FADU Inc. (440110.KQ) - Deep Dive Research Report

Prepared 2026-06-21. Listing venue: KOSDAQ (South Korea). Sector: Technology / fabless semiconductors.

A note on sources for this report. FADU is a Korean small-cap. Like most KOSDAQ companies of its size, it does not publish verbatim English-language earnings-call transcripts. What it does publish, on a quarterly cadence, is a Korean-language earnings disclosure plus an IR news release, supplemented by occasional press briefings (notably its 10th-anniversary press conference in 2025). The "concall" analysis in Sections 7 and 9 is therefore built from FADU's six most recent quarterly earnings releases and management briefings, treated as the functional equivalent of the call. This is flagged where it matters. The insider data in Section 11 is drawn from the MoatMap cross-market disclosure database (Korea/DART feed), which is the canonical source for this gated venue; that scrape is noted as 41 hours stale at the time of writing.

Section 1: What the company does

FADU makes the chip that runs a solid-state drive. Not the storage itself - the NAND flash memory that physically holds the data is made by the likes of Samsung, SK hynix, Micron and Kioxia. FADU makes the controller: the small processor that sits between the host computer and the flash, deciding where every byte goes, correcting errors, managing wear across the flash cells, and serving read and write requests fast enough that a rack of servers does not sit idle waiting for data. If the NAND is the warehouse, the controller is the warehouse manager, the forklift fleet, and the logistics software all at once. An SSD with a mediocre controller is a fast warehouse run badly.

FADU is a fabless company, founded in 2015 in Seoul. Fabless means it designs the chip and hands the blueprint to a foundry (TSMC and others) to manufacture; it owns no fabrication plant. Its entire value is in the design - the architecture of the controller and the firmware that runs on it. The company focuses almost exclusively on the hardest, highest-margin corner of this market: enterprise SSD controllers (eSSD), the controllers that go into the drives bought by hyperscale data centres rather than the ones in your laptop. This is a deliberate and unusual choice. The volume market for client/consumer SSD controllers is dominated by Taiwanese merchant suppliers selling cheap, good-enough silicon. FADU went the other way, aiming at the part of the market where performance-per-watt is worth paying for and where there are only a handful of credible suppliers worldwide.

The core value proposition is power efficiency at data-centre scale. A hyperscaler running hundreds of thousands of SSDs cares enormously about the watts each drive draws, because power and cooling are the binding constraint on how much compute fits in a building. FADU's pitch is that its controllers deliver leading throughput at materially lower power than competing designs, which translates directly into lower total cost of ownership for an operator like Meta or Google. That is the whole game: not "our drive is faster" but "our drive does the same work for fewer watts."

The technical heart of the differentiation is architectural. Where a conventional SSD controller tends to be substantially redesigned each time the PCIe interface jumps a generation (Gen3 to Gen4 to Gen5 to Gen6), FADU built its controller around a pipeline of small parallel processing units (PPUs) layered on top of 64-bit RISC-V cores. An NVMe command is broken into stages, and each stage is handled by a dedicated PPU, with many PPUs running concurrently. This lets FADU optimise each pipeline stage independently and carry the architecture forward across PCIe generations without an overhaul, which is the source of both its claimed power efficiency and its ability to ship new generations relatively quickly for a company its size.

A concrete example of how this actually works commercially: SK hynix uses a FADU controller inside enterprise NVMe SSDs that SK hynix in turn supplies to Meta. FADU does not sell to Meta directly in that arrangement; it sells (or licenses) the controller to SK hynix, which builds the finished drive and books the hyperscaler relationship. FADU earns controller revenue plus, reportedly, technology fees. The same pattern repeats with Western Digital/SanDisk, whose SN861 enterprise drive uses a FADU controller and is qualified for NVIDIA's GB200 NVL72 AI systems. FADU is, in effect, the silicon brain inside other companies' branded drives - and increasingly inside the drives that feed AI training and inference clusters.

The company cannot be understood without its founding-era trauma, which still shapes everything around it. FADU listed on KOSDAQ in August 2023 in one of the most hyped tech IPOs of the year. Within weeks the story imploded: revenue for the quarters straight after listing essentially vanished, and it emerged that major customer orders had been suspended before the IPO without that being disclosed in the prospectus. The stock collapsed, investors sued, and prosecutors eventually indicted executives and the company itself. As of late 2025 FADU sits under a formal Korea Exchange listing-eligibility (delisting) review. So this is a genuinely two-sided story: a company with real, scarce, hyperscaler-validated controller technology, wrapped inside an unresolved capital-markets-fraud and delisting overhang. Both halves are real, and both are covered in detail below.

