Super Micro Computer, Inc. (NASDAQ: SMCI) — The AI Build-Out’s Box-Mover: Record Revenue, 8% Margins, and a Perpetual Need for Capital

Independent equity research. Published 2026-06-10. Fiscal year ends June 30.

⚡ Claude’s Take

This block is the author’s own independent opinion and general information only — it is not investment advice and not a recommendation to buy or sell any security. The analysis that follows takes no position and names no price target outside this block.

Verdict: AVOID at ~$40 — a genuinely booming business that is also a genuinely bad one. Great revenue story, commodity economics, chronic dilution, and an unresolved governance/DOJ overhang. I would only get interested in the low-to-high $20s (~10–13x a realistically diluted, maintained-margin EPS of ~$1.80–$2.00, ≈0.5x sales). Not a high-conviction short, because AI demand is real, the just-completed ~$7B raise removes the near-term liquidity gun, and 18% short interest invites squeezes. Conviction: medium.

The bull pitch — “an AI-infrastructure leader at 12x forward earnings and 0.8x sales” — is a sales-multiple optical illusion. Supermicro’s revenue went from $15B (FY24) to a ~$40B FY26 guide, yet gross profit dollars barely moved ($2.06B → ~$2.3B 9-mo) and net income fell in FY25 despite +47% revenue. That is the financial fingerprint of a price-taking assembler of someone else’s silicon, not a franchise: 8% gross margin is the absence of pricing power. The entire upside case rests on a margin re-rate (DCBBS, enterprise, sovereign mix) that management asserts but two years of results contradict. Meanwhile the model burned $7.6B of operating cash in nine months, inventory doubled to $11.1B, the cash hoard fell to $1.3B, and the company came to market with an emergency ~$7B equity-and-mandatory-convert raise — its tell that it cannot self-fund its own growth. Layer on an auditor that resigned in 2024, a recurrence of the 2018–2020 controls failures, a live SDNY export-control matter, and ~95% of chassis still sourced from a family-affiliated supplier, and you have a permanent, deserved risk discount.

My framing is value trap, not value. The market is pricing the commodity reality roughly correctly; the cheapness only materializes if you believe the unproven margin story. What flips me bullish: two-plus consecutive quarters of sustained double-digit gross margin while customer concentration falls and operating cash flow turns durably positive — proof the DCBBS/enterprise mix is real and the model can self-fund. What flips me more bearish: margins stuck at ~8% as enterprise mix climbs (confirming no pricing power), a restatement out of the independent investigation, or any wobble in NVIDIA allocation or AI capex. Catchy version: selling AI by the ton, at cost-plus — and passing the hat to do it.

1. Executive Summary

Super Micro Computer is the purest public proxy for the volume of the AI server build-out and one of the clearest cases of that volume failing to convert into value. Revenue compounded from $3.56B (FY21) to $14.99B (FY24), $21.97B (FY25), and a ~$40B FY26 guide, with AI GPU platforms now over 80% of sales (Q3 FY26 call, 2026-05-05). On the top line it is among the great hardware ramps of the cycle. On every other dimension that defines business quality, the picture is poor and deteriorating.

Gross margin is in structural collapse: 18.0% (FY23) → 13.8% (FY24) → 11.1% (FY25) → 8.2% (FY26 9-mo), with Q4 FY26 guided to 8.2–8.4%. The decisive datum is that gross-profit dollars went flat while revenue doubled (FY24 $2.06B → FY25 $2.43B → FY26 9-mo $2.28B), and net income actually fell in FY25 ($1,049M vs. $1,153M) despite revenue rising 47%. Forty-seven percent revenue growth produced zero operating leverage. This is the signature of a commoditized integrator passing through expensive third-party silicon (chiefly NVIDIA GPUs) at a thin assembly markup — not a business with pricing power.

The model consumes cash violently as it grows. Operating cash flow was −$2.49B (FY24), +$1.66B (FY25), and −$7.56B in the first nine months of FY26. Inventory swelled to $11.1B and receivables to $8.4B while cash fell from $5.2B to $1.3B, pushing the company to ~$7.5B net debt. On June 9–10, 2026 Supermicro announced a roughly $7B capital raise ($1.25B common + $3.75B mandatory convertible preferred + a $2.0B ATM) to fund working capital — an emergency recapitalization that validates the core finding: it cannot finance its own growth from operations.

There is no durable competitive advantage. Speed-to-market and direct-liquid-cooling are real but replicable execution edges, not barriers to entry; Supermicro is sub-scale versus Dell, HPE, Lenovo and the giant ODMs (Foxconn, Quanta, Wiwynn) who build the same NVIDIA reference racks; its market share proved unstable (ceded during the 2024–25 filing crisis); and a single customer was 63% of one quarter’s revenue. In Greenwald terms — no cost advantage, no demand captivity, no scale-plus-captivity flywheel. The 8% gross margin is the verdict.

The governance ledger is heavy. An auditor (Ernst & Young) resigned in October 2024; the FY2024 10-K was delinquent; the company narrowly avoided Nasdaq delisting (February 2025); this recurred a 2018 delisting and a 2020 SEC revenue-recognition settlement; a March 2026 SDNY indictment of associated individuals over alleged export-control violations is unresolved; and ~95% of chassis are still sourced from a Liang-family-affiliated supplier (Ablecom). None is individually fatal, but together they warrant a permanent risk premium and conservative treatment of every reported figure.

On valuation, the “cheap” multiple is an illusion. At ~$40.64 the equity is ~$24–27B and EV ~$30.5B. The ~0.7–0.8x sales multiple simply prices an ~8% gross / ~3.7% net margin; on earnings (~20x trailing, ~12x forward) and EV/EBITDA (~17x) it is not cheap for a low-quality, cyclical, diluting assembler, and it sits around the 70th percentile of its own valuation history. The forward multiple “works” only if a margin inflection appears that the trailing record denies. This report takes no position and sets no price target; it lays out the embedded expectations and the evidence for and against them.

2. Business Overview

Super Micro Computer, Inc. (“Supermicro,” NASDAQ: SMCI) designs, assembles, and sells server and storage systems and the subsystems that go into them. Incorporated in California in 1993, headquartered in San Jose, and public since 2007, the company built its identity on a modular “building-block” architecture — a library of interoperable motherboards, chassis, power, cooling, and networking components that can be configured into a wide range of “application-optimized” systems faster and at lower design cost than rivals (FY2025 10-K, filed 2025-08-28). That same modularity has now been scaled up one level: the current flagship offering is DCBBS — Data Center Building Block Solutions — a rack- and data-center-scale bundle that wraps the compute racks in direct-liquid-cooling (DLC) plumbing, power shelves, battery backup, networking, on-site deployment/consulting service, and the SuperCloud Composer management-software suite (Q3 FY26 call, 2026-05-05).

What the business actually is today: an NVIDIA-rack integrator. Management states AI GPU-related platforms were over 80% of revenue in Q3 FY26 (Q3 FY26 call, 2026-05-05). The product line is overwhelmingly built around NVIDIA’s rack-scale systems — Charles Liang lists “GB300 NVL72, MNB-300 HGX, B200 NVL4 and inferencing application optimized RTX product lines,” with Vera Rubin / NVL72 SuperCluster next (Q3 FY26 call). There are secondary platforms with AMD (MI350, future Helios/MI400), Intel (Xeon 6+), and Arm, but the economic center of gravity is air- and liquid-cooled NVIDIA HGX/Blackwell racks. (Fact.) Interpretation: despite the “Total IT Solutions” branding, Supermicro is, in substance, a high-velocity assembler/integrator of someone else’s accelerators — a point the financials make unambiguous.

End markets and go-to-market. The company sells into enterprise data centers, cloud/hyperscale (and the fast-growing “NeoCloud” GPU-rental operators), AI model builders, sovereign-AI projects, and 5G/edge (FY2025 10-K). It reaches them both directly and through distributors, value-added resellers, system integrators, and OEMs (10-K, “Sales and Marketing”). Manufacturing is concentrated in San Jose (nearly 4 million sq ft of Bay Area footprint after the newly announced DCBBS campus) with capacity in Taiwan, Malaysia, and the Netherlands (Q3 FY26 call). As of June 30, 2025 the company had 6,238 employees, of whom over 3,200 are in R&D/engineering and 1,733 in manufacturing (FY2025 10-K, “Human Capital”).

Revenue reporting and disclosure quality. Supermicro reports one operating segment. It discloses revenue by channel — “Enterprise/Channel” vs. “OEM Appliance and Large Data Center” — and by geography. In Q3 FY26: Enterprise/Channel was $2.8B (28% of revenue, up from 15% the prior quarter); OEM Appliance/Large Data Center was $7.4B (72%). Geographically the U.S. was 69%, Asia 13%, Europe 7%, Rest of World 11% (Q3 FY26 call). Open Question: the company does not disclose DCBBS, software, or service as a reported revenue line, so investors cannot independently verify the margin-accretion claims that anchor the bull case; software revenue is disclosed only anecdotally ($34M in Q2 FY26 rising to “$46 million booked” in Q3 — i.e., <0.5% of revenue) (Q3 FY26 call).

Revenue is lumpy, project-based, and credit-/working-capital-intensive. This is not a recurring-revenue model. Revenue recognition hinges on customer site readiness — power and networking being installed at the customer’s data center — which management blamed for the sequential Q3 FY26 revenue drop to $10.2B from $12.7B (Q3 FY26 call). The cash-conversion cycle swung to 106 days in Q3 FY26 (from 54), inventory sat at $11.1B, and the company drew on U.S., AR-sale, and a new $1.8B Taiwan revolving facility, ending the quarter in a $7.5B net-debt position (Q3 FY26 call). Interpretation: the model is closer to a financed contract-assembly business than a software-or-IP franchise — a critical fact for every section that follows.

