Marvell Technology MRVL

Reference table

1. What the business actually is — and what the vendor file gets wrong about it

Marvell Technology, Inc. is a fabless semiconductor company headquartered in Wilmington, Delaware, with 7,480 employees and chief executive Matthew J. Murphy. It listed in 2000. Beta is 2.246.

The reporting structure changed, and the vendor did not fully follow it. The 10-K for fiscal 2026 (year ended 2026-01-31, filed 2026-03-11) and the 10-Q for the quarter ended 2026-05-02 both disaggregate revenue into two end markets: data center and communications and other. The vendor's seg_prod block reports, for fiscal 2026, a single line — "Data Center" $6.1003B — against reported revenue of $8.195B, leaving $2.095B (25.6% of the company) simply absent.

The vendor reports one segment totalling $6.100B against $8.195B of revenue; the 10-K reports two end markets; we use the filing. The derived split is corroborated by the filing to within half a percentage point:

Data centre is therefore 74.4% of the company and rising. The fiscal-2025 five-way vendor split (Data Center $4.164B, Enterprise Networking $626.4M, Carrier Infrastructure $338.2M, Consumer $316.1M, Automotive and Industrial $322.4M) does tie exactly to $5.767B of revenue and is used for the prior-year comparison — the defect is confined to the current year, where the vendor captured one of two reported end markets.

What is in the data-centre line. The 10-Q is specific: growth was "driven by AI-related demand for a broad range of our products, including electro-optics, custom, storage, and switching." Four product families, not one programme — which is a materially better risk profile than a single custom-silicon socket, and it is the single most reassuring sentence in the filing.

Two corporate actions dominate the recent history and both are filing-confirmed:

• Divestiture. On 2025-08-14 Marvell completed the sale of its automotive ethernet business to Infineon Technologies AG for $2.5B in cash, recording a pre-tax gain of $1.8B in the third quarter of fiscal 2026, booked in interest income and other, net. This is the single most important fact for reading the reported earnings (Section 2).
• Acquisitions. On 2026-02-02 Marvell completed the acquisition of Celestial AI, Inc., described in the 10-Q as "a provider of a Photonic Fabric technology platform purpose-built for next-generation scale-up interconnect," for total purchase consideration of $3.5B, plus contingent consideration payable in cash and shares through fiscal 2029 on revenue milestones. XConn and one immaterial acquisition closed in the same quarter. Acquisition-related transaction costs of $29.1M were recognised. The 10-Q states plainly: "Revenue and earnings of Celestial since the acquisition date were not material."

Geography — and why the raw geographic data misleads. Fiscal-2026 revenue by shipment destination: China $2.970B (36.2%), Other Countries $2.393B, Taiwan $1.657B (20.2%), United States $1.174B (14.3%). The 10-Q reports that 83% of April-quarter revenue shipped to customers with operations in Asia, up from 74% a year earlier. But the filing explicitly warns against the obvious inference: "these destinations of shipment are not necessarily indicative of the geographic location of the Company's end customers… a substantial majority of the shipments made to China relate to sales to non-China based customers that have factories or contract manufacturing operations located within China." The 36% China figure is a manufacturing-location artefact, not 36% of demand, and any read of export-control exposure from it would be wrong. We state it and decline to draw the conclusion.

2. The most important thing to understand: three separate reasons the reported earnings are not the earnings

Marvell's income statement currently contains three distortions running in different directions at the same time. Untangling them is the difference between a 75x price-to-earnings ratio and a ~153x one.

Distortion one — the $1.8B divestiture gain (inflates). The fiscal-2026 GAAP net income of $2.670B and diluted EPS of $3.07 include the $1.8B pre-tax gain on the Infineon sale, recorded in the quarter ended 2025-11-01. That quarter reported net income of $1.901B on operating income of $357.8M — a 5.3x ratio that is otherwise inexplicable. The 10-K confirms the mechanism directly: "We recognized interest and other income, net of $1.7 billion in fiscal 2026 as compared to interest and other loss, net of $174.4 million in fiscal 2025. The change was primarily due to the $1.8 billion gain on sale."

Distortion two — the contingent-consideration remeasurement (deflates, and is inversely tied to the share price). This is the finding that matters most, and it is available only from the filing. The Celestial contingent consideration "will be remeasured at fair value at each reporting date with changes recognized in Other expense, net." In the April-2026 quarter:

> "Interest and other loss, net increased by $201.4 million… primarily due to a $331.8 million increase in fair value of the contingent consideration liability associated with the Celestial acquisition, partially offset by an unrealized gain of $81.1 million from the forward stock purchase contract."

Because part of the contingent consideration is payable in Marvell common stock, the liability grows as the share price grows. The company entered a cash-settled forward stock purchase contract in April 2026 explicitly "to manage its exposure to changes in the Company's stock price related to the contingent consideration arrangement" — a partial hedge, which produced the $81.1M offsetting gain.