Section 2: Business segments

FADU is close to a single-business company - it makes eSSD controllers - but its revenue arrives through two distinct go-to-market models, and it has a small strategic-optionality arm. It is best understood as two operating lines plus one subsidiary bet.

Controller business (the core, roughly 80% of recent revenue)

This is FADU's reason to exist. The company designs the controller silicon and the firmware, has it fabricated at a foundry, and supplies it to NAND makers and drive builders who integrate it into finished enterprise SSDs. Revenue here comes from selling the controller chips and, in some relationships, technology/licensing fees. In Q1 2026 management said controllers made up roughly 80% of revenue, and they framed that mix shift toward controllers as the reason profitability improved sharply - controllers carry far better economics than reselling assembled drives.

The core capability is the PPU-pipelined RISC-V architecture described above, plus the firmware and the multi-year qualification track record with demanding customers. This is not replicable on a short timeline. Getting an enterprise controller designed in at a hyperscaler requires years of co-engineering, reliability validation, and trust that the supplier will be around to support the part across its service life. FADU has now cleared that bar with at least two of the four global hyperscalers (confirmed for mass-production supply), which is the single most important fact about the business.

Competitively, this segment fights Marvell at the high end, the in-house controller teams of Samsung and SK hynix/Solidigm, and the merchant suppliers (Phison, Silicon Motion) reaching upmarket. FADU wins on power efficiency and on being an independent, customisable alternative to a NAND maker's own captive controller (a hyperscaler that buys NAND from one vendor may not want to be locked into that vendor's controller too). It loses where a customer prefers a vertically integrated drive from a single NAND supplier, or where Marvell's scale and incumbency dominate.

This is unambiguously the margin engine, the growth bet, and the strategic core all at once.

SSD module / finished-drive business (roughly 20% of recent revenue)

Alongside selling controllers to others, FADU also builds and sells some finished SSD modules itself, packaging its controller with NAND into a drive. Management has described new module customers being added in Asia as a driver of certain quarters' revenue. This line gives FADU a way to monetise its controller even where it cannot win a pure controller socket, and to seed relationships, but it carries thinner economics (FADU is buying NAND and reselling assembled product). Management's stated preference, visible in the Q1 2026 commentary, is for the controller mix to rise and the module mix to fall, precisely because the controller is where the value and the margin sit. Treat this as a supporting / seeding line, not the prize.

EEUM subsidiary - CXL and next-gen memory (strategic option, currently frozen)

FADU pursued Compute Express Link (CXL) memory-sharing switch technology through a subsidiary, EEUM. The thesis was that AI data centres would need to pool and share memory across servers, and a CXL switch would be a large new market. As of late 2025 management has deliberately scaled this back: it will take the CXL switch to FPGA proof-of-concept stage and then "freeze it" until the CXL memory-sharing market becomes substantial. The reason given is that NVIDIA's NVLink has strengthened as the memory-pooling fabric inside GPU pods, pushing the broad CXL market further out than hoped. This is a strategic option being kept alive cheaply rather than an active business, and management's willingness to pause it rather than burn cash chasing a market that has not arrived is itself a data point on discipline.

Section 3: Products and business detail

FADU's product catalogue is organised by PCIe generation, with each named controller tied to a bus generation and, increasingly, integrated companion features.

BRAVO (2020) - PCIe Gen3. FADU's first enterprise controller generation. It established the architectural approach and the company's credibility as a controller designer. Now largely superseded but it anchors the family.

DELTA (2021) - PCIe Gen4. The Gen4 controller. This is the generation that carried FADU into serious enterprise engagements as Gen4 became the data-centre standard.

ECHO (2023) - PCIe Gen5. The current volume product and the one that matters most today. ECHO shipped into mass production from late 2024, riding the opening of the Gen5 enterprise market. The AI-data-centre build-out is overwhelmingly a Gen5 story right now, and ECHO is FADU's vehicle into it. Gen5 controller sales are the proximate reason FADU's revenue inflected upward through 2025 and into a profitable Q1 2026.

