Super Micro Computer

Phase A — Understand the business

Company Overview

Super Micro is a San Jose-based integrator of GPU-dense, rack-scale server systems. It designs, builds, and configures complete servers and storage — increasingly liquid-cooled AI racks built around NVIDIA, AMD, and Intel silicon — and ships them as turn-key "Total IT Solutions" to enterprise data centers, cloud providers, and AI builders . The differentiator it claims is the **Server Building Block Solutions®** modular architecture: a common library of subsystems that lets it be *first to market* on each new GPU/CPU generation, and the fact that it is "the only major server, storage, and accelerated compute platform vendor that designs, develops, and manufactures a significant portion of their systems in the United States" (San Jose), with assembly also in Taiwan, the Netherlands, and a new Malaysia plant .

The model is, in plain terms, value-added assembly: buy GPUs/CPUs/memory/SSDs (which the 10-KForm 10-KA company’s audited annual report to the US regulator. The most complete thing it publishes. explicitly says "have represented an increasing portion of our cost of sales"), integrate them into optimized racks with proprietary cooling and management, and sell the system. The bill of materials — above all the GPU — is the dominant cost, which structurally caps gross margin.

Customers: Over 1,000 customers in 100+ countries, but the revenue is brutally concentrated. Four customers each exceeded 10% of net sales in FY2025; in FY2024 it was one; in FY2023 none . In Q3 FY26, the top customer alone was 27.0% of net sales . This is a hyperscaler-driven order book.

Suppliers / contract structure: No long-term supply contracts for most critical components — "we often purchase these materials and components on a purchase order basis" . Payment terms to customers are 30–60 days, occasionally longer for large customers, and the company is now factoring receivables and using "financing arrangements with a customer" to fund the working-capital gap . Two related-party Taiwanese firms — Ablecom (contract manufacturing/chassis, warehousing) and Compuware (distribution, power supplies, contract manufacturing) — are embedded in the supply chain (detailed in Lens 2 and Lens 9).

Supply Chain

Map: upstream silicon & components → contract manufacturing / in-house assembly → SMCI integration → hyperscaler/AI end customer.

• Upstream — the binding input is the NVIDIA GPU. The 10-K names the vendors whose roadmaps it tracks: "NVIDIA Corporation, Intel Corporation, Advanced Micro Devices, Inc., Broadcom Inc., Samsung Electronics, Micron Technology" ``. GPU allocation from NVIDIA is the single most important chokepoint — SMCI's growth is gated by how many Hopper/Blackwell GPUs it can secure. Memory (Samsung/Micron/SK Hynix HBMHigh-bandwidth memoryMemory stacked in layers beside a processor so data reaches it faster. The usual bottleneck in AI chips.), SSDs, chassis, power supplies, CPUs are the rest of the BOM. None are under long-term contract — purchase-order basis only.
• Contract manufacturing — related-party dependency. Ablecom is the primary contract manufacturer for chassis and holds exclusive supplier rights on specific co-developed products; Compuware supplies power supplies and acts as a non-exclusive distributor in Taiwan/China/Australia . SMCI's purchases from the two were 3.3% of COGS in FY25 (down from 6.6% in FY23). Both are majority-dependent on SMCI for their own revenue — and both are building campuses next to SMCI's Malaysia plant . This is a single-source dependency the 10-K flags as a named risk.
• In-house assembly: San Jose (US), Taiwan, Netherlands, Malaysia (new in FY25). US manufacturing is a genuine differentiator for export-sensitive / sovereign-AI customers — and, ironically, the very export-control compliance that the DOJ indictment (Lens 10) alleges was circumvented.
• Downstream — concentrated hyperscaler/AI buyers. The end customers are large cloud and AI infrastructure operators (the four >10% names). The product is the NVIDIA GB200/GB300 NVL72 liquid-cooled rack — Supermicro's GB200 NVL72 packs 72 Blackwell GPUs + 36 Grace CPUs with integrated direct-liquid-cooling (DLC-2, ~98% heat capture) ``.

Chokepoints: (1) NVIDIA GPU allocation — the hard ceiling; (2) HBM supply (Samsung/Micron/SK Hynix); (3) Ablecom exclusivity on certain chassis; (4) data-center "readiness" at the customer end — Q3 FY26 revenue was explicitly held back by "data center and customer readiness together with industry-wide supply chain constraints" ``.