Control and key people. Founder Charles Liang is Chairman, President & CEO; co-founder Sara Liu (his spouse) is an officer; David Weigand is CFO. The 10-K explicitly flags that “concentration of our capital stock ownership with insiders likely limits your ability to influence corporate matters” (FY2025 10-K). Fact. The company is also operating under the shadow of a recent SDNY indictment of former-associated individuals over alleged product diversion to China, an ongoing independent board investigation (Munger, Tolles & Olson; AlixPartners), and the 2024–25 delayed-filing/auditor-resignation episode — context developed in an earlier section.

3. Industry Dynamics

The industry: AI/data-center server systems. Supermicro sits in the system-integration / OEM tier of the AI infrastructure value chain. The chain runs: chip designers (NVIDIA, AMD) → contract fabs (TSMC) and component makers (memory, SSD, networking) → server/rack integrators (Supermicro, Dell, HPE, Lenovo, and the ODMs Foxconn, Quanta, Wiwynn/Wistron) → data-center operators (hyperscalers, NeoClouds, sovereigns, enterprises). The integrator tier’s job is to assemble scarce, expensive accelerators into working, cooled, networked, serviceable racks. (FY2025 10-K, “Competition.”)

The demand backdrop is genuinely large and growing. This is the strongest argument for the industry. Management cites partner framing of a $2–4 trillion AI data-center build-out over 2–3 years and a roughly $200B+ addressable opportunity at ~10% share (BofA conf, 2026-06-02). Even discounting management’s numbers heavily, the secular tailwind is real: Supermicro’s own revenue went from $7.1B (FY23) to a ~$40B run-rate (FY26 guide), and U.S. revenue alone grew 154% YoY in Q3 FY26 (Q3 FY26 call). Fact / Interpretation: the demand pool is vast and expanding — but a large, fast-growing market is precisely the condition that, in Marathon “Capital Returns” terms, attracts a flood of supply.

Industry structure is the problem — this is a brutal place to capture profit. Apply the Greenwald “Competition Demystified” structural lens:

• Barriers to entry are low at the integration layer. As the BofA analyst put it to management directly: “Everybody seems to be doing a data center rack for NVIDIA. You’ve got ODMs, you’ve got EMS companies” (BofA conf, 2026-06-02). Supermicro’s own 10-K concedes rising competition “from original design manufacturers … that benefit from their scale and very low-cost manufacturing and are increasingly offering their own branded products” (FY2025 10-K). When the analyst noted that “anyone can potentially assemble something,” management’s rebuttal was integration, software, and service — a tacit admission that the assembly step itself is not defensible.
• The scarce, value-bearing input is owned by suppliers, not integrators. NVIDIA controls GPU allocation and effectively gatekeeps who gets to build the most-wanted racks; memory/SSD/CPU shortages in the last six months (“double, triple, more than triple”) squeezed every integrator simultaneously (Q3 FY26 call). The profit pool in this chain accrues to NVIDIA (the chip near-monopoly) and to the cloud operators selling tokens — not to the interchangeable box-builders in the middle.
• Buyer power is high and rising. The largest buyers — hyperscalers — increasingly design their own ODM racks, disintermediating the OEM tier entirely. Even NeoClouds and sovereigns are price- and TCO-sensitive on a fungible, NVIDIA-defined platform.
• The product is largely commoditized. Every integrator builds around the same NVIDIA reference architecture (NVL72, HGX). Differentiation collapses to speed, cooling engineering, service, and price — and price competition is the visible result.

The capital-cycle read (Marathon) is unambiguously cautionary. High advertised returns in AI servers during 2023–24 drew capital and capacity from OEMs (Dell, HPE), ODMs (Foxconn, Quanta, Wiwynn), and EMS firms alike. Supermicro itself is pouring capital into capacity — “6,000 racks per month,” eight new Bay Area buildings, ramps in Taiwan/Malaysia/Netherlands (Q3 FY26 call). When many suppliers simultaneously expand into a commoditizing product, the textbook outcome is margin mean-reversion — and the industry’s already-thin integrator margins suggest the reversion is happening in real time, not at some future date. Interpretation, high confidence.

Regulation and cyclicality. Two industry-level overhangs: (1) export controls on advanced compute to China — central to the DOJ indictment and a structural constraint on a global integrator (Q3 FY26 call); and (2) demand cyclicality / concentration — the build-out is funded by a handful of hyperscaler/NeoCloud capex budgets, any wobble in which hits the whole integrator tier at once. The business is also tariff-exposed (management cited tariff cost pressure on Q3 margins).

Verdict: structurally a bad industry at the layer Supermicro occupies. The end-market is enormous and growing, which is why the equity is exciting — but the integration tier itself is structurally unattractive: low entry barriers, a supplier (NVIDIA) that owns the scarce input and the profit pool, powerful buyers who can vertically integrate, a commoditized reference-architecture product, and a textbook capital-cycle flood of competing capacity into the same NVIDIA-rack assembly task. A great market does not make a great industry for the box-builders in the middle. The 8% gross margin is the industry’s verdict, not Supermicro’s alone.

4. Competitive Position

The central question is whether Supermicro has a durable competitive advantage or is a fast, well-run but ultimately price-taking assembler in a commoditizing market. The evidence points firmly to the latter. Working through the Greenwald advantage taxonomy:

Moat candidate 1 — Speed / “Time-to-Online” (TTO) and first-to-market. Management’s headline claim is “the fastest Time-to-Online in the industry” and being “among the first to market” on each new NVIDIA platform (Q3 FY26 call). This is real and is Supermicro’s genuine operational edge: modular building blocks let it field new-GPU systems quickly. But speed is not a structural moat — it is an execution advantage that competitors replicate each cycle, and it does not confer pricing power. The tell: if first-to-market produced durable advantage, it would show up as fat early-cycle margins. Instead, FY26 9-month GAAP gross margin was 8.2% (corroborated by the 8.2–8.4% Q4 guide). A first-mover that earns 8% gross margins is not extracting rent from being first — it is racing to ship at thin spreads before the field arrives.

Moat candidate 2 — Direct-liquid-cooling (DLC) engineering lead. DLC is the most plausible source of real differentiation: thermal/power engineering is non-trivial, and Supermicro is investing in a dedicated DLC-2 clean room (Q3 FY26 call). Interpretation: this is a temporary technology lead, not a structural barrier. Liquid cooling is an industry-wide priority; Dell, HPE, and the ODMs are all building DLC racks for the same NVIDIA platforms. A lead measured in quarters, on a fast-moving public reference architecture, is not the kind of durable intangible (patent thicket, proprietary standard) that sustains excess returns. If DLC were a true moat, Supermicro’s liquid-cooled mix (3,000 of 6,000 racks/month — BofA conf, 2026-06-02) would command premium margins; it does not.

Moat candidate 3 — Scale and breadth of the building-block portfolio. Supermicro’s “broad product portfolio” and “sufficient manufacturing capacity” are listed among its own competitive factors (FY2025 10-K). But this is sub-scale relative to its competitors, which is the opposite of an economies-of-scale moat. The 10-K itself concedes: “most of our competitors have longer operating histories, significantly greater resources, greater name recognition, or deeper market penetration” (FY2025 10-K). Greenwald’s economies-of-scale advantage requires being the largest player in a market with customer captivity; Supermicro is the smaller challenger (management reflexively calls itself “number two,” BofA conf) against Dell, HPE, Lenovo and the giant ODMs (Foxconn, Quanta). Scale here favors the incumbents and the ODMs, not Supermicro.

Moat candidate 4 — DCBBS, software, and service (“captivity”/switching costs). This is the bull’s best hope: bundle SuperCloud Composer software and on-site service to raise switching costs and lift margin (management claims DCBBS gross margin >20% and that DCBBS will be >25% of profit “in the coming few years,” Q3 FY26 call). Skeptical read: the evidence is thin and self-reported. Software revenue is ~$46M/quarter — under 0.5% of revenue (Q3 FY26 call). DCBBS revenue/margin is not separately disclosed, so the >20% claim is unverifiable. SuperCloud Composer competes against incumbents’ management stacks and customers’ own tooling. Until switching costs show up as rising blended margins and stickier, less-lumpy revenue, “captivity” remains an assertion, not a moat. (Open Question.)

Customer concentration: the anti-moat. A real moat produces a diversified, captive customer base. Supermicro’s is the reverse: one 10%+ customer was 63% of Q2 FY26 revenue and 27% of Q3 FY26 (Q3 FY26 call). The JPMorgan analyst put the implication bluntly: “having a concentrated customer with 63% of your revenue is probably … you’re trading off … market share instead of taking … better margin” (JPM conf, 2026-05-18). A vendor that takes 63%-of-revenue mega-orders at low margin to win volume is demonstrating the absence of pricing power, not its presence. Captive customers don’t extract that.

The share-stability test fails. Greenwald’s most reliable moat diagnostic is market-share stability — moats produce steady share; their absence produces volatile share. Supermicro’s AI-server share gained rapidly in 2023–24, then was actively ceded during the 2024–25 delayed-filing/auditor-resignation episode, when it could not file timely and Dell and others took share (consistent with the BofA “number two” framing). Share that swings with an accounting crisis is not protected by any structural barrier — customers switched away and (per management) are switching back. Volatile share = no moat. High-confidence interpretation.