Sierra FC6161 (announced at Flash Memory Summit 2025) - PCIe Gen6. FADU's next-generation flagship, the centre of its 2026 narrative. The published specifications are striking: up to roughly 28.5 GB/s sequential throughput, about 6.9 million random read IOPS, support for capacities up to 512TB, and power draw under 9W. FADU also claims Sierra delivers roughly 2x the power efficiency of ECHO, helped by aggressive use of NVMe power states. Management has signalled plans to integrate DRAM PMIC (power management) functionality into the module, positioning ahead of competitors who still supply those components separately. Gen6 launch is expected through 2026. The heavy Gen6 development cost (a final ~KRW 11 billion R&D charge was flagged in Q4 2025) was the main reason FADU stayed loss-making at the full-year-2025 level even as revenue more than doubled.

On manufacturing: FADU owns no fab. The controllers are fabricated by external foundries (TSMC-class), packaged and tested, and the finished silicon is shipped to customers. The "process knowledge" that protects FADU is not in the fab; it is in the controller architecture, the firmware, and the years of qualification work embedded in each design win. That is the moat such as it is - design IP and customer qualification, not capital equipment.

On geography: FADU is a Korean company selling into a global hyperscale supply chain. Its reach is defined by its customers' reach. Through SK hynix, its controllers travel into Meta's data centres. Through Western Digital/SanDisk, into systems qualified for NVIDIA's flagship AI rack. FADU itself has confirmed mass-production controller supply for two of the four global hyperscalers (the set being AWS, Google, Microsoft, Meta), with management expressing confidence about adding more. New SSD-module customers have been added in Asia. The end market is, in one line, the AI data centre wherever it is being built.

Milestones that changed the business: first enterprise controller (BRAVO, 2020); the August 2023 KOSDAQ IPO (and its near-immediate revenue collapse); ECHO Gen5 mass production from late 2024; the Sierra Gen6 unveil at FMS 2025; and the run of consecutive large orders through 2025 into early 2026 that took cumulative new orders, by April 2026, past the entirety of full-year 2025 revenue.

Section 4: Customers

FADU's customers are not data centres. They are the companies that build the drives the data centres buy: NAND makers and SSD vendors. This is a crucial distinction for understanding both the revenue model and the concentration risk.

The two confirmed anchor relationships are SK hynix and Western Digital/SanDisk. SK hynix integrates FADU controllers into enterprise NVMe SSDs and supplies those drives to Meta, paying FADU controller revenue plus reported technology fees (one figure cited in the press put SK hynix's technology payments to FADU at around KRW 80 billion). Western Digital/SanDisk's SN861 enterprise drive uses a FADU controller and is qualified for NVIDIA's GB200 NVL72 AI system, which is about as strong a third-party validation of an enterprise controller as exists. Through these and other relationships, FADU's silicon reaches Meta and Google. FADU has publicly confirmed mass-production controller supply to two of the four global hyperscalers.

The buying decision inside these customers sits with storage architecture and platform engineering teams, with hyperscaler procurement and reliability organisations as the ultimate gatekeepers. The criteria are specific and unforgiving: performance-per-watt at the drive level, sustained (not just peak) throughput and IOPS, reliability and endurance over a multi-year service life, firmware quality, roadmap credibility for the next PCIe generation, and supply assurance. FADU wins on the first of those above all - power efficiency - and increasingly on roadmap, having a Gen6 part in hand while the market is still ramping Gen5.

Switching costs are high and run in FADU's favour once it is designed in, and against it before. Qualifying a controller into an enterprise drive that a hyperscaler will deploy at scale is a multi-year process involving deep co-engineering and exhaustive reliability validation. Once FADU's controller is qualified into a customer's drive platform, displacing it mid-generation is expensive and risky for the customer. But the same dynamic means every new socket is a slog to win, and a generation transition is the moment a competitor can try to dislodge an incumbent. FADU's architectural ability to carry its design across generations is partly a switching-cost play: it lets a qualified customer move from Gen5 to Gen6 with FADU rather than re-evaluate the field.