Competitive Advantages (moats)

Be honest: the moat here is narrow and contested, and it just got narrower.

• What SMCI genuinely has: (1) Time-to-market — the building-block architecture lets it ship new-GPU racks fast, historically ahead of Dell/HPE; (2) Liquid-cooling depth — datacenter-scale DLC-2 is a real engineering lead as racks cross 120-140kW ``; (3) US manufacturing — unique among majors, valuable for sovereign/regulated buyers; (4) Configurability — the broadest SKU catalog in the industry.
• What it does NOT have: durable pricing power. The product is GPU-led integration — the value the customer pays for is mostly NVIDIA's, not SMCI's. The 10-K is explicit that gross margin fell because of "our strategy to offer competitive pricing to gain market share" ``. A moat that has to be defended by cutting price is not a moat; it is a share grab in a commoditizing layer.
• Bargaining power: Weak on both sides. Upstream, NVIDIA holds the whip (allocation). Downstream, four hyperscalers can dictate terms, demand longer payment, and — critically — switch to Dell. The competitive narrative has flipped in 2026: "Dell is preferred, HPE is acceptable, and SMCI remains in the penalty box," with Dell carrying a ~$43B AI-server backlog and stronger margin discipline (Dell GM ~21.6% vs SMCI ~10%) . On 2026-03-20, "Super Micro Craters 27%, Dell Rises 5%: The AI Server Market Just Got a New Front-Runner" .
• Verdict on moat: Real engineering competence (cooling, speed, US build) but no pricing moat and a structurally commoditizing position between an all-powerful supplier (NVIDIA) and powerful concentrated buyers. The governance overhang (Lens 10) actively erodes the one soft moat that mattered — trust from large enterprise/sovereign buyers.

Segments

SMCI reports a single operating segment — "we operate in one operating segment that develops and provides high-performance server solutions" with the CEO as CODM reviewing consolidated net income ``. So the only meaningful breakouts are product type and geography.

By product type (FY, $M) ``:

This is effectively a one-line business: GPU/AI server systems. Subsystems are shrinking (-17.9% in FY25) as the company deliberately concentrates on full systems. Within systems, "GPU & Super Racks" grew $5,804.0M or 52% YoY in FY25, on a +34% average selling price (H200/H100/B200), with services & software a small recurring sliver (+$102.2M YoY) . In Q3 FY26, AI-GPU products grew **+150.5% YoY** and non-AI categories *shrank 34.9%* over nine months — the mix is going all-AI .

By geography (% of net sales) ``:

Trend & cause: Accelerating into AI systems, with non-US exposure rising in FY25 (Asia 19.4%→25.0%) then snapping back to US-led in Q3 FY26 (68.7%). The Asia growth is exactly the surface where the export-control indictment (Lens 10) lands — ~$2.5B of servers allegedly diverted to China in 2024-2025, the precise period non-US mix was peaking.

---

Phase B — Measure performance

Earnings Result

Income statement (FY, as reported) ``:

The single most important fact about FY25: revenue grew +47% but net income FELL (-9%, $1,152.7M → $1,048.9M), and gross margin collapsed 270bps to 11.1% — a third of the way down from FY23's 18.0%. The cause is in the company's own words: "our strategy to offer competitive pricing to gain market share, change in product and customer mix, and higher manufacturing related expenses" ``. This is the financial signature of a commoditizing integrator buying share with margin.

The latest print — Q3 FY26 (qtr ended 2026-03-31) ``:

Nine months FY26: net sales $27,943.3M, net income $1,052.2M — already exceeding all of FY25 . The quarterly net-income arc: Q1 FY26 ~$168M → Q2 $400.6M → Q3 $483.4M .

The margin rollercoaster is the story of the year. Per the Q3 FY26 call: non-GAAP gross margin recovered to 10.1% in Q3, a 58% sequential jump from ~6.4% in Q2 FY26 . So the path is FY25 11.1% → **Q2 FY26 ~6.4% (collapse)** → Q3 FY26 ~10% (partial recovery). The Q2 collapse drove the 2026-03-20 ~27% crash; the Q3 recovery drove a +24% after-hours pop on 2026-05-05 .