The ROIC / margin test fails. A moat must surface in financial outcomes. The decisive datum: revenue roughly doubled from FY24 ($15.0B) to the FY26 ~$40B run-rate while gross profit barely grew (FY24 gross profit $2.06B; FY25 $2.43B; FY26 9-mo $2.28B). Gross margin fell from 15.4% (FY22) to 8.2% (FY26 9-mo). The historical “traditional compute” margin of 14–17% that management itself referenced (JPM conf, 2026-05-18) has been more than halved by the shift into NVIDIA AI racks. A business adding $25B of revenue with no incremental gross profit has, by definition, no pricing power and no moat — it is converting demand into low-margin throughput. The Q3 FY26 recovery to 10.1% non-GAAP was explicitly driven by transient factors (lower tariffs, non-recurring expedite fees, lower inventory reserve charges) plus mix, not by structural pricing power (Q3 FY26 call, Weigand).

Verdict: no durable competitive advantage — a commoditized, price-taking integrator with a genuine but non-defensible speed/execution edge. Supermicro is fast, technically competent at DLC, and well-positioned for volume — but speed and cooling are replicable execution advantages, not barriers to entry; it is sub-scale versus incumbents and ODMs; its customer base is dangerously concentrated; its share has proven unstable; and the 8% gross margin while revenue doubled is conclusive evidence that it cannot price above cost-plus. DCBBS/software could build a switching-cost moat, but on current disclosure it is a hypothesis, not a fact. In Greenwald terms: no supply/cost advantage (it is sub-scale), no demand captivity (63%-customer churn-prone, volatile share), and therefore no economies-of-scale-plus-captivity flywheel. The market is commoditized and Supermicro is a price-taker within it.

5. Growth History and Forward Opportunities

Growth history — explosive, but increasingly low-quality. Revenue compounded from $3.56B (FY21) to $5.20B (FY22), $7.12B (FY23), $14.99B (FY24, +110%), $21.97B (FY25, +47%), and a $27.94B 9-month FY26 print toward a ~$40B FY26 guide (Q4 target $12B; full-year range $38.9–40.4B) (Q3 FY26 call). On the top line this is one of the great revenue ramps in hardware. (Fact.)

But growth quality has deteriorated sharply, and this is the crux. The Greenwald/Marathon lens insists that growth only creates value when it earns returns above the cost of capital; growth on a commoditized product at sub-cost-of-capital spreads destroys value even as it dazzles. Three facts establish the low quality:

1. Gross profit did not follow revenue. Revenue ~doubled FY24→FY26 (run-rate) while gross profit went $2.06B → $2.43B → $2.28B (9-mo) — essentially flat. Each incremental dollar of AI-rack revenue is arriving at near-zero incremental gross margin. This is the signature of commodity-volume growth, not franchise growth.
2. Growth is funded by the balance sheet, not generated by it. The Q3 FY26 quarter consumed $6.6B of operating cash, pushed inventory to $11.1B, and swung the company into $7.5B net debt, requiring a new $1.8B Taiwan credit line (Q3 FY26 call). Management openly contemplates returning to capital markets if growth re-accelerates: “If we try to double again revenue, then we may need some more help in terms of capital” (Charles Liang, Q3 FY26 call). Growth that requires perpetual external financing at thin margins is fragile.
3. Growth is lumpy and concentration-driven. The sequential Q3 drop (revenue −19% q/q on “site-readiness delay”) and the 63%→27% single-customer swing show revenue lurching with individual mega-deals (Q3 FY26 call). High-quality growth is broad-based and predictable; this is neither.

Forward opportunities — large in size, unproven in economics. Management’s growth narrative rests on:

• Continued AI/NVIDIA platform cycle (Vera Rubin/NVL72, GB300) and AMD MI400/Helios — real volume, but the same low-margin commodity dynamic, with NVIDIA controlling allocation.
• DLC adoption as racks get hotter — plays to Supermicro’s strength, but is industry-wide and not margin-protected.
• DCBBS + software + service — the only forward driver that could improve economics (management targets >20% DCBBS gross margin, >25% of profit). This is the lever that matters most for growth quality, but it is small (<0.5% of revenue in software) and undisclosed at the segment level (Q3 FY26 call). Open Question.
• Enterprise diversification — enterprise/channel jumped to 28% of Q3 revenue from 15% (+45% q/q), which management frames as higher-margin and stickier (Q3 FY26 call). If durable, this is the single most encouraging quality signal in the data; one quarter is not a trend.
• Sovereign AI — Spain/Catalonia, Korea, the DataVolt deal (BofA conf, 2026-06-02). Large potential, slow-moving, lumpy, and politically/export-control exposed.
• TAM — a ~$200B opportunity at 10% of a $2–4T build-out vs. a ~$40B run-rate (BofA conf). Ample headroom if economics can be made to work.

The variant question is not “can it grow?” but “can it grow profitably?” Supermicro has proven it can capture AI-server volume; it has not yet proven it can capture profit. Management’s entire forward case is a margin-mix story — DCBBS, software, service, enterprise — layered on top of a low-margin NVIDIA-rack base. Each is plausible; none is yet evidenced in the consolidated numbers, where gross margin is still ~8%.

Verdict: low-quality growth — spectacular top-line, value-questionable economics. The revenue ramp is genuine and the market is huge, but the growth to date is commodity-volume growth: flat gross profit on doubled revenue, financed by debt and inventory, lumpy and customer-concentrated. It will create shareholder value only if the unproven margin-mix shift (DCBBS/software/service/enterprise) actually lifts and sustains blended margins toward double digits — a transition management asserts but has not yet demonstrated at scale or disclosed transparently. Until the consolidated gross-profit line inflects upward with revenue, this is growth that is impressive to watch and dangerous to underwrite.

6. Financial Quality

Super Micro’s income statement tells a story that, read top-line only, looks like one of the great growth stories of the AI cycle — and, read line-by-line, reveals a low-margin, capital-devouring assembler whose economics are deteriorating as it scales. Revenue compounded from $3.56B in FY21 to $14.99B in FY24 to $21.97B in FY25, and the first nine months of FY26 already total $27.94B, against full-year guidance of roughly $40B (company filings; FY26 guide per Q3 FY26 call, 2026-05-05). A ~6x revenue increase in four years is real and not in dispute. The problem is that essentially none of the qualities investors associate with a durable franchise — margin expansion, operating leverage, cash generation, returns on capital, share-count discipline — travel alongside that growth. In several cases they move sharply the wrong way.

Gross margin is in structural collapse, and gross-profit dollars have stopped growing. GAAP gross margin fell from 18.0% in FY23 to 13.8% in FY24, 11.1% in FY25, and 8.2% in the first nine months of FY26, with management guiding Q4 to 8.2–8.4% (filings; Q3 FY26 call). This is the single most important number in the analysis. The most telling cut is gross profit in dollars: $2.06B (FY24) → $2.43B (FY25) → $2.28B (FY26 9-mo) (filings). In other words, the company nearly doubled revenue from FY24 toward a ~$40B run-rate while generating roughly the same absolute gross profit. [Interpretation] Every incremental dollar of AI-server revenue is arriving at near-zero incremental gross margin — the textbook signature of a price-taking integrator passing through expensive third-party silicon (predominantly NVIDIA GPUs) with a thin assembly markup, not a business with pricing power. Management attributes the compression to customer mix (a few hyperscale-scale accounts at concession pricing), new-platform ramp costs (Blackwell/liquid-cooling), and inventory reserves, but the trend long predates any single ramp and points to the underlying economics rather than a transient.

Operating leverage is negative. Operating margin fell from 10.7% (FY23) to 8.1% (FY24) to 5.7% (FY25) to 4.6% (FY26 9-mo) (filings). Scale is supposed to spread fixed opex over a larger base; here, collapsing gross margin has overwhelmed any opex efficiency, and opex itself carries meaningful stock-based compensation (~$114M/quarter guided for Q4) (Q3 FY26 call) that flatters GAAP-to-non-GAAP bridges investors lean on. The clearest evidence that this is not a compounder: net income actually fell year-over-year in FY25 — $1,153M (FY24) to $1,049M (FY25) — despite revenue rising +47% (filings). A business whose absolute profit shrinks while revenue grows by half is, by definition, not scaling profitably. FY26 9-month net income of $1,052M annualizes only modestly above the FY25 level on revenue that will be up another ~80%. [Interpretation] The market is being asked to pay for revenue scale that is not converting into earnings scale.

Cash flow is the most damning chapter. Because Super Micro buys GPUs, memory and components, builds them into systems, and carries the receivables until customers pay, growth consumes cash rather than producing it. Cash from operations was −$2,486M in FY24, +$1,660M in FY25, and −$7,557M in the first nine months of FY26 (filings). The FY25 positive figure is the exception that proves the rule — it coincided with a period of slower sequential growth and inventory drawdown, not with durable cash conversion. The FY26 figure is staggering: the company burned over $7.5B of operating cash in nine months. CapEx is modest (PP&E was only ~$608M at March 2026 against $23.5B of total assets — genuinely asset-light) (Q3 FY26 10-Q), so this is not a factory-building story; the cash is being absorbed almost entirely by working capital. [Interpretation] Free cash flow is deeply negative in every genuine growth year, and the faster the company grows, the more cash it incinerates.