Concentration is real and is the dominant customer risk. A small number of NAND/drive partners (SK hynix, WD/SanDisk, and a short list of others) account for the bulk of revenue, and ultimately a handful of hyperscalers drive the end demand. This concentration is exactly what detonated in 2023: when major-customer orders were suspended, revenue did not just dip, it collapsed to near zero for two quarters. The flip side is that concentration here is partly a reflection of quality - there are only a few customers in the world who buy leading-edge enterprise controllers at all, and winning even two of the four hyperscalers is a genuine achievement. But investors should hold both thoughts: the customer set is elite, and it is dangerously narrow.

Contract structure is a mix of controller supply (priced per unit, tied to the customer's drive shipment volumes), technology/licensing fees in some relationships, and lower-margin finished-module sales. Because end demand is driven by hyperscaler capex cycles and specific drive-platform qualifications, revenue is lumpy - large orders arrive in clusters (FADU described four consecutive months of large orders into late 2025/early 2026) rather than as a smooth recurring stream. This lumpiness is structural to the business and is why a single quarter tells you little.

Section 5: Competitive landscape

The enterprise SSD controller market is concentrated. Roughly six vendors account for the large majority of total SSD controller revenue, and the enterprise/data-centre tier is narrower still. FADU competes in that top tier, which has three kinds of rival: the independent high-end merchant leader (Marvell), the Taiwanese volume merchants moving upmarket (Phison, Silicon Motion), and the captive in-house controller teams of the NAND giants (Samsung, SK hynix/Solidigm, Kioxia, Micron).

Marvell is the most direct competitor at the high end. Its Bravera SC5 family targets exactly FADU's territory - PCIe Gen5 NVMe data-centre controllers - and Marvell brings enormous scale, a broad data-centre silicon portfolio, and deep hyperscaler relationships. FADU's counter is power efficiency and being a focused, independent alternative; Marvell's edge is scale, breadth, and incumbency. This is the matchup that most defines FADU's ceiling.

Phison is the merchant controller leader by unit volume, strong across client and enterprise, and has pushed into branded high-capacity enterprise drives (the Pascari line, up to 122TB). Phison's strength is breadth and volume; against FADU it competes more on the strength of a full-stack offering than on best-in-class performance-per-watt at the very top.

Silicon Motion dominates client/OEM controller supply and is a formidable volume player, though its centre of gravity is client rather than the leading-edge enterprise tier FADU targets. It is more an adjacent threat (should it push harder into enterprise) than a head-to-head rival today.

Captive in-house teams - Samsung, SK hynix/Solidigm, Kioxia, Micron - are the subtlest competitive factor. Each NAND maker can build its own controller, and a vertically integrated drive is the default option a hyperscaler weighs against a FADU-plus-NAND combination. FADU's existence depends on the argument that an independent best-of-breed controller, decoupled from the NAND vendor, is worth choosing - which is most compelling precisely to the hyperscalers who want to multi-source NAND without being captive to any one vendor's controller. Note the nuance: SK hynix is simultaneously a major FADU customer and an owner of in-house controller capability. That coopetition is a structural feature of FADU's world.

Barriers to entry are high but not absolute. Designing a qualified enterprise controller takes years, deep firmware expertise, foundry access at leading nodes, and - hardest of all - the customer qualification track record. FADU's own history shows it can be done by a startup, but it also shows how brutal the path is. The barrier that most protects FADU now is not the design itself but the installed base of qualified design wins at two hyperscalers, which a new entrant would need years to replicate.

Where FADU is strong: performance-per-watt, an architecture that carries cleanly across PCIe generations, a Gen6 part already unveiled, and independence from any single NAND vendor. Where it is exposed: scale and balance-sheet depth versus Marvell; customer concentration; and a governance/listing overhang that no competitor carries.

Market-cap figures are approximate, move daily, and are provided only as peer-size reference as of June 2026.

Section 6: Industry

The demand driver behind FADU is the single biggest spending wave in technology right now: AI data-centre build-out. Training and serving large AI models requires feeding enormous, fast, hot data to GPUs, and that data increasingly lives on enterprise NVMe SSDs. Every AI server rack needs high-performance, high-capacity, power-efficient storage, and every one of those drives needs a controller. As hyperscalers pour capex into AI infrastructure, the enterprise SSD - and therefore the enterprise SSD controller - is a direct beneficiary. This is the tailwind FADU is riding, and it is the reason a company that nearly died in 2023 returned to profit by Q1 2026.