Beat/miss in Q3 FY26: EPS beat, revenue missed. Non-GAAP EPS $0.84 vs ~$0.61 consensus (+37.7%), but revenue $10.2B missed the ~$12.39B expected (-17.7%) — the shortfall pushed to deferred revenue on "data center and customer readiness" . Sequentially, **revenue fell from ~$12.7B in Q2 FY26 to $10.2B in Q3** — a yellow flag on order timing/lumpiness .

Balance-sheet flags (FY25 year-end) ``:

• Cash $5,169.9M (up from $1,669.8M) — but funded by debt issuance, not operations.
• Inventory $4,680.4M — held ~flat YoY despite +47% revenue (~78 days of COGS ``).
• AR $2,203.9M — declined despite the revenue surge (a genuine positive vs the receivables-outrunning-revenue red flag); allowance for credit losses $0.
• Convertible debt $4,725M; total liabilities $7,716.6M; equity $6,301.9M.
• Operating cash flow turned negative over 9M FY26 ("net cash used in operating activities") as inventory + AR built; the gap was plugged by $4,010.2M of net new debt ``.

Guidance: Management guided Q4 FY26 EPS ~$0.62 and Q3 FY27 EPS ~$0.87 ``. Tone shifted constructive on margin recovery but defensive on the revenue push-out.

Earnings Calls (sentiment trend)

No transcripts on disk (transcripts=0); this lens is ``.

• Through FY24 (early 2024): euphoric — "AI everywhere," DLC leadership, raising long-term revenue targets; the stock ran to a split-adjusted ~$2,884 rebased peak.
• Late 2024 (the blackout): the narrative broke. After the 2024-08-27 Hindenburg report and EY's 2024-10-24 resignation, SMCI went delinquent — no timely 10-K/10-Qs — and management spent calls on remediation, the Nasdaq exception, and re-assertions of integrity rather than operations ``.
• 2025 (re-baselining): tone = "we're past it, demand is intact, Blackwell ramp coming." Heavy emphasis on order book and capacity (Malaysia).
• Q2 FY26 call (Feb 2026): the margin collapse to ~6.4% forced a defensive, explain-the-mix posture.
• Q3 FY26 call (2026-05-05): management leaned hard into the margin recovery to ~10.1% and the $39B order pipeline, framing the revenue miss as timing (deferred to future quarters) ``.

Phrases they keep saying: "Building Block Solutions," "first to market," "DLC/liquid cooling leadership," "design wins," "data center building blocks at rack scale." What they stopped saying: the swaggering multi-year revenue targets of the 2024 peak. The recurring tell across calls is demand is never the problem — margin, supply, and "customer readiness" are. That is consistent with a price-taking integrator.

Comps

Peer set from the company's own FY2025 compensation peer group (Marvell, Micron, HPE, HP, Jabil, Sanmina, Western Digital, NetApp, CDW) `` plus the obvious direct rival Dell.

Read: SMCI looks cheapest on forward P/E (~10-13x) but most expensive on EV/EBITDA (~22.7x) — the divergence is the tell. The P/E rests on a forward EPS that consensus has flat YoY (FY26 ~$1.70 vs FY25 ~$1.72) while EV/EBITDA captures the reality that ~10% gross margins crush EBITDA ``. Against Dell's 21.6% GM and HPE's cleaner book, SMCI's "cheapness" is a discount for margin fragility and governance risk, not a bargain. Do not anchor a bull case on the low P/E — it is low for reasons.

Stock-Price Catalysts (>5% moves, ~5yr)

The five-year cumulative-return table in the 10-K (rebased $100 at 6/30/2020) ``:

• FY23 (→$877): the AI-server breakout — GPU-rack demand inflected. The defining up-leg.
• FY24 (→$2,885): mania peak; included the 10:1 forward split (effective 2024-10-01) and S&P 500 inclusion hype ``.
• 2024-08-27: Hindenburg short report (accounting manipulation, sibling self-dealing, sanctions evasion) → sharp drop ``.
• 2024-10-24 → 2025-02: EY resignation, delinquency, near-delisting → deep drawdown (the FY25 close at 1,726 is down ~40% from the FY24 peak).
• 2026-03-19/20: DOJ indictment of co-founder/director Wally Liaw — stock ~-33% (-$10.26/share), the largest single event ; same window Q2 FY26 margin collapse to ~6.4% (-27% on 2026-03-20) .
• 2026-05-05: Q3 FY26 — EPS beat / margin recovery → +24% after-hours ``.
• 2026-06-09/11: $7B equity raise announced/priced → -26% over two days ``.