The balance sheet shows exactly where the cash went — and how fragile the structure became. Between June 30, 2025 and March 31, 2026:

• Inventory rose from $4.68B to $11.10B (+137%) (10-Q).
• Accounts receivable rose from $2.20B to $8.41B (+281%) (10-Q).
• Cash fell from $5.17B to $1.29B (10-Q).
• Total assets grew $14.02B → $23.45B; stockholders’ equity rose only $6.30B → $7.58B (10-Q).

Inventory and receivables together swelled by roughly $12.6B in nine months, funded by draining cash and adding debt. [Interpretation] An $11.1B inventory balance in a market defined by rapid GPU generational transitions (Hopper→Blackwell→Rubin) is itself a risk: a single platform shift or demand air-pocket can force write-downs on stranded components — precisely the dynamic management already cites as a margin drag. The $8.4B receivables balance, against thin margins, concentrates credit risk in a small number of large AI buyers.

Returns on capital are mediocre and falling. Third-party aggregators report TTM ROE ~17.9% and ROA ~5.5% (third-party aggregate; not reconciled to filings — Assumption). Even taking the ROE at face value, it is unremarkable for a business sold to investors as an AI thoroughbred, it is declining as the equity base is diluted, and the 5.5% ROA is the more honest figure for a working-capital-heavy assembler — it says the enormous asset base ($23.5B) throws off very little profit per dollar deployed. [Interpretation] Under the Greenwald lens, returns this thin on this much capital, with no pricing power and a commodity input pass-through, are the financial fingerprint of an absence of competitive advantage, not its presence.

Quality-of-earnings caveat (material). Three factors warrant conservative treatment of every reported number. First, accounting history: the auditor Ernst & Young resigned in October 2024, citing unwillingness to rely on management/board representations; the FY24 10-K was delayed; a special committee reviewed and reported no evidence of fraud; BDO was appointed and delinquent filings were brought current in February 2025, narrowly avoiding Nasdaq delisting (public filings/8-Ks, 2024–2025). This followed a prior SEC revenue-recognition matter settled in 2020 (~$17.5M, with a CEO clawback) (SEC, 2020). Second, related-party manufacturing: Ablecom and Compuware — Taiwan companies affiliated with the Liang family — supply a large share of chassis and subassemblies; Ablecom manufactured ~95.5% of SMCI’s chassis in the nine months ended Mar-31-2026, and combined related-party purchases were 2.1% of cost of sales (9M FY26) vs. 3.4% (9M FY25) (10-Q, Note 10). The CEO also carries an unsecured personal loan (originated ~$12.9M, Oct-2018) from the spouse of a related-party principal (10-Q). Related-party manufacturing embedded in COGS makes the already-thin gross margin harder to independently verify. Third, mechanics: the combination of razor-thin margins, multi-billion-dollar working-capital swings, large inventory reserves, and SBC-heavy opex means small estimate changes (reserve timing, cost allocation) can move the modest net income line materially. [Open Question] How much of reported gross margin and inventory carrying value would survive a fully independent, conservative restatement remains an open question that no outside analyst can fully close.

Verdict — do economics improve with scale? No; they deteriorate. This is the rare case where the data answers the framework question unambiguously and negatively. Across the largest revenue ramp in the company’s history, gross margin halved, gross-profit dollars went flat, operating margin more than halved, net income fell in FY25, operating cash flow swung to a ~$7.6B nine-month burn, and the balance sheet’s cash cushion was nearly exhausted to fund inventory and receivables. The business gets less profitable and more capital-hungry as it grows — the opposite of a scale economy. Super Micro is a low-margin, working-capital-intensive assembler riding a demand wave, not a high-return compounder, and its financial quality should be assessed accordingly.

7. Capital Allocation

Capital allocation is the bridge between business value and shareholder value, and Super Micro’s record is that of a serial consumer of external capital, not a returner of it. The company has never paid a dividend and has never repurchased stock; it is structurally a net issuer (filings). That is not inherently wrong for a fast-growing business — but it raises the bar: capital raised must earn a return above its cost, and here the underlying economics (an earlier section — thin, falling margins; mediocre ROIC; negative growth-year FCF) make that case difficult to sustain.

The financing model: borrow and dilute to fund working capital. Because growth burns operating cash, Super Micro funds itself through a stack of converts, bank lines, and equity. As of March 31, 2026 the debt structure comprised (Q3 FY26 10-Q):

• Convertible notes, $4.725B face: 2028 ($700M, 2.25%, conversion $61.06); 2029 ($1,725M, 3.50%, conversion $83.44); 2030 ($2,300M, 0.00%, conversion $55.20). All are out-of-the-money at the $40.64 reference price.
• Bank revolvers/lines, ~$4.069B: JPMorgan revolver $2.0B, CTBC revolver $1.763B, Taiwan bank lines $0.306B; plus ~$44.5M term loans.
• Total debt ~$8.8B; net debt ~$7.5B against just $1.29B cash.

The 2029 convert is itself a cautionary tale of weak structuring: originally issued at a 0.00% coupon in February 2024, it was amended in February 2025 to add a 3.50% cash coupon, recognizing a $30.3M extinguishment loss (filings). [Interpretation] A company forced to renegotiate a zero-coupon convert into a cash-pay instrument within a year — during the audit/delisting episode — was paying up for capital from a position of weakness, not strength.

The June 2026 recapitalization. On June 9–10, 2026 the company announced a roughly $7B capital raise: a $1.25B common-stock offering; 75M depositary shares representing a Series A Mandatory Convertible Preferred ($50 liquidation preference each = $3.75B, plus ~$0.56B over-allotment of 11.25M shares), auto-converting around June 2029 and carrying a cumulative dividend payable in cash or stock; and a $2.0B at-the-market (ATM) equity program (424B5 ×2, S-3ASR, 8-K, 2026-06-09/10; underwriters JPMorgan/Goldman/Citi; reference price $40.64). Use of proceeds: working capital, component purchases for AI orders, and general corporate purposes. [Interpretation] This is, in plain terms, an emergency recapitalization: the balance sheet had ~$7.5B net debt and only $1.29B cash, and the operating engine had just consumed $7.6B of cash in nine months. The mandatory convertible — which will become equity in 2029 regardless of price, and whose dividend can be paid in still more stock — is a particularly dilutive instrument, chosen because straight equity at $40.64 (a fraction of the post-split highs) is expensive and additional senior debt would strain an already-levered, thin-margin balance sheet. The raise validates the financial-quality thesis: the company cannot self-fund its own growth.

Dilution is chronic and about to accelerate. On a split-adjusted basis (10-for-1 split, October 2024), shares outstanding went from ~565M (Jun-23) to 642M (Jun-24) to 624.7M (Jun-25) to 673.6M (Mar-26) — roughly +19% over three years, before the June 2026 raise (filings). The new common offering, the $2.0B ATM, the mandatory convertible (which converts to ~$4.3B of equity), the stock-or-cash preferred dividend, and the existing $4.725B of converts (in-the-money above ~$55–83) layer on materially more potential dilution. [Interpretation] Per-share value is being continuously diluted to fund inventory and receivables that earn a thin, declining margin — the worst combination for compounding.

M&A and R&D — the one relatively clean part of the record. Super Micro is an organic assembler; M&A has been minimal, so there is no record of value-destructive empire-building, and R&D is run lean as a percentage of sales (filings). [Interpretation] This avoids one common capital-allocation sin, but it is faint praise: the absence of bad acquisitions is not evidence of good allocation when the core activity — funding negative-FCF growth with serial dilution — is itself questionable.

Insider behavior signals no conviction. The Form 4 record since 2025 shows no open-market purchases by insiders. Activity consists of gifts (code G; Liang/Liu), option exercises with tax withholding, and RSU grants (M/F/A; Weigand, Clegg, Xiao, Cheung), alongside heavy Form 144 planned-sale volume (SEC Form 4/144 corpus). [Interpretation] Notably, no insider bought stock on the open market into a $40.64 reference price during a ~$7B dilutive raise — the cheapest, most direct signal of management confidence is absent. Founder-CEO Charles Liang’s ~13.8% economic interest (including related holdings) does align him with the equity, but his alignment coexists with the related-party manufacturing arrangements (Ablecom/Compuware) and the personal loan from a related-party principal’s spouse, which complicate the incentive picture rather than clarify it (10-Q, Note 10; DEF 14A).

Governance overhang. The 2024 auditor resignation and delinquent-filing episode, layered on the 2018–2020 SEC revenue-recognition matter, mean capital allocation is being conducted by a management/board that has twice strained the credibility of its own financial reporting (public filings; SEC 2020). [Open Question] Whether the post-BDO control environment is now robust enough to govern an $8.8B debt stack, $11.1B of inventory, and a serial equity-issuance machine is unproven.

Verdict — has management allocated capital intelligently? No, on the evidence available. Management has built a financing model that requires perpetual external capital — converts repeatedly restruck on unfavorable terms, ~$4B of bank lines drawn, and a ~$7B emergency equity-and-mandatory-convert raise — to fund working capital that earns a thin and falling return. There are no buybacks or dividends because there is no surplus cash to return; growth is financed by chronic, accelerating dilution; insiders are not buying; and the whole apparatus operates under an unresolved governance and related-party overhang. The cleanest favorable point — disciplined avoidance of large M&A — does not offset a model that systematically converts shareholder capital into low-return inventory and receivables. This is capital consumption in service of revenue scale, not intelligent capital allocation in service of per-share value.

8. Changes and Headwinds — Last Two Years

The two years to June 2026 reshaped Super Micro twice over: a revenue explosion that turned a mid-cap server vendor into a ~$40B AI-infrastructure assembler, and a governance crisis severe enough to threaten the company’s continued listing. Both must be weighed together, because the same demand wave that drove the top line also drove the financial and reputational fragility now embedded in the equity.