The market structure has two relevant technology cycles layered on the demand wave. First, the PCIe generation transition: the industry is in the thick of the Gen4-to-Gen5 shift now, with Gen6 next. Each transition is a chance to win or lose sockets and tends to favour suppliers who arrive early with a working part - which is why FADU's Sierra Gen6 unveil matters strategically beyond its near-term revenue. Second, relentless capacity growth (drives scaling toward hundreds of terabytes) and an unforgiving focus on power efficiency, because data-centre power is the gating resource.

On market size: the SSD controller market overall is a multi-billion-dollar market with a concentrated vendor base, the top handful of suppliers taking the clear majority of revenue, and analysts project continued growth driven by data-centre and AI demand (Technavio, Research and Markets, and others cover the enterprise SSD controller segment specifically). FADU plays only in the enterprise/data-centre slice, the highest-value and most defensible part of that market.

Where FADU sits in the supply chain: it is upstream of the drive but downstream of the foundry and the NAND maker. NAND makers supply the memory; FADU supplies the controller IP/silicon; drive vendors (often the NAND makers themselves, or WD/SanDisk) assemble the finished SSD; hyperscalers buy the drive. FADU's leverage point is that the controller is the part of the drive that most determines performance-per-watt, giving a small company outsized influence over a strategically important component.

Regulation here is less about product approval and more about two things specific to FADU's situation: export-control and geopolitics around advanced semiconductors and AI infrastructure (which shape where and to whom leading-edge storage can be sold), and Korean capital-markets regulation, which after the FADU IPO fiasco the FSS used as a case study to overhaul the domestic IPO disclosure process. The latter is less an industry factor than a company-specific overhang (covered in risks).

Cyclicality: the broader memory/storage industry is famously cyclical, swinging with NAND pricing and capex cycles. But the AI-driven enterprise demand FADU serves has, so far, behaved more like a secular up-cycle than a normal commodity swing. The risk is that this is still a capex cycle - hyperscaler spending can pause, drive qualifications can slip, and order flow (as FADU itself proved in 2023) can stop abruptly. Tailwinds: AI capex, the Gen5/Gen6 transition, the power-efficiency premium. Headwinds: the cyclicality of NAND/storage spend, and concentration of demand in a few hyperscalers whose buying is lumpy.

Section 7: Growth triggers

Drawn from FADU's six most recent quarterly earnings releases and management briefings (its functional concall record; see the source note at the top).

• Adding new hyperscaler customers in 2026. Management has stated FADU is confirmed for mass-production supply to two of the four global hyperscalers and expects to add more during 2026, with that expansion underpinning a step-change in revenue from Q1 2026 onward. (Q4 FY2025 release, Feb 2026; reiterated Q1 FY2026 release, May 2026 - a repeated trigger.)

  "In 2026, FADU expects to add multiple global hyperscaler customers, driving a substantial increase in revenue beginning in the first quarter." (paraphrased from the Q4 FY2025 / early-2026 management commentary)

• Sierra Gen6 controller launch through 2026. The PCIe Gen6 Sierra FC6161 (28.5 GB/s, ~6.9M IOPS, sub-9W, up to 512TB) was unveiled at FMS 2025; management guides commercial ramp through 2026, positioning FADU ahead of the Gen6 transition. (Q3 FY2025 and Q4 FY2025 releases, Nov 2025 / Feb 2026.)

• Gen5 (ECHO) mass-production ramp. ECHO entered mass production from late 2024 and Gen5 controller revenue has risen steadily as the Gen5 enterprise market opened - cited repeatedly as the proximate revenue driver. (Q2 FY2025 through Q1 FY2026 releases - a repeated trigger across at least four periods.)

• Record order backlog converting to revenue. Cumulative new orders by April 2026 had already surpassed full-year 2025 revenue, and management flagged four consecutive months of large orders into early 2026 as forward visibility. (Q1 FY2026 release, May 2026.)

  "As of April, cumulative new orders have already exceeded the whole of last year's revenue." (paraphrased from the Q1 FY2026 release)

• Mix shift toward higher-margin controllers. Controllers reached ~80% of revenue in Q1 2026 and management expects the controller share to keep rising, structurally improving profitability. (Q1 FY2026 release, May 2026.)

• Integrated DRAM PMIC module in Gen6. Management announced plans to integrate DRAM power-management (PMIC) into the module, a feature competitors still supply separately, as a differentiator for the Gen6 generation. (Late-2025 management briefing.)