What the tape reacts to: for most names it's earnings; for SMCI it is overwhelmingly governance and financing events — auditor resignations, DOJ actions, dilutive raises — layered on a high-beta AI-demand substrate. The market treats SMCI as a credibility stock first and a fundamentals stock second. That is the single most important behavioral fact for sizing any position.

---

Phase C — Judge people & books

Management

• Charles Liang — Founder, President, CEO, Chairman. Founded SMCI in 1993; built it from a motherboard maker into a ~$22B AI-server integrator. Beneficially owns ~13.5% (~66.7-82.8M shares, ~$2.0B); co-founder/spouse Sara Liu owns ~12.5% — combined founder-family stake ~26%, insiders ~28% total ``. Skin in the game: maximal. This is a true founder-operator with enormous personal stake — a stabilizer for strategy continuity, but also a key-man risk (the 10-K names losing Liang as a specific risk factor) and a governance-concentration risk (founder-controlled board was the exact subject of EY's independence concerns).
• Track record: genuinely impressive on growth/time-to-market; deeply blemished on controls and disclosure. SMCI has already been through one SEC accounting saga — a 2020 SEC settlement over premature revenue recognition (2014-2017), in which the company paid $17.5M and Liang personally reimbursed ~$2.1M without admitting/denying ``. The 2024-2026 cycle is, in part, history rhyming.
• Capital allocation: Reinvests aggressively into inventory/capacity; no dividend, no consistent buyback (a one-off $200M repurchase at $40.89 in June 2025 was concurrent with the 2030 convert placement — a structuring move, not a return-of-capital signal) ``. ROE optically high but on thin and now-diluting margins. The 2028/2030 converts + the June 2026 $7B raise mean the equity has been the funding shock-absorber.
• Red flags (this is where it stacks): related-party manufacturing with the CEO's family (Liang + Liu own ~10.5% of Ablecom; Liang's brother Steve Liang owns ~35% of Ablecom and sits on Compuware's board) ; a thin senior bench for a hyperscale company; and the board-independence questions EY raised. The bench was *forced* to deepen in FY25 — new CFO-adjacent hires, promoted controller to **Chief Accounting Officer**, hired a VP of technical accounting/external reporting .
• Archetype: classic founder-operator who outran his control environment. The implication for this stage: the growth instinct is real, but the company has repeatedly shown it cannot be trusted to build the controls and governance that a $20B+ public company requires — and the market has now twice repriced that.

Forensic Red Flags

This is the lens that defines SMCI. Act as a forensic analyst — the file is thick.

Accounting-statement risks:

• Gross-margin volatility (11.1% FY25 → ~6.4% Q2 FY26 → ~10% Q3 FY26) `` — extreme swings in a low-margin integrator make a single quarter's "beat" low-signal; the margin is a function of mix and price concessions management does not fully control.
• Inventory $4.68B held flat on +47% revenue with rising write-downs — $239.3M of inventory valuation write-downs over 9M FY26 (+50% YoY), explicitly for "excess and obsolete inventory with either insufficient demand or reduced net realizable value" ``. In a fast-moving-GPU business, obsolescence risk is real and recurring.
• AR allowance = $0 at FY25 year-end despite a $2.2B receivable concentrated in four customers ``. Defensible if those are hyperscalers, but a zero allowance on that concentration is aggressive.
• Operating cash flow negative over 9M FY26, plugged by $4.0B of new debt and receivables factoring (new factoring fees appearing) `` — earnings are real but cash conversion is poor and getting worse as the business scales working capital faster than it generates cash.
• Tariff exposure as a new structural cost: tariff expense +608% in Q3 FY26 (+$89.4M) and +891% over 9M (+$273.6M) `` — a margin headwind outside management's control.
• Deferred revenue / revenue-timing: Q3 FY26 revenue was pushed out (the miss) into deferred revenue — and revenue timing is the precise lever SMCI was sanctioned for in 2020. Worth watching, not yet a finding.