The consequence is counter-intuitive and should be stated bluntly: today's 12.81% share-price gain mechanically worsens the GAAP result Marvell will report on 2026-08-27. Management's own guidance already anticipates it — GAAP EPS of $0.37 ±$0.05 against non-GAAP EPS of $0.93 ±$0.05, a $0.56 gap. (Guidance is management framing and is half-weighted throughout this dive.)

Distortion three — the customer warrant recognised as contra-revenue (deflates the top line). During fiscal 2025 Marvell issued a warrant to a customer for up to 4.2 million shares at an exercise price of $87.77, with a seven-year exercise term and a five-year vesting term. The 10-Q: the shares "vest primarily based on the customer's achievement of qualifying product revenue milestones and are recognized as a reduction to revenue as qualifying revenues are recognized." Total grant-date fair value $227.6M ($54.44 a share); 0.9 million shares vested as of 2026-05-02. Reported revenue is therefore net of a non-cash charge that grows as that customer buys more — success is partially self-taxing at the revenue line.

Putting numbers on it. The quarterly record, with both bases side by side:

Read the last two highlighted rows together. In the quarter with the divestiture, GAAP EPS was $2.19 against $0.76 non-GAAP. In the quarter with the contingent-consideration remeasurement, GAAP EPS was $0.0386 against $0.80 non-GAAP. The business did roughly the same thing in both.

The three trailing multiples, all of them defensible and all of them different:

• Trailing GAAP (Q2 FY26 through Q1 FY27): net income $194.8M + $1,901M + $396.1M + $34.5M = $2.526B → the vendor's priceToEarningsDilutedRatioTTM of 75.2x. Flattered by the divestiture; rejected as a lens.
• Trailing non-GAAP (earn_cal): $0.67 + $0.76 + $0.80 + $0.80 = $3.03 → 72.1x. Excludes $590.8M of annual stock-based compensation, which is a real cost paid in dilution — see the share count rising from 856.2M to 893.3M.
• Our derived underlying GAAP: strip the $1.8B gain and add back the $331.8M remeasurement, both at the trailing effective tax rate of 13.29%: $2.526B − $1.561B + $0.288B ≈ $1.25B, or roughly $1.40 a share → ~153x. Labelled as our derivation, not a reported figure, and stated to make one point only: the genuine current-earnings multiple is roughly double what any screen shows.

This dive values Marvell on forward non-GAAP EPS, because that is the basis the estimate block is on and the basis on which management guides — while stating explicitly that stock-based compensation of $590.8M (7.5% of fiscal-2026 revenue, 42% of that year's free cash flow) is excluded from it and is not free.

3. Growth and margin — the part of the story that holds up

Revenue. $2.969B (FY2021) → $4.462B → $5.920B → $5.508B → $5.767B (FY2025) → $8.195B (FY2026, +42.1%). Note the FY2023-FY2025 plateau: revenue actually fell in fiscal 2024 and barely recovered in fiscal 2025, with operating losses in both. This is not a company that has grown smoothly; it is a company that spent two years going nowhere and then inflected. The knowledge base's own framing — "left for dead in the LLM boom" — is factually accurate against this series.

Margin. Gross margin: 50.1% (FY2021), 46.3%, 50.5%, 41.6%, 41.3%, 51.0% (FY2026) — and 52.1% in the April-2026 quarter, up 1.8 percentage points year on year, which the 10-Q attributes to "better cost absorption driven by higher revenues, partially offset by a shift in product mix." The single ugly data point in the record is Q3 FY2025 at 23.0% gross margin with a −$702.8M operating loss; that is a restructuring/impairment-affected quarter in the middle of the plateau and it is a useful reminder of what this business looks like when the cycle is against it.

Operating leverage is real but modest so far. Operating income: −$720.3M (FY2025) → $1.338B (FY2026), a $2.06B swing on $2.43B of incremental revenue. But research and development remains 25.5% of revenue on a trailing basis and rose $144.6M year on year in the April quarter, "primarily due to higher overall spending… including increased employee compensation and related costs, primarily driven by growth in headcount including the addition of new employees from our recent acquisitions." Marvell is buying growth and staffing to it.

The forward path (consensus, epsAvg basis — the ebitAvg rows are rejected, see Section 8):

Two observations. First, the FY2027 and FY2028 rows are unusually tight — a $0.28 spread on $4.05, and a $1.04 spread on $6.25, across 24-25 analysts. The street agrees about the next two years to a degree that is rare in semiconductors, which raises the cost of being wrong in either direction. The FY2029 row is the opposite: $5.28 to $11.55 on 13 analysts, a 2.2x spread. That is where the genuine uncertainty sits and it is what section 6a has to price.

