Where Can an AI Infrastructure Bottleneck Beat Memory? Can FADU Open P, Q and a New Segment Together?

A follow-up on FADU as an AI eSSD controller bottleneck. The note tests whether FADU can earn a higher multiple than memory by opening P, Q, cost leverage and new segments such as Gen6 SSD controllers, PMIC and CXL.

📚 Connected context This is a follow-up to Is FADU Korea’s Sandisk Beta?. The previous note tested foreign flow and Sandisk correlation. This note asks the more important question: where can an AI infrastructure bottleneck earn a longer-duration multiple than memory? Related context: AI infrastructure multiple map, Goldman token demand vs JPM memory ASP peak-out, and Marvell/Broadcom bottleneck preview.

TL;DR

An AI infrastructure bottleneck does not beat memory just because demand is strong. It needs four things:

  1. P: pricing power or ASP uplift.
  2. Q: volume growth and customer programs.
  3. C: cost leverage and margin expansion.
  4. A new segment that expands the addressable market.

Memory already has strong P and Q. The problem is the market eventually worries about ASP peak-out. To earn a higher multiple than memory, a company must prove customer qualification, repeat revenue, new product expansion and margin leverage.

FADU has part of that setup. AI data centers are increasing demand for enterprise SSDs and SSD controllers. In 1Q26, FADU reported KRW 59.5bn revenue and KRW 7.7bn operating profit, turning profitable. About 80% of revenue came from controllers, and backlog at the end of March 2026 is estimated at roughly KRW 175.9bn. (The Bell)

But the conclusion is not buy now.

FADU could become a stronger AI infrastructure bottleneck than a simple memory cycle stock, but it still has to prove that through 2Q revenue, mix and margin.

The key number is 2Q26 revenue.

2Q26 revenueInterpretation
Below KRW 70bnWeak conversion from orders to revenue
KRW 80bn-90bnNot bad, but not enough for a strong rerating
KRW 90bn-105bnHealthy order-to-revenue conversion
Above KRW 100bnStrong signal that FADU is an AI storage bottleneck, not just a theme

Flows are strong but not clean. Over the last five trading days, foreigners bought about KRW 211.2bn net, but program net buying was also about KRW 203.4bn. That means the buying may include program or basket flow rather than pure long-only accumulation.

The practical stance is Watch / conditional entry. I would look for defense around KRW 107,000, recovery into KRW 112,000-117,000, continued foreign buying, easing institutional selling, and 2Q26 revenue above KRW 90bn.


1. What Kind of Bottleneck Can Beat Memory?

The best AI infrastructure investments are not simply important components. They are bottlenecks where pricing, volume, margins and market scope improve together.

AxisMeaningMemoryWhat FADU Must Prove
PASP and pricing powerStrong from DRAM/NAND price increasesGen5/Gen6 controller ASP, TCO premium
QShipments, orders, customer programsStrong HBM, DDR5, eSSD demandFast order-to-revenue conversion
CCost and margin structureHeavy capex and cycle riskFabless fixed-cost leverage and controller gross margin
New segmentNext product poolHBM4, SOCAMM, eSSDGen6 SSD controller, PMIC, CXL, low-latency SSD

Memory is still the main lane. But every memory cycle faces the same question: is this peak earnings, or can it last longer?

FADU can escape that question only if it is seen not as a NAND price beta, but as a controller and firmware platform that solves AI data-center storage bottlenecks.


2. FADU: Q Is Starting to Show Up

The biggest change in FADU is that orders are starting to become visible revenue.

[Fact] FADU reported 1Q26 revenue of KRW 59.5bn, operating profit of KRW 7.7bn and net profit of KRW 10.2bn. Operating margin was 12.9%. Based on user-provided research, controller revenue was about 80.4% of total revenue and finished SSD products were about 16.6%. (The Bell)

[Fact] Backlog at the end of March 2026 is estimated at USD 116.2mn, or about KRW 175.9bn. The 2026 delivery portion is estimated at USD 95.1mn, or about KRW 144.0bn. (DataTooza)

[Fact] On May 28, 2026, FADU disclosed an enterprise SSD controller supply contract worth roughly KRW 46.5bn, with the contract period running from May 28, 2026 to January 1, 2027. (The Elec)

These numbers matter because Q is no longer just a story. It is showing up through orders, backlog and revenue.

The caveat is that FADU’s 2025 revenue was KRW 92.4bn, so a contract around KRW 9bn can cross the KOSDAQ disclosure threshold. More May disclosures do not automatically mean new customer explosion. Some may be larger or rolling purchase orders from existing overseas NAND customers.


3. 2Q26 Is the Real Test

The user-provided contract-period allocation model points to about KRW 103.8bn of mechanical 2Q26 revenue. A more practical investment range is KRW 90bn-105bn.

The model matters because it back-tested well for 1Q26: it implied roughly KRW 59.1bn versus actual revenue of KRW 59.5bn. It is not perfect, but it is a useful order-to-revenue sanity check.