• New SSD-module customers in Asia. Addition of new module customers in the Asian market was cited as supporting certain 2025 quarters. (Q2 FY2025 / H1 release, Aug 2025.)

Section 8: Key risks

Delisting / capital-markets-fraud overhang (high probability of continued overhang, potentially catastrophic tail). This is the defining company-specific risk. The Korea Exchange has launched a substantive listing-eligibility review after finding the IPO documents contained false statements or material omissions, and executives plus the company have been indicted under the Capital Markets Act for concealing the pre-IPO suspension of major-customer orders (SK hynix, SpaceX) while projecting ~KRW 120 billion of 2023 revenue against actual post-listing quarterly revenue of tens to hundreds of millions of won. A class action by investors is ongoing. The mechanism of harm is twofold: a tail risk that the shares are ultimately delisted, and a persistent governance discount and management-distraction drag even if they are not. Management's public posture - that "technology, not sales, is what matters" and a denial of the charges - does not resolve the legal exposure. This risk is unique to FADU; no competitor carries anything like it.

Customer concentration (high probability of volatility, moderate-to-severe impact). A handful of NAND/drive partners and, behind them, a few hyperscalers drive nearly all demand. The 2023 collapse is the proof of concept for how violently this can hurt: when orders were suspended, revenue went to near zero for two quarters. Lumpy, order-driven revenue means any single large customer pausing or delaying a drive qualification can swing a year. FADU has acknowledged the lumpiness implicitly by emphasising order backlog rather than smooth recurring revenue.

Competition from Marvell and from captive in-house controllers (moderate probability, structural). Marvell's scale and incumbency at the high end, and the ever-present option for a NAND maker to use its own controller (SK hynix being both customer and in-house controller owner), cap FADU's pricing power and socket count. The mechanism: at each PCIe generation transition, a customer re-evaluates, and a better-resourced rival or an internal team can take the socket. FADU's defence is performance-per-watt and architectural carry-forward, but it is fighting from a smaller balance sheet.

Capex cyclicality of AI/storage demand (moderate probability, moderate impact). FADU's recovery is built on the AI data-centre build-out. If hyperscaler capex pauses or the NAND cycle turns, the enterprise SSD demand that drives FADU softens. The current secular framing could revert to ordinary cyclicality. Management's own emphasis on Gen6 R&D timing (the ~KRW 11 billion charge that kept 2025 loss-making) shows the business still front-loads heavy investment against demand it must then realise.

Technology-roadmap execution and R&D burn (moderate probability, moderate impact). FADU's edge depends on shipping each new generation on time and at the promised power efficiency. Heavy Gen6 development costs kept full-year 2025 in operating loss despite revenue more than doubling. A slip in Sierra's qualification, or a failure to hit its claimed efficiency in customer testing, would erode the core differentiation and the 2026 revenue case simultaneously. Management's decision to freeze the EEUM/CXL effort shows R&D discipline, but also that the company must husband resources.

Section 9: Walk the talk

The six reporting periods used here are: Q1 FY2026 (released ~12 May 2026), Q4/FY2025 (~Feb 2026), Q3 FY2025 (~Nov 2025), Q2/H1 FY2025 (~Aug 2025), Q1 FY2025 (~May 2025), and Q4/FY2024 (~Feb 2025). The most recent is well within 90 days of today. FADU does not publish verbatim English transcripts, so this assessment is built from its quarterly earnings releases and management briefings across those six periods.

The honest starting point is that FADU's management carries a credibility deficit that no other company in this coverage universe carries: the entire reason it is under a delisting review is that the pre-IPO disclosures were found to contain false statements about customer orders and revenue. Any "walk the talk" assessment has to be read against that backdrop. What follows judges the post-scandal management record - the period in which the company has been trying to rebuild trust through delivery rather than promises.

On that narrower test, the recent record is markedly better than the founding-era record. Through FY2024 and into early FY2025, management's central message was that the ECHO Gen5 controller would enter mass production and drive a revenue recovery off the near-zero 2023 base. That was delivered: ECHO did ramp from late 2024, and FY2024 revenue roughly doubled off the trough, with FY2025 revenue more than doubling again to a company record. The directional promise - "Gen5 mass production restores the revenue trajectory" - was kept, and it was kept across multiple consecutive quarters of rising revenue (Q1 2025 through Q4 2025 each built on the prior).