Internal-control & disclosure (the core):

• EY engaged Mar 2023; resigned 2024-10-24 mid-audit (their first SMCI audit, no report ever issued). EY's resignation letter, quoted in the 10-K: "we are resigning due to information that has recently come to our attention which has led us to no longer be able to rely on management's and the Audit Committee's representations and to be unwilling to be associated with the financial statements prepared by management" — citing doubts on COSO Principle 1 (integrity) and Principle 2 (board independence) ``.
• Special Committee (Cooley LLP + forensic firm Secretariat Advisors) cleared management on 2024-12-02 — "the evidence reviewed... did not give rise to any substantial concerns about the integrity of our senior management or the Audit Committee" — and stated it did not believe EY's conclusions were supported by the facts ``. Note the structural conflict: a board-appointed committee clearing the board over the objection of the resigning auditor.
• BDO USA engaged 2024; issued a CLEAN opinion on FY25/FY24 ("present fairly, in all material respects") and Deloitte audited the FY23 comparative ``. So the financials are audited and clean as of FY25.
• BUT material weaknesses remain UNREMEDIATED as of 2025-06-30. Management concluded that ICFR was NOT effective and disclosure controls were NOT effective due to material weaknesses (IT general controls, segregation of duties, manual journal entries) ``. Remediation is underway (ERP SOD redesign, ITGC scope expansion, new CAO/VP, journal-entry controls) and they believe one weakness (journal-entry approval) is remediated — but the rest are not. A clean opinion with live material weaknesses means investors are relying on BDO's substantive testing, not on a functioning control system.
• Delinquency / near-delisting: failed to timely file FY24 10-K + two 10-Qs; Nasdaq granted an exception; all filed by 2025-02-25 ``.

Regulatory findings (required sub-section):

• SEC EDGAR EFTS (LR + AAER), 2021-06-20 → 2026-06-20: ZERO findings in the search window . (The 2020 SEC accounting settlement pre-dates this window — see Lens 9, .)
• 10-K Item 3 (Legal Proceedings) incorporates Note 13 "Commitments and Contingencies" by reference and gives no quantified specific litigation — but the entire Item 9 disclosure (auditor resignation, material weaknesses, delinquency) is the de facto legal-risk surface ``.
• Non-SEC enforcement — THE headline finding (web): On 2026-03-19, the U.S. Attorney for the Southern District of New York unsealed an indictment of three SMCI-associated individuals for an export-control conspiracy — diverting ~$2.5B of AI servers to China (2024-2025) in violation of U.S. export controls. Indicted: Yih-Shyan "Wally" Liaw (co-founder, SVP Business Development, board director); Ruei-Tsang "Steven" Chang (Taiwan sales manager); Ting-Wei "Willy" Sun (contractor) . **Supermicro itself is NOT a defendant**; it placed the two employees on administrative leave, terminated the contractor, and **Liaw resigned from the board on 2026-03-20**; an independent board investigation was launched . This substantiates the sanctions-evasion strand of the August 2024 Hindenburg report — the most serious of the short thesis's claims is now a federal indictment.
• Securities class action: Bhuva v. Super Micro Computer, No. 26-cv-02606 (N.D. Cal.), class period 2024-04-30 to 2026-03-19, alleging the company concealed export-control violations and internal-control weaknesses ``. Multiple firms soliciting lead plaintiff.
• DOJ + SEC subpoenas (document requests) received late 2024 following the short report; company cooperating; unresolved ``.

Conclusion: Financials are currently audited-clean, but the company carries (1) live, unremediated SOX material weaknesses, (2) a co-founder/director under federal criminal indictment for export-control violations, (3) an active securities class action, (4) open DOJ/SEC investigations, and (5) a history of a prior SEC accounting settlement. This is the densest forensic-risk stack in large-cap AI hardware.

---

Phase D — Project & stress-test

Forward Projection

Build from actuals + guidance. FY ends June 30. 9M FY26 net income is already $1,052.2M; management guided Q4 FY26 EPS ~$0.62 ``.