2Q26 revenueJudgment
Below KRW 70bnConversion is slower than expected
KRW 70bn-80bnNot a disaster, but not a rerating signal
KRW 80bn-90bnLower end of expectations
KRW 90bn-105bnGood zone; conversion is working
Above KRW 100bnStrong surprise; supports a 2026 revenue path above KRW 300bn

Mix matters too. The model estimates roughly KRW 86.2bn of controller revenue and KRW 17.6bn of finished SSD revenue in 2Q. If controller mix remains around 80%, FADU is easier to value as a controller/firmware bottleneck rather than a finished SSD company.


4. P Is Not Fully Proven Yet

FADU needs pricing power. But NAND price increases are not the same as controller ASP increases.

Higher NAND prices can support enterprise SSD investment and long-term supply commitments, which helps FADU. But controller ASP does not move like NAND spot prices. FADU’s P must come from:

DriverMeaning
Gen5 controller performanceRead performance per watt, thermals, reliability and firmware
Gen6 transitionHigher ASP and qualification barrier
Customer-specific designTailored controller and firmware for hyperscaler needs
TCO reductionMore work per watt and lower heat
PMIC/CXL expansionNew products beyond the SSD controller

Shinhan Securities framed FADU as an SSD controller company that can lower data-center total cost of ownership and reportedly set a KRW 130,000 target price. (NewsPim)

[Inference] The direction is right. But the market wants numbers. If P is real, it should show up in ASP, controller gross margin, repeat orders and Gen6 commentary from 2H26 onward.


5. New Segments: Gen6, PMIC and CXL

FADU needs a next product pool to earn a higher multiple.

Current revenue is still centered on Gen5 enterprise SSD controllers. Gen6 is upside, not base revenue. PMIC and CXL are also options, not proven segments.

SegmentCurrent interpretationInvestment meaning
Gen5 eSSD controllerCore current revenue2026 execution proof
Gen6 eSSD controller2027+ upsideHigher ASP and qualification barrier
Finished SSD productsSome revenue contributionHelps Q, but margin needs proof
PMICPower-management optionCould expand the storage power-efficiency TAM
CXLMemory expansion/server architecture optionLong-duration optionality

Some sell-side scenarios point to KRW 300bn revenue in 2026, more than doubling in 2027, and KRW 800bn-900bn by 2028. If that happens, FADU becomes an AI storage platform, not just a controller vendor.

But this is still a bull case. The market needs 2Q revenue, 2H orders, Gen6 customer evidence and margin proof.


6. Flow: Strong, But Not Clean

Source: Research OS local DB and Kiwoom investor_flow_raw_daily, as of the June 2, 2026 close. Price uses prices_daily.

PeriodRetailForeignersInstitutionsFinancial investmentTrustPrivate fundPension etc.
5D-KRW 61.8bn+KRW 211.2bn-KRW 148.3bn-KRW 67.6bn-KRW 25.1bn-KRW 43.2bn-KRW 10.3bn
10D-KRW 86.2bn+KRW 333.9bn-KRW 235.6bn-KRW 71.5bn-KRW 46.4bn-KRW 106.4bn-KRW 9.6bn
20D-KRW 241.3bn+KRW 547.5bn-KRW 276.7bn-KRW 61.5bn-KRW 11.3bn-KRW 204.8bn+KRW 1.0bn

Recent daily flow also shows persistent foreign absorption.

DateForeignersInstitutionsRetailInterpretation
2026-06-02+KRW 38.0bn-KRW 16.9bn-KRW 18.7bnForeign defense, retail/institution selling
2026-06-01+KRW 37.2bn-KRW 33.7bn-KRW 3.5bnHeavy institutional selling
2026-05-29+KRW 29.1bn-KRW 22.8bn-KRW 5.0bnForeign-only absorption
2026-05-28+KRW 53.7bn-KRW 35.2bn-KRW 16.8bnHeavy foreign buying during weakness
2026-05-27+KRW 53.3bn-KRW 39.8bn-KRW 17.8bnForeign buying, institutional distribution

The issue is program flow.

ItemRecent state
5D program net buy+KRW 203.4bn
10D program net buy+KRW 269.1bn
5D short-sale valueKRW 82.9bn, average ratio 5.88%
10D short-sale valueKRW 112.7bn, average ratio 4.52%
Stock loan balance1.656mn shares on June 1; June 2 data collected as zero and is unreliable

Foreign net buying and program net buying are almost the same size. So recent foreign buying should not be treated as pure long-only accumulation yet.


7. Valuation: Expensive on 2026, Reasonable Only If the Path Opens

At the June 2, 2026 close of KRW 107,100, FADU’s market cap is about KRW 5.37tn. Using user-provided consensus/research scenarios:

Revenue baseMarket cap / revenueInterpretation
2026E KRW 300bn~17.9xHigh growth already priced in
2027E KRW 600bn~8.9xRequires customer expansion and doubling revenue
2028E KRW 800bn~6.7xExplainable if platformization works
2028E KRW 900bn~6.0xStrong bull case

The stock is expensive on 2026 revenue. It becomes more reasonable only if 2027 and 2028 growth materialize.