Second, and flagged rather than smoothed: the FY2031 row shows revenue FALLING 22.8% from FY2030's $28.308B to $21.861B, on more analysts (22 versus 13). Either the street's long-range model contains an explicit cyclical downturn, or the row is contaminated. We do not use it, and we note that a five-year estimate block that turns down at the end is at least an honest reminder that this industry has cycles.

4. Balance sheet, the debt correction, and the $2.0B instrument the vendor cannot see

As at 2026-01-31 (fiscal-2026 year end), with 10-Q updates to 2026-05-02:

• Cash and equivalents $2.639B (2026-01-31); $3.8B at 2026-05-02 per the 10-Q, "$1.2 billion higher than our balance at January 31, 2026"
• Short-term investments $0
• Inventory $1.388B · total assets $22.285B · total liabilities $7.977B · total equity $14.308B
• Goodwill $11.062B + intangible assets $2.051B = $13.113B, or 58.8% of total assets
• Book value per share $20.65; tangible book value per share $1.69

Correction one — total debt is overstated by $319.7M. The 10-K states, twice and unambiguously: "As of January 31, 2026, we had a total of $4.5 billion debt outstanding, which consisted of senior notes outstanding" and "total borrowings outstanding of $4.5 billion, consisting of senior notes outstanding, of which $499.8 million are due within twelve months." The vendor's totalDebt is $4.790B, arrived at as short-term debt $499.8M + long-term debt $3.971B + capitalLeaseObligations $319.7M. The first two sum to $4.471B, which matches the filing (and the $499.8M current portion matches to the dollar). The vendor has added lease obligations on top of a debt figure the company defines as senior notes only. The vendor reports total debt of $4.790B; the 10-K shows $4.5B of senior notes; we use $4.471B and treat leases separately. This is the documented capitalLeaseObligations failure mode, caught exactly as the data contract predicts.

The revolving credit facility was amended and restated in June 2025 at $1.5B capacity and was undrawn at 2026-01-31, available through 2030-06-30, with covenants in compliance. In fiscal 2026 Marvell repaid $590.6M of the 2026 Term Loan and, on 2025-06-30, issued $500.0M of 2030 senior notes and $500.0M of 2035 senior notes.

Corrected net debt: $4.471B of senior notes less the 10-Q's $3.8B of cash at 2026-05-02 ≈ $0.671B net debt. The vendor's trailing enterprise value of $192.867B implies net debt of $1.433B. Corrected enterprise value (common only) ≈ $192.105B.

Correction two, and the larger one — $2.0B of convertible preferred sits entirely outside the vendor's market capitalisation. From the 10-Q:

> "On March 31, 2026, the Company completed the issuance and sale of 2.0 million shares of Series A Convertible Preferred Stock… to NVIDIA Corporation… for an aggregate purchase price of $2.0 billion. Each share… has a stated value of $1,000 and is initially convertible, at the option of the holder, into shares of the Company's common stock at an initial conversion price of approximately $91.84 per share into an aggregate maximum of approximately 21.8 million shares of common stock… As of May 2, 2026, no shares of Series A Preferred Stock had been converted."

The preferred participates in dividends and liquidation on an as-converted basis, has no redemption or preemptive rights, and voted on an as-converted basis at the 2026-06-25 annual meeting (per the 8-K of that date). The 10-Q carries a dedicated risk factor: "our issuance of preferred stock to NVIDIA may reduce the market price of our common stock."

At $218.59 the conversion price of $91.84 is 138% in the money. Those 21.8M shares are worth $4.765B against the $2.0B paid four months ago.

• Vendor market capitalisation (common only): $191.434B
• Fully converted market capitalisation: ≈$196.20B
• Fully converted enterprise value: ≈$196.87B

Every per-share and enterprise-value multiple in the vendor file understates the capitalisation by roughly 2.5%. That is not fatal, but it is real, it is undisclosed by the feed, and it compounds: bookValuePerShareTTM, priceToBookRatioTTM (10.58x), evToSalesTTM (22.13x) and enterpriseValueMultipleTTM (41.76x) are all computed on a share count that excludes 21.8M deep-in-the-money shares. We use the fully converted figures throughout: 22.6x trailing revenue on the corrected enterprise value of $196.87B against trailing revenue of $8.718B.

Non-equity tripwire — checked and passed for the common. MRVL common is ordinary equity: price $218.59 (not par-like), beta 2.246 (not low), dividend $0.24 a year yielding 0.11% (nominal, not fixed-income), volume 33.1M shares, and a 52-week band of $62.31 to $316.43 — a 5.1x range, the opposite of narrow. It is unambiguously common stock. The relevant caution here is the mirror image: the company has $2.0B of convertible preferred outstanding that the equity feed does not surface, which is why Section 4 exists.