The accounting and governance saga (Fact). On August 27, 2024, Hindenburg Research published a short report alleging accounting manipulation, undisclosed related-party dealings through Liang-family-affiliated entities (Ablecom, Compuware), and sanctions/export-control circumvention (Hindenburg Research, “Super Micro,” Aug-27-2024). Within days SMCI delayed its FY2024 10-K, filing an NT 10-K. The escalation that followed is what distinguishes this episode from an ordinary short attack: in October 2024 auditor Ernst & Young resigned, stating it was “unwilling to be associated with the financial statements prepared by management” and citing concerns over management’s and the Audit Committee’s representations (SMCI 8-K, Oct-30-2024). (Interpretation: an incumbent Big Four auditor resigning mid-engagement — rather than merely qualifying an opinion — is among the most severe negative governance signals a public company can emit, and is materially worse than the underlying Hindenburg allegations themselves.) A Special Committee, advised by independent counsel, subsequently reported it had found no evidence of fraud or misconduct by management or the board, while recommending some remediation (SMCI press release, Dec-2-2024). BDO USA was appointed as the new auditor in November 2024. Facing a Nasdaq delisting deadline for delinquent filings, SMCI submitted a compliance plan, received an extension, and on February 25, 2025 filed the delinquent FY2024 10-K plus the FY2025 Q1 and Q2 10-Qs, regaining compliance and avoiding delisting (SMCI 8-K, Feb-25-2025).

This was not a first offense (Fact). SMCI was delisted by Nasdaq in 2018 for late filings (relisted 2020), and in August 2020 settled SEC charges (~$17.5M) over premature revenue recognition; CEO Charles Liang reimbursed the company ~$2.1M under the the Sarbanes-Oxley Section 304 clawback provision (SEC AAER, Aug-2020). (Interpretation: the 2024–25 crisis is best read not as a one-time aberration but as a recurrence of a long-standing pattern of weak financial controls and aggressive practices at the same company under the same CEO — a base-rate problem, not a one-off.)

The export-control overhang is fresh and unresolved (Fact). On March 19, 2026, the U.S. Attorney’s Office for the SDNY unsealed an indictment of three individuals employed by or associated with the company in connection with an alleged conspiracy to commit export-control violations — alleged diversion of products to China (SMCI 8-K, Mar-20-2026; Q3 FY26 10-Q, filed May-11-2026). The company states it is not a defendant or a target and is cooperating; independent directors retained Munger, Tolles & Olson and AlixPartners for an independent investigation, ongoing as of the Q3 FY26 call (Q3 FY26 earnings call, May-5-2026). Management stated it does not currently believe a restatement is required and that it understands there has been no change in NVIDIA allocation (ibid.). On June 9, 2026, SMCI filed an 8-K supplementing its risk factors specifically to reflect this matter (SMCI 8-K, Jun-9-2026). (Open Question: the investigation is unfinished, so the “no restatement / no company involvement beyond the three indicted individuals” position is a management hypothesis subject to BDO review, not an audited conclusion.)

The related-party structure persists through the crisis (Fact). Ablecom — owned by the CEO’s brother Steve Liang and family, with Charles Liang and his wife also holding interests — manufactured ~95.5% of the chassis SMCI purchased in the nine months ended Mar-31-2026 (10-Q, May-11-2026). Combined Ablecom/Compuware purchases were 2.1% of cost of sales (9M FY26) vs. 3.4% (9M FY25) (ibid.). Separately, in October 2018 the CEO personally borrowed ~$12.9M from Chien-Tsun Chang, spouse of Steve Liang, an unsecured loan with no maturity date (ibid.). (Interpretation: the dollar magnitude of related-party purchases is modest relative to COGS, but the ~95% chassis-sourcing concentration in a family-affiliated supplier — exactly the structure Hindenburg flagged — remains intact, leaving the central governance criticism unaddressed in substance.)

The business changes (Fact). Revenue rose from $14.99B (FY24) to $21.97B (FY25) to a ~$40B FY26 guide ($38.9B–$40.4B) (10-K FY25; Q3 FY26 call), with AI GPU platforms >80% of revenue. Strategically the company is pivoting from box-builder toward rack-scale “Data Center Building Block Solutions” (DCBBS) and direct-liquid-cooling (DLC), with capacity expansion across San Jose (~8 buildings, ~4M sq ft), Taiwan, Malaysia and the Netherlands, targeting >6,000 racks/month (Q3 FY26 call). Enterprise mix jumped from 15% to 28% of revenue (Q2→Q3 FY26), and the company is pursuing sovereign-AI deals (DataVolt, Spain/Catalonia, Korea). To fund this, the balance sheet was rebuilt with serial dilution and leverage: a $1.725B 0% 2029 convert (Feb-2024) repriced to 3.50% in Feb-2025 at a $30.3M loss, a $700M 2.25% 2028 convert, a $2.3B 0% 2030 convert (Jun-2025), a JPM $2.0B revolver and CTBC $1.763B facility, and in June 2026 a ~$7B raise ($1.25B common + $3.75B mandatory convertible preferred via 75M depositary shares + a $2.0B ATM).

Verdict — net weaken the thesis. The demand-side change is genuinely positive: SMCI capitalized on the AI build-out faster than any peer, and the DCBBS/enterprise pivot is a credible attempt to escape pure box-building. But it is overwhelmed by the governance ledger. An auditor resignation, a near-delisting, a recurrence of the 2018–2020 controls failures, an unresolved DOJ export-control matter, an intact ~95% related-party manufacturing dependency, and a balance sheet leveraged and diluted to fund hyper-growth at 8% gross margins together raise the company’s risk premium structurally. The last two years made SMCI bigger and more strategically ambitious — and simultaneously less trustworthy and more financially fragile. On balance, the changes weaken the investment case.

9. Risk Analysis

SMCI carries an unusually dense and correlated risk stack: it is a low-margin, working-capital-hungry assembler, dependent on a single supplier (NVIDIA) for its key input and on a handful of customers for its revenue, operating under a live DOJ matter and a recent auditor change, and funding hyper-growth through repeated dilution and leverage. The danger is not any one item but their correlation — a demand pause, a margin disappointment, and a financing window closing could arrive together.

High-likelihood / high-impact cluster. Four risks sit in the top-right corner. (1) Customer concentration — a single customer represented 63% of Q2 FY26 revenue and the revenue shortfall in Q3 was attributed to that customer’s site-readiness delay (Fact, Q3 FY26 call). This is not diversified demand; it is a small number of large orders whose timing the company does not control. (2) Gross-margin collapse — margins have fallen by more than five points in two years to an ~8% handle (Fact, 10-Ks), the financial signature of an undifferentiated integrator with no pricing power. (3) Working capital / liquidity — the business consumes cash violently as it grows (9M FY26 operating cash flow of −$7.56B, inventory $11.1B against just $1.29B cash; Fact, 10-Q). The cash-conversion cycle blew out from 54 to 106 days in a single quarter (Q3 FY26 call). (4) Dilution — funding that cash consumption requires near-continuous capital-market access, evidenced by the June-2026 ~$7B raise on top of three outstanding convertible tranches (Fact).

(Interpretation: the most dangerous feature is the correlation. The same large customers that create concentration risk also drive the lumpy revenue that strains working capital; the working-capital strain forces dilutive financing; and the financing is only available while AI sentiment and the equity price hold up. A demand air-pocket would simultaneously impair revenue, trap inventory, and close the financing window — the textbook fragility of a leveraged, low-margin, hyper-growth assembler. CFO Weigand’s own comment that he “always hope[s] we need to go back to the markets for more money” because it signals growth (Q3 FY26 call) is candid confirmation that the model is structurally external-capital-dependent.)

Tail risks (lower likelihood, severe impact). A restatement (management says unlikely, but the investigation is unfinished — Open Question), a downgrade in NVIDIA allocation, or an adverse turn in the DOJ matter each individually could be thesis-ending; none is the base case on current evidence, but each is non-trivial given the company’s history. Catastrophic-loss risk is real but not central: the balance sheet is leveraged (net debt ~$7.5B) yet covenant/maturity pressure is manageable near-term, and the convertibles are long-dated (2028–2030); the June-2026 raise further reduces near-term liquidity risk. The realistic downside is severe multiple and earnings compression, not insolvency — though a closed financing window during a working-capital surge is the scenario that could turn a downturn acute.

10. Variant Perception

Consensus belief (Fact/Interpretation). The sell-side is cautious-to-mixed, not euphoric: the average price target (~$37) sits below the ~$40.64 quote, and short interest is elevated at 18.1% of float. Consensus broadly holds that SMCI is a fast-growing but structurally low-quality beneficiary of AI infrastructure spend — a real volume play with a recurring governance discount. The own-history valuation percentile composite of ~70 indicates the stock is not cheap relative to its own range, despite a ~12x forward P/E that bulls cite as inexpensive. (Interpretation: the market is not pricing SMCI as a clean compounder; it is pricing it as a cyclical, governance-flagged box assembler with optionality — which is roughly the correct frame.)