Where management was candid rather than promotional: through 2025 they repeatedly told investors that the company would stay loss-making at the operating line because of front-loaded Gen6 R&D, and they quantified it (the ~KRW 11 billion final Gen6 development charge in Q4 2025). That is the opposite of the 2023 pattern - guiding down on profitability and explaining why, rather than papering over it. They then guided that 2026 would inflect to profit as Gen5 scaled and the controller mix rose.

Management framed 2026 as the year FADU "enters a profit phase from the first quarter, with substantial revenue expansion as multiple hyperscaler customers are added." (paraphrased, Q4 FY2025 release)

That specific, datable, near-term promise was then tested immediately - and met. Q1 FY2026 delivered the guided turn: a return to operating and net profit, controllers at roughly 80% of revenue (the mix shift management had promised would drive margins), and a step-up in revenue. Crucially, management's claim that cumulative orders by April had already surpassed all of 2025's revenue is a falsifiable, specific statement of the kind the company was caught misrepresenting in 2023 - and this time it lined up with reported results. Delivering the exact quarter they pointed to, on the exact metrics they named, is the single strongest piece of rehabilitated credibility in the file.

The discipline cuts the other way too, in a good sense: management said it would freeze the EEUM/CXL switch programme rather than keep spending into a market that had not arrived, and it did, openly explaining that NVIDIA's NVLink had pushed the CXL opportunity out. A management team that overpromises does not voluntarily shelve a flagship "future" project and tell investors the addressable market is not there yet.

The promise-vs-outcome pattern:

The plainly stated assessment: on the post-scandal record, this is management that has done largely what it said, including hitting a specific near-term profit-inflection quarter on the metrics it named, and that has shown unusual willingness to guide down and to shelve projects honestly. That earns real credit. But it must be weighed against the fact that the company is under a delisting review precisely because its founding-era disclosures were found to be false. The correct read is a management team rebuilding credibility through delivery, from a starting point of a serious, unresolved integrity failure. Recent execution is genuinely good; the trust deficit is genuinely large; both are true at once.

Section 10: Shareholder friendliness index

Dividends. FADU has not paid a dividend in any of the last three financial years (FY2023, FY2024, FY2025). This is unsurprising and not a knock: the company was loss-making at the operating line through FY2025 as it absorbed heavy Gen6 R&D, only returning to profit in Q1 FY2026. A pre-profit fabless semiconductor company reinvesting every won into the next controller generation is doing the right thing for shareholders by not paying out. There is no dividend trend to assess because there has been no dividend.

Buybacks and dilution. Two windows must be stated separately. For the trailing ~90 days, the MoatMap database recorded zero buyback activity by FADU (window since 2026-03-23). For anything older, a search of the company's disclosures and Korean financial coverage turned up no announced or executed share-repurchase programme over FY2023-FY2025 either; FADU has not been buying back stock. On the other side of the ledger, the share count has more likely drifted upward than down over the period: the August 2023 IPO issued new shares, and as a venture-backed, R&D-heavy company FADU uses equity-linked compensation (the insider activity in Section 11 is executives selling shares they hold). There is no evidence of share count shrinking; if anything it has grown modestly from the IPO and equity compensation. No buyback claim of any kind can be made in FADU's favour for any window in the last three years.

Verdict: Hoards Capital (by necessity, not greed). FADU returns no capital - no dividend, no buyback - because until Q1 2026 it had no profit to return and is plowing everything into the controller roadmap that is the entire investment case. This is appropriate for the stage, but as a pure capital-return screen the company scores at the bottom.

Section 11: Insider activities

Source: MoatMap cross-market disclosure database (Korea/DART feed), the canonical source for this gated venue. The most recent scrape is flagged 41 hours stale (last refresh 2026-06-19 23:00 UTC), so filings from the last day or two may be missing. All transactions below are officer dealings reported via DART's 임원·주요주주 소유상황보고서 (Officer & Major Shareholder Securities Holdings Report).

Recent transactions (most recent first):

Buys - read the signal. There were no insider purchases in the last 12 months. Zero buys. There is therefore no bullish conviction signal to flag here; the most powerful tell in this section (open-market insider buying) is absent.