5. Cash flow and capital returns — funded by an asset sale, not by operations

The arithmetic that matters: in fiscal 2026, free cash flow of $1.396B was less than the $2.245B returned to shareholders. The gap was covered by the $2.5B of Infineon proceeds — the 10-K confirms "Net cash provided by investing activities of $2.1 billion in fiscal 2026 was primarily driven by net proceeds from sale of our automotive ethernet business of $2.5 billion." The record buyback was funded by selling a business, not by operating the remaining one. That is not illegitimate, but it is not repeatable, and any read of the fiscal-2026 buyback as a run-rate is wrong.

And the buyback did not shrink the share count. Diluted shares: 856.2M (Q4 FY26) → 893.3M (Q1 FY27), +37.1M or +4.3% in a single quarter, because Celestial and XConn were paid for partly in stock and the assumed Celestial equity plan added further option and restricted-stock overhang. $2.040B of repurchases in fiscal 2026 did not offset the dilution from one quarter of acquisitions. On top of that sit 21.8M as-converted preferred shares and up to 4.2M customer-warrant shares.

Capital return in the April-2026 quarter, per the 10-Q: "We returned $253.8 million to stockholders… through $200.0 million in repurchases of shares of our common stock and $53.8 million in cash dividends." Annualised, that is roughly $1.02B, or 0.53% of the market capitalisation — immaterial as a return component at this valuation.

Free-cash-flow yield is 0.87% and price to free cash flow is 115x. Both are accurate and both simply confirm that nothing about this investment case rests on current cash generation.

6. Valuation — priced in or room?

At $218.59 (fully converted market capitalisation ≈$196.20B, corrected enterprise value ≈$196.87B):

Peer context. The vendor peer set for MRVL is close to useless — Fortinet, Corning, Infosys, Monolithic Power, Strategy Inc, Cloudflare, NXP, Roblox, Synopsys and TE Connectivity are a grab-bag of unrelated businesses, and no peer multiples are supplied. No peer-multiple comparison is drawn. The genuinely useful comparison is within this batch: AMD trades at 22.5x trailing sales and MRVL at 22.6x — two very different businesses at an identical revenue multiple, which is worth holding in mind. The vendor's own rating model scores Marvell B− with an overall score of 2, its price-to-earnings and price-to-book components both at 1 out of 5 — the model is registering the valuation, not the growth.

6a. What today's price assumes (the inversion)

At $218.59 — 54.0x January-2027 consensus and 8.5x January-2029 consensus revenue — the price embeds roughly the following falsifiable claims:

• Revenue compounds ~41.5% a year for three more years, reaching $23.233B by January 2029. (Consensus-derived: $8.195B actual FY2026 → $11.520B / $16.696B / $23.233B, on 29/29/26 analysts.) The company itself has guided "to roughly 40% growth for the next few years" per a knowledge-base claim sourced to the chief executive — management framing, half-weighted. This is the load-bearing assumption and it is remarkably close to what management says, which is itself a warning: the market is currently paying full price for guidance being met.
• Non-GAAP EPS reaches $6.245 by January 2028 and the market still pays ~40x it. (Consensus EPS; the 40x is our number.) At today's price that is 35.0x — so the base case requires only that the multiple not compress much, not that it expand.
• Gross margin holds near 52-53% on a GAAP basis (company guidance for the July quarter is 52.1-53.1% GAAP, 58.25-59.25% non-GAAP — half-weighted) while custom silicon grows as a share of mix. Custom XPU work is structurally lower margin than electro-optics; a mix shift toward it is a margin headwind the consensus path does not obviously accommodate.
• No large customer insources. (Our reading of the 10-Q risk factors, which name this explicitly.) This is the most fragile assumption in the price and the one no number in the file can bound. Three customers at 75% of gross receivables means the loss of one is not a growth-rate question, it is a step change.
• The 2026-05-18 caution — that these names had reached "exhaustion levels" and that "much is already priced in" — does not recur. (Knowledge-base claim, Stephanie Guild, skill 0.8, conviction 55.) It was made when the stock was near its high and it was correct; the stock is 30.9% below that high. The price today assumes the drawdown is over.

6b. The return bridge (why the multiple moves)

Expected return over the 12-24 month horizon decomposes as: non-GAAP EPS growth (+54% to January 2027, +54% again to January 2028) + multiple COMPRESSION (from 54.0x FY27E to an assumed ~40x FY28E) + shareholder yield (~0.5%, immaterial) − dilution (~4% a quarter in the acquisition quarter, and structurally 1-2% a year from stock compensation).