The strongest bull case. SMCI is a $40B-revenue business guiding toward $50B+ on a credible AI-infrastructure tailwind, trading at ~12x forward earnings and ~0.7–0.8x sales — multiples that embed almost no margin recovery. The bull argues the 8% gross margin is a trough, depressed by tariffs, expedite fees, and inventory reserves, with Q3 FY26 non-GAAP gross margin already recovering to 10.1% from 6.4% (Q3 FY26 call). The mix shift toward higher-margin enterprise (15%→28% in one quarter), DCBBS (management guides to >20% of net income within two years, at >20% segment margins), and recurring management software ($10M→$46M/quarter) gives a path back to a “sustainable double-digit gross margin model” (Q3 FY26 call). First-mover DLC/rack-scale positioning and the fastest “Time-to-Online” let SMCI win the earliest Vera Rubin/GB300 deployments. On this view, modest margin normalization on a much larger revenue base produces an earnings inflection the ~12x multiple does not reflect.

The strongest bear case. SMCI is a commoditized integrator with no durable moat: an 8% gross margin is the absence of pricing power, share is unstable, and DLC/rack integration is replicable execution, not a barrier to entry. The OEMs (Dell, HPE, Lenovo) and ODMs (Foxconn, Quanta, Wiwynn) compete for the same NVIDIA allocation. The business is cash-destructive (9M FY26 OCF −$7.56B), serially dilutive (~$7B raised in June 2026 alone), and reliant on a single customer (63% of Q2 FY26 revenue) and a single supplier. Net income actually fell in FY25 ($1,049M vs. $1,153M in FY24) despite +47% revenue — growth that did not translate to profit. Overlaying this is an unresolved governance/DOJ/related-party overhang and a history (E&Y resignation, 2018 delisting, 2020 SEC settlement) that justifies a permanent discount. On this view, the “cheap” forward multiple is a value trap: it capitalizes peak-cycle volumes and an assumed margin recovery that the competitive structure will not permit.

The 3–5 assumptions that matter most (Interpretation).

1. Is 8% the trough or the structure? The entire bull case rests on durable margin recovery toward double digits. The bear holds 8% is what commoditized assembly earns at scale.
2. Will DCBBS/enterprise/software meaningfully re-rate the mix? Management’s >20%-of-net-income-from-DCBBS claim, if realized, changes the business model; if it stalls at a few points of profit, it does not.
3. Does demand stay durable, or is this peak AI capex? ~80%+ of revenue depends on the AI GPU build-out continuing.
4. Does the company retain unfettered NVIDIA allocation and clean capital-market access through the DOJ/governance overhang?
5. Is there a restatement or a worse-than-expected DOJ outcome? A binary that would override the operating story entirely.

What would falsify each side.

• Falsifies the bull: two-to-three more quarters of gross margin stuck near 8% as enterprise/DCBBS mix rises (proving mix shift does not lift margin); a customer defection following the indictment; or a forced, deeply dilutive raise at a depressed price. The single most decisive disconfirming datapoint would be margins failing to hold double digits even as enterprise mix climbs — that would confirm the no-pricing-power thesis.
• Falsifies the bear: sustained double-digit gross margins across multiple quarters with a visibly diversifying customer base (no single customer >~25%), positive operating cash flow as growth moderates, and a clean close to the independent investigation with no restatement. The most decisive bullish datapoint would be operating cash flow turning durably positive while revenue still grows — proving the model can self-fund and is not a perpetual capital-markets-dependent.

(Interpretation: the variant-perception edge here is narrow. Consensus already understands both the AI tailwind and the governance discount; the genuine debate is the single question of whether ~8% gross margin is cyclical or structural. An investor with a differentiated, evidence-based view on margin durability — not on AI demand, which is well understood — is the only one with a real variant perception in this name.)

11. Valuation Discussion (Embedded Expectations)

Super Micro presents the most common trap in hardware investing: a low multiple of sales that masks a high multiple of earnings on a low-quality earnings base. At ~$40.64 (Jun-9-2026 ref) and ~673.6M shares (Mar-2026, pre-raise), equity value is ~$24–27B and enterprise value ~$30.5B against ~$7.5B net debt (total debt ~$8.8B less $1.29B cash) (brief; 10-Q 2026-03-31). The central analytical task is not to declare SMCI “cheap” or “expensive” on any single metric, but to back out what the current price requires — and to test that requirement against the trailing record.

Where SMCI trades, and what each multiple is telling you.

The valuation tension is entirely a function of which line of the income statement you anchor to [Interpretation]. On the top line, ~0.7–0.8x sales screens as deep value. But that low multiple is the market correctly pricing an 8%-gross-margin business: a dollar of SMCI revenue is worth a fraction of a dollar of Dell or HPE revenue because far less of it survives to gross profit, and almost none to free cash flow. On the bottom line, ~20x trailing / ~12x forward earnings is not a value multiple for a commoditized, working-capital-hungry, governance-scarred, diluting hardware assembler. The sub-1x sales multiple and the ~20x earnings multiple are the same fact viewed from two ends — reconciled by an ~3.7% net margin. The own-history percentiles (~70th) confirm the stock is not even cheap relative to its own past (third-party valuation data).

Peer comparison — SMCI is the lowest-quality node in the OEM set.

The comp table delivers the verdict the headline sales multiple obscures: SMCI carries the lowest gross margin of the OEM peer set (~8% vs. Dell ~21%, HPE ~33%, Lenovo ~16%) yet trades at a forward P/E in line with or above HPE and only modestly below Dell [Interpretation]. Investors are paying a near-OEM earnings multiple for a business whose cost structure resembles the Taiwanese ODMs (Quanta, Wiwynn) far more than it resembles Dell. The “cheapness” is a sales-multiple illusion; on the quality-adjusted earnings basis where it should be judged, SMCI is not discounted enough to compensate for its inferior economics, cyclicality, customer concentration, and the audit/restatement overhang.

The embedded-expectations problem: the price requires a margin re-rate that the results deny. To justify ~$24–27B of equity, the market is underwriting both (a) sustained ~$45–50B revenue (FY27E Street) and (b) a step-change in profitability — because at the current margin structure the earnings are not large enough to support the price at a defensible multiple. The trailing record is the problem: FY24 revenue $14.99B; FY25 $21.97B (+47%); FY26 9-month $27.94B toward a ~$40B guide — yet net income FY25 fell to $1,049M from $1,153M in FY24 despite revenue rising 47%, and FY26 9-month net income of $1,052M is essentially flat against the prior full year. Explosive top-line growth has produced zero operating leverage [Interpretation]. So the forward ~12x P/E on ~$3.20 FY27E EPS embeds a compound assumption: revenue ~$45–50B and net income roughly doubling from a ~$1.4B run-rate to ~$2.1–2.2B — the margin half of which the trailing two years directly contradict.

Scenario analysis (transparent math; diluted basis). Dilution must be confronted first. The ~673.6M Mar-2026 share count is already stale: the June-2026 financing adds ~31M common shares ($1.25B @ ~$40.64), a $3.75B mandatory convertible preferred (75M depositary shares, auto-converts ~2029, ~$4.3B forward equity, ~75M common-equivalent), and up to ~49M shares via the $2.0B ATM. The three existing converts ($700M @ $61.06; $1,725M @ $83.44; $2,300M @ $55.20) are out-of-the-money today but cap upside — a recovery toward those strikes pulls in ~52M+ more shares. Scenario math below uses a pro-forma fully-diluted count of ~830M (673.6M + ~31M + ~49M + ~75M) [Assumption; Open Question: exact depositary-share conversion ratio].

[Assumption — all scenario inputs are explicit assumptions; net income = Rev × net margin; EPS = NI ÷ 830M diluted.]

The decisive variable is margin, not revenue [Interpretation]. In the base case, even growing revenue to $46B — itself a heroic +15% on the $40B FY26 guide — produces only ~$1.83 diluted EPS at an 8.5% gross margin, which at 12–14x supports roughly the current equity value, no more: at maintained margins the current price already capitalizes the growth. The bull case requires gross margin to inflect to ~11% — a level SMCI has not sustained in the AI era — to justify a higher price; a pure DCBBS-mix/enterprise/sovereign re-rate bet. The bear case — margins compressing toward true ODM economics (~7.5% GM) — collapses EPS to ~$1.15 and the equity well below the current mark. In every column the ~25%+ dilution is shared across the recovery.

Cross-check and verdict on embedded expectations. EV/EBITDA of 17.3x on ~$1.58B TTM EBITDA is expensive, not cheap — a growth-software-adjacent multiple on a commodity-assembly EBITDA base that does not reliably convert to free cash flow (working capital absorbs it). The market is paying ~0.7–0.8x sales because the business earns ~8% gross / ~3.7% net — pricing the commodity reality correctly, not an inefficiency to exploit. The forward ~12x P/E only “works” if one accepts a margin inflection that has not appeared in two years of results. What must be true for the current price to be conservative is precisely what the trailing financials contradict: durable margin expansion alongside sustained ~$45–50B revenue. (No price target; no recommendation — scenarios and embedded expectations only.)

12. Fact vs. Interpretation Table

13. Open Questions

1. Is ~8% gross margin the trough or the structure? Everything turns on this. Two-to-three quarters of data on enterprise/DCBBS mix vs. blended margin will answer it.
2. What are DCBBS revenue and segment margins, actually? Management claims >20% gross margin and a path to >20% of net income, but discloses no segment line. Unverifiable today.
3. Exact terms of the June 2026 mandatory convertible — coupon rate, conversion ratio/floor/cap, and therefore the precise share count it will create in 2029. Priced after this report’s as-of date.
4. Will the largest customer(s) stay? A 63%→27% single-customer swing in one quarter; the durability and identity of the anchor accounts is unknown.
5. Does the independent investigation (Munger Tolles / AlixPartners) close cleanly with no restatement, and does the SDNY export-control matter stay confined to the three indicted individuals?
6. Does NVIDIA allocation remain unaffected by the export-control overhang? Management asserts yes; it is not independently verifiable.
7. What is normalized free-cash-flow conversion once growth moderates? FY25 (+$1.66B OCF) hints the model can generate cash when not hyper-growing, but one year is thin evidence.
8. How much of inventory ($11.1B) is at risk of obsolescence across the Blackwell→Rubin transition?