FY2026 (base): 9M EPS ≈ $1.08 (Q1 ~$0.25 + Q2 ~$0.60 + Q3 $0.72, share-count-adjusted) + Q4 guide ~$0.62 → **~$1.70** — squares with consensus FY26 ~$1.70. Revenue ~$41B ``. Note: this is BEFORE the dilution from the June 2026 $7B raise, which lands in FY27+ and lowers per-share figures.

Consensus sits at FY26 ~$1.70 and FY27 ~$2.54 (+49%) — i.e., consensus is roughly my base→bull on FY27, pricing a real margin recovery + order conversion. **The swing variable is gross margin, not revenue.** A 2-point GM move on ~$45B revenue is ~$900M of gross profit — that single line determines whether FY27 EPS is $1.60 or $2.90. Dilution from the mandatory-convertible preferred (converts ~June 2029) suppresses FY27-28 per-share figures regardless.

No our model forecast logged (per --watchlist rule — breadth mode produces dossiers only). If logged, the base call would be: "SMCI FY26 non-GAAP EPS >= $1.70, p=0.55, resolves 2026-06-30."

Bull vs Bear

Bull case. Demand is not the question — it is overwhelming. SMCI sits on ~$39B of recent AI-server orders from 20+ customers , is a leader in the one thing the Blackwell/Rubin era *requires* (datacenter-scale direct-liquid cooling at 120-140kW/rack), and is first-to-market on each NVIDIA generation. Blackwell is sold out into mid-2026 industry-wide . If margins normalize to even 11-12% and the order book converts, FY27 EPS of ~$2.50-2.90 against a ~$30 stock is a ~10-12x P/E on a hyper-growth AI-infrastructure name — optically cheap. The $7B raise, however painful, funds the working capital to capture the orders (management's "build for 2027" framing). US manufacturing is a structural edge for sovereign/regulated AI. The contrarian bull: the market is so fixated on governance that it is ignoring a genuine demand super-cycle the company is uniquely equipped to serve.

Bear case (2-3 permanent-impairment risks).

1. The margin is the business model, and it's ~half of Dell's. ~10% gross margin in a layer where the GPU is the value means SMCI is a thin-margin integrator buying share with price. If Dell (21.6% GM, $43B backlog) decides to compete harder on the rack, SMCI has no margin cushion to defend with ``. This is structural, not cyclical.
2. Governance is a permanent discount, not a one-off. A co-founder/director federally indicted for diverting $2.5B of servers to China , on top of unremediated SOX material weaknesses , an EY integrity-driven resignation, an active class action, and a prior 2020 SEC settlement — this is a pattern. Sovereign/enterprise buyers who care about export compliance now have a reason to choose Dell. Trust, once a soft moat, is now a liability.
3. Dilution + cash burn. Operating cash flow is negative; the company is funding growth with $4B+ of debt and a $7B equity raise that dilutes existing holders meaningfully ``. The per-share growth story is being taxed by the share count required to fund it.

Pre-mortem (18 months out, thesis broke). It's December 2027. SMCI's stock is down another 40%. What happened: the DOJ matter widened (the company itself, not just individuals, drew a deferred-prosecution agreement or a fine); a hyperscaler customer shifted its rack business to Dell, and with four customers at ~75% of revenue, the loss of one cratered growth; gross margin never sustainably cleared 10% because every Blackwell-rack win was priced to the bone; and the mandatory-convertible preferred converted into a much larger share count, so even decent net income produced mediocre EPS. The order book was real — but it converted at margins that didn't reward shareholders.

Are multiples too high? On EV/EBITDA (~22.7x) yes; on P/E (~10-13x) no — but the low P/E is a governance + margin discount, correctly applied. The market is not mispricing SMCI cheap; it is pricing real, specific risks.

Contrarian view (what the market refuses to see): The bull and bear are both right simultaneously — and the market keeps oscillating violently between them (±25-33% on single events) because it cannot hold two true things at once. The thing being missed is that this is not an investment, it is a trade on a binary: if the DOJ matter resolves cleanly (company never charged, individuals only) and margin holds ~10-11%, the stock re-rates hard off a depressed base. If the DOJ matter implicates the company, the equity is impaired regardless of the order book. There is no "own the compounder" middle path here — the distribution is bimodal.

Devil's Advocate (short-seller)

You are a short-seller dismantling the bull.