Our base case assumes the multiple COMPRESSES, and that is deliberate. The entire expected return is earnings growth; we are explicitly not asking the market to pay more per dollar of earnings than it does now. From $218.59 to $250 is +14.4% while forward earnings roughly double — meaning the forward multiple falls from 54.0x (FY27E) to 40.0x (FY28E). If the market instead holds 54x on the January-2028 number, the price is $337 and the bull case is met without any thesis change. If it compresses to 25x — the multiple at which the stock traded in a less enthusiastic regime — the price is $156 and the bear case is met without any earnings change.

That is the honest shape of this investment: the earnings path is unusually well agreed and the multiple is not. The variance in outcomes is almost entirely multiple variance, which is another way of saying it is sentiment variance — and this stock has just demonstrated, with a 30.9% drawdown followed by a 12.81% day, exactly how much sentiment variance it carries.

Dilution deserves a line in the bridge because it is unusually large here. Between the acquisition stock, the assumed Celestial equity plan, $590.8M of annual stock-based compensation, 21.8M as-converted preferred shares and up to 4.2M warrant shares, the share count is going up, not down, and the buyback has not been sufficient to offset it. Per-share earnings growth will lag total earnings growth, and the consensus EPS figures already reflect analysts' own dilution assumptions — which may or may not include the preferred.

6c. Variant perception (where we differ, what would surprise)

• We differ from the street on what today's move means, in the constructive direction. The consensus target of $251.21 is +14.9% from here, which is almost exactly our base case — so on level we have no edge and say so. Where we differ is on structure: today's 12.81% took Marvell from 38.7% below its 52-week high to 30.9% below it, while the other semiconductor names in this batch gapped at or through their highs. We think buying a 12.81% day that ends below the 50-day moving average is a categorically different act from buying a 12.81% day that ends at a record, and we are willing to act on this one where we would not on some others. Watchable number: reclaim of the 50-DMA at $240.15.
• We differ on the reported earnings, sharply. Every screen shows a trailing price-to-earnings ratio of 75x. Strip the $1.8B Infineon gain and add back the $331.8M contingent-consideration remeasurement and the underlying figure is roughly 153x — double. We think a material part of the crowd owns this stock on a P/E that does not exist. Symmetrically, the GAAP print on 2026-08-27 will look bad by construction (guided $0.37 against $0.93 non-GAAP), and a sell-off on that GAAP number would be an opportunity rather than information. That is a specific, dated, falsifiable variant view. Watchable number: the contingent-consideration line inside Other expense, net, in the FQ2 filing.
• We differ on the balance sheet in two places. The vendor's total debt of $4.790B includes $319.7M of lease obligations the 10-K excludes from its own $4.5B senior-note figure; and the vendor's $191.434B market capitalisation excludes 21.8M as-converted preferred shares worth $4.765B. Neither correction changes the verdict, but together they mean every enterprise-value multiple on every screen covering this name is wrong in the same direction.
• We do NOT differ on the growth, and we want to be clear about that. The 42.1% fiscal-2026 revenue growth, the 46% data-centre growth and the 28% April-quarter growth are all confirmed by the filings in the company's own words. We are not making a contrarian growth call. Our variant perception is entirely about entry structure and about what the reported earnings actually are — which is a smaller edge than a growth call, and the verdict is Tactical rather than Core partly for that reason.
• Positive surprise that would force a re-rate: data-centre revenue growth re-accelerating above the 27% posted in April; Celestial revenue becoming material (the 10-Q says it was not); the strategic preferred converting, which would signal the holder wants common exposure rather than a security; or a disclosed new custom-silicon programme that broadens the customer base away from three. Watchable number: the customer-concentration disclosure in the next 10-Q — if the number of 10%-plus customers goes from three to four or five, that is a genuine de-risking.
• Negative surprise that would break the thesis: a large customer insourcing, disclosed or inferable from a guidance cut; gross margin falling below 50% on GAAP as custom mix rises; an equity offering following the 2026-07-09 prospectus supplement; or the FY2029 estimate range resolving toward its $5.28 low rather than its $11.55 high, which would put the stock at 41x rather than 24x on that year.

Synthos fair values

All three anchors are multiples of the January-2028 consensus non-GAAP EPS of $6.245 (25 analysts — the best-covered forward year that is far enough out to matter), cross-checked against January-2027 EPS of $4.051 and against enterprise value to January-2029 revenue.

