-specific execution risk but still leaves investors exposed to a shared downturn in memory pricing; it is not three uncorrelated bets.

Micron vs SK Hynix vs Samsung: Which AI Memory Stock Looks Strongest?

Our business winner is SK hynix. Half of HBM revenue, proven shipments and relationships with large AI customers give it the clearest exposure to the hardest-to-replace memory inside accelerators. At the Seoul ordinary share price, roughly 10× trailing operating profit makes a more convincing combination of franchise and valuation than Micron's roughly 21×. It would be our first business to study for focused AI memory exposure. The condition is crucial: we would not carry that valuation conclusion unchanged over to a Nasdaq ADR trading at a 39% parity premium. Samsung's fast HBM gains are the strongest reason the SK hynix lead could narrow.

Samsung is our most interesting challenger at its Korean ordinary-share price. It is already number one in total DRAM and NAND revenue, has lifted its HBM share sharply and trades at about 9.8× trailing operating profit and 3.0× book. If HBM qualification continues to improve while the phone and foundry businesses recover, its breadth could turn from a distraction into an advantage. The risk is that its current profit remains heavily dependent on one exceptionally favourable memory cycle. Calling it “diversified and safe” would overstate what the latest quarter shows.

Micron is the hardest price to defend today, despite excellent execution. Its roughly $100 billion of minimum contracted multiyear revenue gives it a credible answer to the old memory-cycle problem and its recent cash generation is substantial. But its shares have risen about 590% in a year and cost about twice as much relative to recent operating profit as the Korean ordinary shares. We would need to see repeated evidence that contracts protect margins after supply improves before treating that premium as comfortable. The September 30 results will be an immediate test; the June guidance must never be passed off as September results.

What Would Change This Ranking?

1. HBM qualification and share: Does SK hynix hold its lead as Samsung ships more HBM4, or does Samsung's Q2 share gain continue? Check revenue share and actual qualified shipments, not sample announcements alone.
2. Conventional DRAM and NAND prices: If selling prices cool while shipments rise, the profit improvement may reverse even as revenue stays high. TrendForce's contract-price reports and company gross margins are more useful than a single stock-market move.
3. Contracts versus real cash: Track deliveries under Micron's agreements, operating cash flow after fab spending and whether customer deposits convert into profitable sales. Do the same for SK hynix's longer agreements.
4. Capacity discipline: Follow Micron's US and Asian projects, SK hynix's Korean and US plans and Samsung's HBM and foundry spending. Today's shortage is valuable partly because supply takes time to build; tomorrow's oversupply is the opposite side of that investment.
5. The security's own price: Recheck SKHY against Seoul and the exchange rate, then recalculate valuation using a reasonable through-cycle profit. This matters more than declaring a winner from a static market-share chart.

The lesson of this three-way contest is simple: SK hynix currently makes the strongest HBM argument, Samsung has the greatest scale and catch-up potential, and Micron has turned supply tightness into remarkable earnings and contracts. The investor's edge comes from asking how much of each advantage remains when memory is no longer scarce and how much the share price already assumes it will.