The required proof:

  1. 2Q26 revenue above KRW 90bn.
  2. Controller mix above 75-80%.
  3. Operating margin improving toward the mid-teens.
  4. New segment evidence from Gen6, PMIC or CXL.

8. Red Team

RiskExplanation
Existing customer repeat ordersMay disclosures may reflect larger rolling orders, not new customer explosion
Low disclosure thresholdKRW 9bn contracts look large because 2025 revenue was low
Customer concentrationA few large customers can create valuation discount
Gen6 delayWeakens the new-segment thesis
Program-driven flowForeign buying may include basket/program activity
Institutional selling and shortingCould cap upside
NAND price illusionNAND ASP up does not automatically mean controller ASP up

The most important line:

A good company and a good entry price are not the same thing.


9. Practical Strategy

Current stance: Watch / conditional entry.

ConditionMeaningAction
Defense around KRW 107,000Recent foreign-buying zone holdsKeep watching
Recovery to KRW 112,000-117,000Near-term range recoveryConsider first entry
Foreign net buying for 3+ daysPersistence beyond program flowFlow quality improves
Easing institutional sellingLess overhead supplyUpgrade possible
2Q26 revenue above KRW 90bnOrder-to-revenue conversion proofCore trigger
2Q26 revenue above KRW 100bnStrong surpriseMultiple expansion possible
Break below KRW 103,000 with weaker foreign buyingFlow failureReassess

FADU is not a cheap stock. It is a stock that can become cheaper if numbers catch up.


Final View

There are AI infrastructure bottlenecks that can beat memory. But they need P, Q, margin leverage and new segments together.

FADU is a candidate.

Q is starting to show up through backlog, 1Q revenue and May contracts. P is still being proven through controller performance, Gen6, TCO reduction and customer-specific firmware. New segments such as PMIC, CXL and Gen6 are still options.

Flows are strong, but not clean. Foreign buying is large, but program buying is almost the same size.

So the final judgment is:

FADU can become a better AI infrastructure bottleneck than a simple memory beta, but it must prove that with 2Q26 revenue of KRW 90bn-105bn, sustained controller mix, easing institutional selling and continued foreign demand.

If the numbers follow, FADU can move beyond “Korea’s Sandisk beta” and become Korea’s representative AI storage bottleneck stock.


Evidence Classification

[Fact]

  • FADU 1Q26 revenue of KRW 59.5bn, operating profit of KRW 7.7bn and net profit of KRW 10.2bn are based on public coverage and user-provided research. (The Bell)
  • End-March 2026 backlog of about USD 116.2mn, or KRW 175.9bn, is based on user-provided research and public data summaries. (DataTooza)
  • FADU disclosed a KRW 46.5bn enterprise SSD controller contract on May 28, 2026. (The Elec)
  • June 2, 2026 close of KRW 107,100 and 5D flow of foreign +KRW 211.2bn, institution -KRW 148.3bn and retail -KRW 61.8bn are from Research OS local DB and Kiwoom investor_flow_raw_daily.
  • 5D program net buy of KRW 203.4bn is from Research OS local DB program_trade_daily.

[Inference]

  • FADU’s Q is becoming visible, but P still needs proof through controller ASP, Gen6 and TCO premium.
  • Recent buying likely includes program/basket flow because foreign net buying and program net buying are nearly the same size.
  • 2Q26 revenue of KRW 90bn-105bn is the key zone for proving order-to-revenue conversion.

[Speculation]

  • 2027 revenue of KRW 600bn and 2028 revenue of KRW 800bn-900bn are bull-case sell-side/user-provided scenarios, not company guidance.
  • Gen6, PMIC and CXL becoming meaningful new segments remains unproven.
  • FADU deserving a higher multiple than memory is still a thesis before 2Q and 2H execution proof.

[Blocked]

  • Customer-level orders, product ASP, controller gross margin and Gen6 customer names are not publicly confirmed.
  • June 2 stock-loan balance was collected as zero and is unreliable; June 1 balance of 1.656mn shares is treated as the latest usable value.
  • Full sell-side reports were not independently re-read in full; this note uses user-provided summaries plus public coverage.

Coverage Health

  • Price and flow data use Research OS local DB and Kiwoom-derived flow surfaces. Investor flow is stored in million_krw; program and short-sale data are stored in KRW and converted to KRW 100mn units.
  • Financial and contract figures use public coverage plus user-provided research notes. Official filing verification can be added in a later update.
  • This is research, not investment advice. FADU is a volatile growth stock; chasing without 2Q revenue and flow confirmation creates unfavorable loss asymmetry.
Built with Hugo
Theme Stack designed by Jimmy