• Bear ~$156 — 25x FY28E. Cross-checks: 38.5x FY27E; 6.7x FY29E revenue; and it happens to sit almost exactly on the street's low target of $155. The scenario: a large customer insources or delays, revenue growth decelerates to 20%, custom-silicon mix pulls gross margin below 50%, and the multiple compresses to what the market paid for this business before the inflection. −28.6%.
• Base ~$250 — 40x FY28E. Cross-checks: 61.7x FY27E; 10.8x FY29E revenue; 26.9x FY29E EPS. The scenario: the consensus path is met, the multiple compresses from 54x to 40x as it should when growth matures, and the return is earnings growth net of dilution. +14.4%, plus a negligible 0.5% shareholder yield. Note this sits 0.5% below the street consensus of $251.21 — we have essentially no edge on level and say so plainly.
• Bull ~$345 — 55x FY28E. Cross-checks: 85.2x FY27E; 14.9x FY29E revenue. The scenario: the co-packaged-optics transition arrives on the timeline the technical claims imply, FY2029 resolves toward the upper half of its $5.28–$11.55 range, the customer base broadens, and the market continues to pay a scarcity multiple for the only merchant supplier positioned on both custom silicon and optical interconnect. +57.8%.

Base is 14.4% above spot; asymmetry roughly 2.0:1 to the upside (28.6% down, 57.8% up). That asymmetry, not the base case, is what justifies the position — and it exists only because the stock is 30.9% below its high. It would not exist at $316.

7. Knowledge base — 38 raw hits, 33 used, 5 discarded, and a breadth problem worth naming

Raw hits: 38. Used: 33. Discarded: 5. Searches covered MRVL and Marvell (word-boundary, entities first, then claim text) across 51,928 records.

Discards, with reasons:

• 1 text-only match (2026-04-26) where the entity list is a different company and Marvell appears inside a claim arguing that a competitor would not be outperformed — the claim is about the other company, not this one, and it is demoted rather than used as MRVL conviction.
• 2 claims carrying the thinker label quarantine_misattributed (2026-03-08 and 2026-06-04). Our base has flagged these itself as attribution failures. We do not use claims our own attribution layer has quarantined, even though both are bullish and one is specific ("wrote a paper on it at $90 and now ~$165, still has a doubling to go"). Using a quarantined claim to support a Buy would be exactly the failure the quarantine exists to prevent.
• 2 further claims demoted to zero weight as generic sector commentary in which Marvell is one name in a list of six.

The composition of what remains is the finding, and it is a mixed one:

Two attribution problems must be stated. First, five claims dated 2026-06-02 are shelved in our base under the thinker jensen_huang — the channel — while the speaker field correctly records "Matt Murphy", who is Marvell's own chief executive. Under the four-lane attribution policy these are management claims and are half-weighted here, not independent third-party conviction. Attributing them to the channel would have counted the company's own advocacy as external validation. This is precisely the channel-versus-speaker defect and it is caught and corrected here.

Second, 17 of 33 usable claims come from a single thinker. Direction is not the issue; breadth is. This is conviction in one analyst's persistent, evolving, publicly-updated position — valuable, but it is one opinion sampled seventeen times, not seventeen opinions.

The strongest independent claims, verbatim:

> 2025-12-07 · bullish · conviction 70 · Jordi Visser (skill 2.0) · entity: MRVL

> "Marvell, left for dead in the LLM boom, is finally breaking out as it becomes part of the data-center build — AI silicon/photonics, aided by the Celestial acquisition for optical interconnect."

This claim is now checkable, and it checks out. It was made five days after the Celestial merger agreement was signed (the 10-Q dates the Agreement and Plan of Reorganization to 2025-12-02) and roughly two months before the deal closed on 2026-02-02 for $3.5B. The claim's premise is confirmed by the filing.

> 2026-05-23 · bullish · conviction 70 · Jordi Visser · entity: MRVL

> "Marvell only broke out 2 months ago, not long in the tooth; the optical side is so important it can be a Micron-type name over the next 2 years."

> Registered test: "Marvell becomes a Micron-type winner over the next 2 years"

> 2026-05-09 · bullish · conviction 70 · entities: MRVL, MU

> "Marvell will outperform Micron from here; only trimmed ~10% of Marvell for discipline — the optical-fiber/interconnect data-center need is still building."

> Registered test: "Marvell outperforms Micron going forward"

This test is now gradeable and it is currently FAILING. From 2026-05-09 to today, Micron closed at $892.67 (+7.62% today alone, market capitalisation $1.008T on a memory supercycle); Marvell is 30.9% below its 52-week high. Both names are covered in this batch and the divergence is stark. We report this because the same lane supplies most of our bullish evidence, and a lane whose registered test is failing must be weighted accordingly.

The one cautionary claim, and it was right:

> 2026-05-18 · neutral · conviction 55 · Stephanie Guild via darius_dale (skill 0.8) · entities: MRVL, COHR

> "On technicals, high-flying tech names like Marvell and Coherent have gotten too high, near exhaustion levels; a pullback is healthy and much is already priced in."

Made near the highs; the stock is now 30.9% below them. A lane in which the single dissenting voice was vindicated and the dominant voice's registered test is failing is not a lane that should carry a Core rating, however bullish its net direction. It is a Moderate.