14. What Must Be True

For the bull case to win (the stock materially re-rates upward):

• Gross margin must inflect and hold in double digits (≥10–11%) for several consecutive quarters, driven by a real, disclosed mix shift to DCBBS/software/enterprise — not by transient items (tariffs, expedite, reserve releases).
• Revenue must sustain toward $45–50B+ while the customer base diversifies (no single customer >~25%).
• Operating cash flow must turn durably positive as growth normalizes, ending the serial-dilution cycle.
• The independent investigation must close with no restatement and NVIDIA allocation must remain intact.
• Falsification test: if gross margin remains stuck near 8% over the next two-to-three quarters even as enterprise mix climbs, the bull thesis is dead — it would prove the mix shift does not lift blended margin, i.e., no pricing power.

For the bear case to win (the stock de-rates / underperforms):

• Gross margin compresses back toward true ODM economics (~7–8% or below) as hyperscaler/NeoCloud buyers exert pricing power on a fungible NVIDIA-rack product.
• The working-capital/dilution treadmill continues — further raises at depressed prices — while net income fails to scale with revenue.
• A demand air-pocket in AI capex, a customer loss, an adverse DOJ/restatement outcome, or an NVIDIA-allocation change crystallizes.
• Falsification test: if operating cash flow turns durably positive while revenue still grows, and gross margin holds double digits with a diversifying customer base, the bear thesis is broken — the model would have proven it can self-fund and command a margin.

The two falsification tests converge on the same observable: the trajectory of consolidated gross margin and operating cash flow over the next three quarters. That, not AI demand (which is well understood), is where the variant perception lives.

15. Source Appendix

Primary and high-quality sources underpinning this memo (full list in the standalone source appendix):

• SEC EDGAR filings (CIK 0001375365): FY2025 Form 10-K (filed 2025-08-28); Form 10-Q for the quarter ended 2026-03-31 (filed 2026-05-11) and prior FY26 10-Qs; Form 8-Ks (2024-10 auditor change; 2025-02 delinquent-filing cure; 2026-03-20 indictment disclosure; 2026-06-09 risk-factor supplement); DEF 14A proxy; Forms 3/4/5 and 144 (insider transactions); S-3ASR and two 424B5 prospectus supplements for the June-2026 offerings (2026-06-09/10). Financial series reconciled via EDGAR XBRL (revenue tag RevenueFromContractWithCustomerExcludingAssessedTax).
• Company event transcripts (earnings calls & investor conferences): Q3 FY26 earnings call (2026-05-05); Q2 FY26 (2026-02-03); Q1 FY26 (2025-11-04); Q4 FY25 (2025-08-05); BofA Global Technology Conference (2026-06-02); J.P. Morgan TMT Conference (2026-05-18).
• Market/quantitative data: third-party market-data provider (2026-06-09/10); public market data (yfinance) (price, EV, peer cross-check). Third-party aggregated data treated as non-primary and reconciled to filings where material.
• Third-party research referenced as context (not evidence): Hindenburg Research, “Super Micro” (2024-08-27); SEC AAER re 2020 settlement.

Facts cited to filings/transcripts are primary. Interpretations and assumptions are labeled inline. Management commentary is treated as hypothesis, validated against filings and external data per the firm’s research standard.

APPENDIX A — Standard Diligence Questionnaire

Super Micro Computer, Inc. (NASDAQ: SMCI) — answers as of 2026-06-10. Fiscal year ends June. All figures grounded in verified company filings, third-party market-data providers (flagged where unreconciled), and the body analysis. Labeled Fact / Interpretation / Assumption where it matters. No price target, no buy/sell.

General

What thoughtful questions have other investors asked about this company? The most penetrating questions cluster around three themes. (1) Earnings quality and trust — given the E&Y resignation (Oct-2024), the delinquent FY24 10-K, the 2020 SEC settlement, and the March-2026 SDNY indictment of three associated individuals, can the reported numbers be relied upon, and is a restatement truly off the table? (2) Whether revenue growth is real or financed — how much of the +47% FY25 / ~$40B FY26 revenue ramp is durable demand versus inventory pre-build and receivables stuffing funded by debt and equity issuance? (Interpretation: with OCF of -$7.56B over 9M FY26 and inventory more than doubling, this is the central question.) (3) Whether there is any economic moat — investors repeatedly ask why gross margin is collapsing (18.0% FY23 → 8.2% 9M FY26) if SMCI genuinely has differentiated liquid-cooling/rack-scale IP. These are the right questions; the answers in this report are largely unfavorable.

Cyclicality & Earnings Nature

Are earnings at a cyclical high or low? Interpretation: revenue is at a cyclical high (AI capex super-cycle) while margins and earnings quality are arguably at a cyclical low. Net income actually fell FY24→FY25 ($1,153M → $1,049M) despite revenue rising +47% — a hallmark of a business capturing volume at the bottom of its own value chain while margin compresses. Earnings are highly exposed to the durability of hyperscaler/NVIDIA GPU spend.

Driven by external environment or internal actions? Fact/Interpretation: overwhelmingly external. The revenue surge is a direct function of NVIDIA GPU allocation and the AI build-out, not internally generated advantage. SMCI’s internal actions (DLC, building-block speed-to-market) help win share at the margin but do not create the demand and have not protected pricing.

How stable are revenues? Fact: highly unstable and concentrated — one customer was 63% of Q2 FY26 revenue and 27% in Q3 FY26, and quarterly results have historically swung sharply. Interpretation: revenue is lumpy, allocation-dependent, and project-driven (large-data-center deployments), the opposite of recurring.

Outlook for products/services? Fact: management guides ~$40B FY26 revenue. Interpretation: top-line outlook is strong in dollars but the product (GPU servers/racks) is a pass-through of NVIDIA silicon with thinning take-rate; the “service” content (SuperCloud Composer software, DLC integration) is small relative to hardware and not yet a margin engine.

How big will this market be — growing/shrinking, domestic/international? Fact: the AI server/data-center infrastructure market is large and growing; US was 69% of SMCI revenue. Interpretation: the market grows, but SMCI’s slice is the commoditized integration layer where ODMs (Foxconn, Quanta, Wiwynn) and OEMs (Dell, HPE) compete the spread toward zero. A growing market does not guarantee growing profit for an undifferentiated assembler.

Business Quality & Competitive Moat

Is the industry getting more or less competitive? Fact/Interpretation: more competitive. Dell and HPE have aggressively entered AI-server racks, ODMs sell direct to hyperscalers, and all chase the same NVIDIA allocation. SMCI’s margin collapse is the clearest evidence of intensifying competition and an absence of pricing power.

How profitable is the business (ROIC/ROE)? Fact: Third-party data providers report TTM ROE ~17.9% and ROA ~5.5% (unreconciled, third-party). Interpretation: headline ROE flatters a thin-margin business levered by $8.8B of debt and rapidly growing equity; on an unleveraged/ROIC basis the economics are weak — net margin ~3.7% and falling, op margin 4.6%. Returns are being manufactured by balance-sheet expansion, not durable profitability.

How profitable is the industry — how many competitors, barriers to entry? Fact: peer gross margins — Dell ~21%, HPE ~33%, Lenovo ~16%, ODMs ~5-8%; NVIDIA owns the profit pool. SMCI at 8.2% sits near the ODM floor. Interpretation: barriers to entry in box assembly are low; the only scarce input is GPU allocation, which NVIDIA controls — so the integrator captures little of the value.

Can the business be easily understood? Fact: the core model (buy GPUs/components, integrate into servers/racks, ship) is simple. Interpretation: but the financials are not easy to trust given the audit history, related-party web (Ablecom/Compuware/Liang family), and aggressive working-capital swings — comprehensibility of the business ≠ reliability of the accounts.

Can it be undermined by foreign low-cost labor? Fact/Interpretation: yes — it largely already is. Asian ODMs (Foxconn, Quanta, Wiwynn) operate the same assembly at ~5-8% margins and sell directly to hyperscalers, structurally capping SMCI’s pricing. Ablecom (Taiwan, Liang family) makes ~95.5% of SMCI’s chassis, underscoring that the manufacturing is low-cost-geography commodity work.

Do brands matter? Interpretation: minimally. In GPU racks the brand that matters is NVIDIA; the integrator’s brand carries little premium with sophisticated hyperscaler/large-data-center buyers (72% of revenue is OEM/large-data-center). The collapsing gross margin confirms negligible brand pricing power.

Nature of competition? Interpretation: competition is on speed-to-market, allocation access, and price — all replicable. SMCI’s historical edge was first-to-market on new NVIDIA platforms and early DLC scale; both are being matched by Dell, HPE, and ODMs, leaving price as the swing variable.

Customer switching costs? Interpretation: low. Buyers can and do multi-source server racks; the standardized NVIDIA reference architectures reduce lock-in. The 63%→27% single-customer swing across two quarters is itself evidence that demand is not sticky.