Sells - work out the why. Five executives (four 상무-level, one 전무-level) sold shares in a tight four-week window from 8 May to 2 June 2026, immediately after FADU reported its strong Q1 FY2026 results (~12 May) and as the stock traded around the KRW 92,000-112,000 range. The reasons are not disclosed in the filings, so this is inference rather than fact: the timing - a cluster of officer sales right after a blow-out result that returned the company to profit and lifted the share price - is most consistent with executives realising gains and diversifying after a strong post-results rally, the kind of equity-compensation monetisation common after a recovery. None of the sales is individually large relative to shares outstanding (each is 0.02% of the company or less), and the dollar amounts are modest for senior executives (the largest, Woo Yeong-jae's, was under US$1m). There is no disclosed 10b5-1-style pre-set plan in this venue, and no disclosed estate, gift, or block-trade explanation. Read honestly: reason not formally disclosed; pattern most consistent with post-rally profit-taking.

Net assessment. Insiders were net sellers over the window, with activity broad-based across five different executives rather than concentrated in one - which slightly raises the noise but the sizes are individually small and the timing (right after a strong result and price recovery) points to ordinary diversification rather than a signal about the business. There were no offsetting buys, and notably no purchase by the CEO or any founder, which would have been the strongest possible vote of confidence given the delisting overhang. The cluster of small officer sells is a mild concern rather than a red flag: it is the kind of selling that follows a good quarter and a rallied stock, not panic exit-sized blocks, but the complete absence of any insider buying - in a company whose entire equity story hinges on rehabilitating trust - means there is no bullish counter-signal to lean on.

Section 12: Scenarios

Bull case. FADU's architecture and power-efficiency lead prove durable across the PCIe generation transition. The two confirmed hyperscaler design wins expand to three or four through 2026 and 2027, exactly as management has guided, and the record order backlog converts cleanly into rising controller revenue. The Sierra Gen6 controller qualifies on schedule, hits its claimed 2x efficiency gain in customer testing, and arrives ahead of competitors as the Gen6 enterprise cycle begins - so FADU, for once, is the early supplier into a transition rather than a challenger fighting an incumbent. The controller mix keeps climbing past 80%, profitability compounds as the heavy Gen6 R&D rolls off, and the AI data-centre storage wave stays in its secular up-phase. Critically, the legal and delisting overhang resolves without a delisting - the listing-eligibility review concludes that the post-scandal company is viable, the governance discount narrows, and FADU re-rates from a tainted recovery story into a credible independent enterprise-controller franchise. In this world, FADU is the one independent controller designer the hyperscalers trust to stay ahead on watts, and SK hynix and WD/SanDisk are just the first two of many drives carrying its silicon into AI racks.

Base case. Management delivers roughly what it has guided. Gen5 (ECHO) continues to ramp, the controller mix stays favourable, and FADU sustains the profitability it reached in Q1 2026 across the year, with lumpy, order-driven quarters rather than a smooth line. Sierra Gen6 launches through 2026 but ramps gradually, with real revenue contribution arriving later than the unveil excitement implies. FADU adds perhaps one more meaningful customer relationship but does not sweep all four hyperscalers; Marvell remains the high-end incumbent and the NAND makers keep optionality on in-house controllers, capping FADU's pricing power. The CXL/EEUM bet stays frozen and contributes nothing. The delisting review grinds on as an unresolved overhang that keeps a discount on the equity but does not end in delisting. FADU ends the period as a real but mid-scale independent controller supplier - genuinely profitable, genuinely validated, but still concentrated in a few customers and still living under a legal cloud.

Bear case. The order lumpiness that defines this business turns against FADU. A hyperscaler delays or re-sources a drive qualification, a Gen6 slip lets Marvell or a NAND maker's in-house team take the socket FADU was counting on, and the AI-storage demand wave cools from secular to cyclical as hyperscaler capex pauses. Revenue, which has always arrived in clusters, gaps down the way it did in 2023 - the structural concentration risk re-expressing itself. Heavy R&D commitments made against expected demand strand the company back in losses. And the worst tail: the Korea Exchange listing-eligibility review concludes against FADU and the shares face delisting, or the criminal case and class action produce penalties and management upheaval that gut the company's ability to execute. In this scenario the technology may still be good, but the combination of a single customer pulling back, a missed generation, and an unresolved governance failure proves that good silicon is not enough when the customer base is this narrow and the integrity overhang is this deep.