The technical claims — sourced to the chief executive, half-weighted, but structurally useful:

> 2026-06-02 · bullish · conviction 82 · speaker: Matt Murphy (Marvell CEO — COMPANY VOICE, half-weighted)

> "The copper wall is moving into the rack itself; at 400 gig you can no longer fully connect a rack with copper, so intra-rack links go optical (CPO), driving an explosion in optical demand."

> Registered test: "at 400 Gbit/s rack connections can no longer be fully copper and become optical"

This is the clearest statement of the physical mechanism behind the exponential score. It comes from the company and is discounted as such — but it is a testable engineering claim rather than a forecast, and it is the sort of assertion that is either true or not.

One knowledge-base figure we could not reconcile, reported rather than suppressed. A 2026-07-19 claim states Marvell was "down ~62% in the correction but still his biggest semi position." Against our price data the maximum drawdown from the 52-week peak is −30.9% ($316.43 to today's $218.59), and even measuring to the recent low we cannot produce −62%. We cannot reconcile the figure and therefore do not use it; the positioning statement it accompanies (still the largest semiconductor position besides one other name) is used, the number is not.

8. Data integrity — what we rejected from the vendor file and why

Eight material items. Two are filing contradictions.

1. seg_prod reports ONE end market for fiscal 2026 against reported revenue 34% larger — FILING WINS. The vendor supplies "Data Center" $6.1003B only. The 10-K and 10-Q both disaggregate into two end markets, data center and communications and other. The vendor reports $6.100B of segment revenue against $8.195B of total revenue; the 10-K shows two end markets; we use the filing and derive communications and other at $2.095B. The derived growth rates (+46.5% and +30.7%) match the 10-K's stated +46% and +31% — a clean corroboration. The fiscal-2025 five-way split does tie to revenue and is used for comparison.

2. totalDebt overstated by $319.7M — FILING WINS. The vendor reports $4.790B, comprising short-term debt $499.8M + long-term debt $3.971B + capitalLeaseObligations $319.7M. The 10-K states total borrowings of $4.5B "consisting of senior notes outstanding," of which $499.8M are due within twelve months. The vendor reports total debt of $4.790B; the 10-K shows $4.471B of senior notes; we use $4.471B. This is the documented lease double-count.

3. Market capitalisation and every EV multiple exclude $2.0B of deep-in-the-money convertible preferred. 2.0 million Series A preferred shares issued 2026-03-31, convertible at ~$91.84 into ~21.8 million common shares — 138% in the money at $218.59, worth $4.765B. The vendor's $191.434B market capitalisation is common-only. We use a fully converted capitalisation of ≈$196.20B and enterprise value of ≈$196.87B throughout, and flag that priceToBookRatioTTM (10.58x), evToSalesTTM (22.13x) and enterpriseValueMultipleTTM (41.76x) are all computed on the understated count.

4. Trailing GAAP earnings multiples are economically meaningless — REJECTED as a lens. priceToEarningsRatioTTM 74.37x and priceToEarningsDilutedRatioTTM 75.15x are arithmetically correct on trailing GAAP net income of $2.526B, but that figure contains a $1.8B pre-tax divestiture gain in one quarter and a $331.8M contingent-consideration charge in another. Our derived underlying figure is roughly $1.25B, giving ~153x. Both the vendor number and ours are stated; the vendor number is not used.

5. est.ebitAvg is internally impossible across every forward year — REJECTED. FY2027: EBITDA $2.999B, EBIT $324.8M, net income $3.532B. Net income exceeding EBIT by an order of magnitude is arithmetically impossible in the absence of enormous non-operating income, and the same pattern repeats in FY2028 ($4.347B / $470.7M / $5.671B), FY2029 and FY2030. All ebitAvg and ebitdaAvg rows are rejected and no EV/EBITDA multiple appears in this dive. epsAvg is used exclusively, per the data contract. Note also ratios_ttm.ebitMarginTTM of 38.28% against operatingProfitMarginTTM of 16.20% — the same defect in the trailing block; the operating figure is the one consistent with the income statement.

6. The FY2031 estimate row shows revenue falling 22.8% on MORE analysts than FY2030 — flagged, not used. $28.308B (FY2030, 13 analysts) → $21.861B (FY2031, 22 analysts), with EPS falling from $11.921 to $9.400. Either the street models an explicit cycle in that year or the row is contaminated. No conclusion in this dive rests on FY2030 or FY2031.

7. The 52-week high is reported inconsistently within the same file. quote.yearHigh is $329.88; the computed tech block reports hi52 of $316.43 with pct_from_hi of −30.92%, which reconciles exactly to $316.43 at today's price. quote.yearLow is $61.44 against tech.lo52 of $62.31. We use the tech block values ($316.43 / $62.31) because they are internally consistent with the computed drawdown, and we note that using the quote's $329.88 would make the drawdown −33.7% — directionally identical and, if anything, more favourable to the entry argument.