Financial Condition & Balance Sheet

Assets not fully recognized on the balance sheet? Interpretation: no material hidden asset of note. If anything, the concern runs the other way — the $11.1B inventory (up from $4.68B) and $8.41B AR (up from $2.20B) carry impairment/obsolescence and collection risk in a fast-moving GPU-generation cycle (Hopper→Blackwell transition risk). The intangible “DLC IP” is not a balance-sheet asset and, given commoditization, may be worth little.

Off-balance-sheet liabilities? Fact: related-party arrangements (Ablecom/Compuware manufacturing) and operating commitments exist; related-party purchases were ~2.1% of COGS in 9M FY26. The CEO holds a ~$12.9M unsecured personal loan from the spouse of a related-party principal (a governance flag, not a corporate liability). Open Question: full extent of purchase commitments to suppliers and any unconsolidated entity exposure warrants scrutiny given the related-party density.

How conservative is the accounting? Interpretation: not conservative, and the track record demands skepticism — 2018 delisting, 2020 SEC settlement (~$17.5M; CEO ~$2.1M clawback), E&Y resignation (Oct-2024), delinquent FY24 10-K. Aggressive revenue recognition was the prior issue; the current cash-conversion-cycle blow-out (54→106 days) and OCF deeply negative while net income is positive is a textbook earnings-quality red flag (Fact: 9M FY26 OCF -$7.56B vs. net income +$1,052M).

How CapEx-hungry? Fact: asset-light on PP&E (~$608M). Interpretation: but extremely working-capital-hungry — the real cash sink is inventory and receivables, not plant. This is the more dangerous form of capital intensity because it scales with revenue and is financed by debt/equity, producing deeply negative FCF in growth years.

Capital Allocation & Management

How much FCF, how used, philosophy? Fact: FCF is deeply negative in growth years (OCF FY24 -$2.49B, FY25 +$1.66B, 9M FY26 -$7.56B; minus modest capex). Interpretation: the company consumes cash to fund working capital and is financing growth with debt and equity rather than internal generation — a “grow at any cost / capture share” philosophy, not value-disciplined capital allocation.

Significant acquisitions? Fact: no material acquisitions noted; growth is organic (allocation-driven). Not applicable as a capital-allocation concern, but the absence of M&A also means no diversification away from the thin-margin core.

Buying back shares? Fact: no buybacks. Interpretation: appropriate given the cash burn, but it removes a shareholder-return lever and underscores that the equity is a funding source, not a return vehicle.

Issuing shares to insiders / dilution? Fact: heavy issuance — the June-2026 ~$7B raise ($1.25B common + $3.75B mandatory convertible preferred auto-converting ~2029 + $2.0B ATM) plus three convertible tranches (2028/2029/2030) and a 10-for-1 split (Oct-2024). Interpretation: shareholders face substantial dilution; the mandatory-convertible preferred and ATM are dilutive overhangs. Insider ownership is ~13.8% (Charles Liang, incl. related), aligning the CEO’s wealth with the stock but not insulating outside holders from dilution.

Director/management compensation? Open Question: detailed comp metrics require the proxy (DEF 14A). Fact: founder-CEO Charles Liang and co-founder/spouse Sara Liu hold significant equity; CFO David Weigand. Interpretation: founder-led with concentrated insider stake — alignment via ownership, but the related-party loan and family-supplier (Ablecom) web complicate the incentive picture.

Motivations of management? Interpretation: growth- and share-maximization, consistent with founder control. The combination of a controlling founder, related-party suppliers, a personal loan from a supplier-affiliated party, and a checkered governance/audit history means outside investors must weigh entrenchment and conflict-of-interest risk heavily.

Valuation & Market Data

ADR/MLP/K-1? Fact: none — SMCI is a Delaware C-corporation listed on NASDAQ; standard 1099 treatment, no K-1.

Dividend policy? Fact: no dividend (and none expected near-term given cash burn). Note: the new mandatory convertible preferred carries a cumulative dividend payable in cash or stock — a claim ahead of common, not a return to common holders.

How profitable is the business? Fact: marginally and deterioratingly so — gross margin 8.2% (9M FY26, from 18.0% in FY23), op margin 4.6%, net margin ~3.7%, net income roughly flat-to-down despite revenue more than tripling FY23→FY26. Interpretation: low and structurally pressured profitability for the scale of revenue.

Is net income diverging from cash from operations? Fact: yes, sharply — 9M FY26 net income +$1,052M vs. OCF -$7.56B. Interpretation: this is the single most important earnings-quality signal in the file; positive accrual earnings alongside large cash outflows (driven by inventory/AR build) warrant heavy skepticism about earnings durability.

Risks & Downside

What would cause the stock to decline? A negative finding from the ongoing independent investigation (Munger Tolles/AlixPartners) or any restatement; a single large customer (63% of Q2 FY26) reducing or cancelling orders; the Hopper→Blackwell GPU transition stranding the $11.1B inventory; further gross-margin compression below ~8%; a working-capital reversal forcing distressed financing; loss/curtailment of NVIDIA allocation; or dilution from the ATM/mandatory-convertible. Short interest of 18.1% of float and beta 1.68 amplify moves both ways (52-week range $19.48-$62.36).

Risk of a catastrophic loss? Interpretation: real and above-average. The convergence of customer concentration, inventory/receivable build financed by debt, an indictment of associated individuals, a history of accounting failures, and a single-supplier (NVIDIA) dependency creates multiple independent paths to a severe drawdown. Convertibles are all OTM (refs $55.20-$83.44 vs. ~$40.64), so debt would be repaid in cash, pressuring liquidity (cash only $1.29B).

Chance of a total loss? Interpretation: low but non-trivial — higher than for a typical profitable large-cap. The company is cash-flow-negative with $8.8B debt and a fragile trust profile; a restatement combined with a demand air-pocket and refinancing stress is the tail scenario. Assumption: absent fraud-level revelations, the equity is more likely to suffer a large drawdown than a zero.

Recent News & Events

Has the business environment changed recently? Fact: yes. Competitive intensity rose (Dell/HPE AI-rack entry), gross margin stepped down materially through FY25-FY26, and SMCI executed a ~$7B June-2026 capital raise — signaling the working-capital model now requires large external funding. The AI-capex tailwind remains but the terms of trade for the integrator have worsened.

Acquisitions? Fact: none material.

Change in accounting policies / auditor? Fact: E&Y resigned Oct-2024; BDO appointed; FY24 10-K delinquency cured Feb-2025; Nasdaq delisting avoided. The March-2026 SDNY indictment of three associated individuals (export-control/diversion to China) prompted an independent investigation (Munger Tolles, AlixPartners) that is ongoing; the company is not a defendant and management says no restatement is expected (Interpretation: management’s “no restatement” view is a hypothesis, not yet auditor-confirmed — treat as an open risk).

Recent changes — new markets, facilities, management? Fact: continued capacity expansion to support DCBBS rack-scale and DLC volumes; product push into direct-liquid-cooling and SuperCloud Composer software; 10-for-1 stock split (Oct-2024). Management remains founder-led (Liang/Liu/Weigand). Interpretation: the strategic direction (rack-scale + liquid cooling) is sound directionally but is being matched by competitors and has not arrested margin erosion.

APPENDIX B — Source Appendix

Company: Super Micro Computer, Inc. (NASDAQ: SMCI) — CIK 0001375365 Report date: 2026-06-10. Fiscal year ends June 30. Independent research; primary sources listed below.

Sources are grouped by type. Primary sources (SEC filings, company transcripts) take precedence; third-party aggregated data is reconciled to filings and labeled. Facts in the memo cite the underlying filing/transcript; interpretations and assumptions are labeled inline.

1. SEC filings — primary (EDGAR, CIK 0001375365)

Financial time series reconciled via EDGAR XBRL company facts (data.sec.gov): revenue tag RevenueFromContractWithCustomerExcludingAssessedTax; plus GrossProfit, OperatingIncomeLoss, NetIncomeLoss, NetCashProvidedByUsedInOperatingActivities, InventoryNet, CashAndCashEquivalentsAtCarryingValue, StockholdersEquity.

2. Company event transcripts — primary (earnings calls & investor conferences)

3. Quantitative / market data — third-party (reconciled to filings)

• Third-party market-data provider (fundamentals snapshot + own-history valuation percentiles) (accessed 2026-06-09/10): market cap, EV, TTM margins, ROE/ROA, short interest, ownership, own-history valuation percentiles. Treated as non-primary and reconciled to SEC filings.
• Third-party financial news aggregator (accessed 2026-06-10): June-2026 offering coverage and sentiment skew.
• public market data (yfinance) (accessed 2026-06-10): price, EV, peer cross-check (DELL/HPE). Unofficial; reconciled to filings.

4. Third-party research referenced as context only (NOT evidence)

• Hindenburg Research, “Super Micro Computer: Fresh Evidence Of Accounting Manipulation…” (2024-08-27) — short report; referenced for the timeline of allegations, not as a factual basis.
• U.S. SEC Accounting and Auditing Enforcement Release re the 2020 settlement (~$17.5M; CEO the Sarbanes-Oxley Section 304 clawback ~$2.1M) — referenced for the prior-controls-history base rate.

5. Analytical frameworks applied

• Greenwald & Kahn, Competition Demystified — barriers-to-entry / advantage-type taxonomy; market-share-stability and ROIC tests.
• Chancellor (Marathon), Capital Returns — supply-side capital-cycle analysis; high returns attract capital and mean-revert.

All URLs resolve under https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0001375365 for filings, and https://data.sec.gov/api/xbrl/companyconcept/CIK0001375365/... for XBRL facts. Accessed 2026-06-10.