• The revenue is borrowed from the future and bought with price. SMCI "wins" by undercutting — its own 10-K says so. A backlog assembled at ~10% gross margin is not a quality backlog; it is volume rented from Dell at a discount the moment Dell wants it back ``.
• Concentration is a loaded gun. Four customers ≈ 75% of AR; top customer 27-32% of sales ``. Hyperscalers dual-source and squeeze. Lose one and the growth story inverts overnight — and they now have a governance reason to diversify away.
• The moat bulls cite is NVIDIA's, not SMCI's. Strip out the GPU and what's left is sheet metal, power supplies, cooling, and integration — replicable, and Dell/HPE/ODMs (Foxconn, Quanta, Wistron) all do it. DLC leadership is a 12-18 month lead, not a decade moat.
• The worst capital-allocation/governance moves are a pattern, not an accident: related-party manufacturing with the CEO's brother (Ablecom 35%-owned by Steve Liang) ; a 2020 SEC accounting settlement; an EY resignation over integrity; a board-appointed committee clearing the board; and now a co-founder/director indicted for export-control crimes . The kindest read is chronically weak controls; the unkind read is a culture that repeatedly skirts the rules to drive sales — which is exactly what the DOJ alleges ("all to drive sales and generate revenues in violation of U.S. law").
• Assumptions that must hold for ~$30: (1) DOJ never charges the entity; (2) margin sustainably ≥10%; (3) no major customer defection; (4) the $7B raise is the last dilution; (5) SOX weaknesses get remediated without a restatement. That is five independent things going right.
• If growth disappoints 20-30%: at ~10% gross margin, operating deleverage is vicious — opex doesn't flex down as fast as a thin gross profit, so a revenue miss compresses net margin disproportionately (already visible: FY25 net income fell on +47% revenue). EPS could halve on a 25% revenue shortfall ``.
• The single permanent-impairment scenario: DOJ/SEC concludes the company (not just individuals) bears responsibility for the export-control scheme → corporate penalty + export-privilege restrictions + hyperscaler flight. Plausibility: non-trivial given an indicted director and an active SEC investigation. That is the short's home run.

Management Questions (ordered by information value)

1. On the DOJ indictment: What specific controls failed to allow ~$2.5B of servers to reach China, were they the same controls flagged as SOX material weaknesses, and what is the company's exposure to a corporate charge or deferred-prosecution agreement?
2. When will all SOX material weaknesses be formally remediated and re-tested, and will FY26's 10-K carry a clean ICFR opinion with no live weaknesses?
3. What is the structural, sustainable gross-margin floor for the AI-rack business once mix and price concessions settle — and how does it get above 11% durably?
4. What share of revenue is the single largest customer today, and what concrete steps reduce four-customer ≈75% concentration over the next 8 quarters?
5. Why was a $7B equity raise (vs. more debt or vendor financing) the right funding choice, and is this the last equity dilution before the order book self-funds?
6. After the mandatory-convertible preferred converts in 2029, what is the fully-diluted share count, and what FY29 net income makes the dilution accretive?
7. How exposed is the order book to NVIDIA GPU allocation, and what is your committed Blackwell/Rubin allocation through CY2027?
8. With Dell carrying a $43B backlog and 21.6% gross margins, what stops them taking your hyperscaler share at a price you can't match?
9. What governance changes followed Liaw's resignation — board independence, related-party (Ablecom/Compuware) oversight, export-compliance leadership?
10. How much of the Q3 FY26 revenue miss is genuinely deferred (recognized in Q4/Q1) vs. lost, and what is the actual conversion timeline of the $39B order figure?
11. What is the realistic free-cash-flow path — when does operating cash flow turn durably positive given the inventory/AR working-capital intensity?
12. What is your tariff-mitigation strategy as tariff expense scales with revenue (+891% over 9M FY26)?
13. How do you defend the related-party manufacturing arrangements with Ablecom/Compuware given the self-dealing allegations — and is there a plan to unwind or arms-length them?
14. What inventory write-down run-rate should we expect, and how do you manage GPU-generation obsolescence (Hopper→Blackwell→Rubin) on a $4.7B inventory?
15. What does the credit-rating path look like — when do you expect investment grade (the revolver's IG trigger is no earlier than Sep 2026), and what does that do to financing cost?

---