8. Revenue is reported net of a contra-revenue customer warrant, and no screen adjusts for it. Up to 4.2M shares at $87.77, grant-date fair value $227.6M, recognised as a reduction to revenue as the customer's qualifying revenue is earned; 0.9M vested at 2026-05-02. Reported revenue understates gross commercial activity by a non-cash amount that grows with that customer's purchases. Stated, not adjusted for, because the filing does not quantify the period charge.

Disclosure the filing contains but the extracted text lost. The 10-K and 10-Q both introduce a table of "net revenue attributable to significant customers… whose revenues represented 10% or more of total net revenue" — but the table values did not survive text extraction, so we have the accounts-receivable concentration (three customers, 75%) and not the revenue concentration. We state the limitation rather than inferring the revenue figure from the receivables figure, which would be unsound given differing payment terms.

Not defects, correctly reported: dividendYieldTTM of 0.11% on a $0.24 annual dividend is consistent with the cash-flow statement's $205.1M of dividends paid; intangiblesToTotalAssetsTTM of 62.07% is consistent with the balance sheet on a trailing basis (58.8% at fiscal year end, higher after the Celestial goodwill addition); netDebtToEBITDATTM of 0.31x is directionally right even though the EBITDA input is suspect.

9. Technicals — and the entry question this batch is really about

• Price $218.59, up 12.81% from $193.775, on 33.1M shares, with a day range of $208.00 to $222.67 — closing within 1.8% of the day's high, which is the constructive detail.
• −30.9% from the 52-week high of $316.43. Also +250.8% above the 52-week low of $62.31. The annual range is 5.1x wide, which is the volatility this position carries.
• BELOW the 50-day moving average of $240.15 by 9.0%. ABOVE the 200-day moving average of $136.25 by 60.4%. The 50-DMA is declining; the 200-DMA is rising steeply.
• RSI 54.3 — merely neutral after a 12.81% day, which is the clearest single indicator of how deep the preceding decline was. MACD −13.32, still negative.
• Relative performance: 3-month +33.6% vs SPY +7.6%; 6-month +177.9% vs SPY +11.1%; 12-month +193.6% vs SPY +24.3% and QQQ +30.8%. This name has already produced an extraordinary run and then surrendered roughly a third of it.

Today's move and what it does to the entry

This is the load-bearing paragraph of the dive, and the honest answer is that Marvell is the one semiconductor name in this batch where today's gap improves rather than degrades the entry — but only relatively.

Today's closing moves across the batch: MRVL +12.81%, INTC +10.92%, LRCX +7.85%, MU +7.62%, KLAC +6.95%, AMAT +5.48%, AMD +7.00%, ASML +4.22% — while AMZN fell 2.32%. The move is real, it is broad, and it is the central entry question for the whole batch.

What distinguishes Marvell is where the gap started and where it ended. Marvell opened at $208.00 and closed at $218.59 — and finished the day still 30.9% below its 52-week high and still 9.0% below its 50-day moving average. It did not break out. It bounced inside an existing drawdown. Buying it tomorrow means paying 12.81% more than yesterday's close for something that remains a third below where the same crowd was willing to pay for it four months ago.

That is a genuinely different act from buying a name that gapped to a record, and it is why this is the semiconductor position in the batch we are most comfortable opening — while still refusing to open it all at once.

Three things argue against buying the whole position today:

1. The 50-DMA at $240.15 has not been reclaimed. Until it is, today reads as a bounce, not a trend change. That is 9.9% above the close.

2. Earnings land 2026-08-27, 23 days out, and management has already guided the GAAP number to $0.37 against $0.93 non-GAAP. A GAAP print that looks like a collapse is expected — but it will be reported as a collapse, and the mechanism (Section 2) means today's rally makes it worse.

3. The president and chief operating officer sold 10,000 shares at $180.50 on 2026-08-03 — the session before this one (Section 10).

Two argue for starting now:

1. The asymmetry only exists at this price. At $218.59 the bear/bull spread is −28.6% / +57.8%. At $316 it was roughly −51% / +9%. The drawdown is the opportunity.

2. RSI at 54.3 after a 12.81% day means the move has not created a technically stretched condition. There is no overbought signal to wait out.

Practical read: stage it. A reasonable construction is roughly a third now, a third on either a reclaim of the 50-DMA at $240.15 or a retest toward the recent lows, and the final third after the 2026-08-27 print — explicitly accepting that a bad-looking GAAP number is the most likely path to the better price. This is an accumulation with a known event inside it, not a trade.

10. Insiders — one discretionary sale, one day before the move

The reading, stated carefully. Seven of the eight transactions belong to a single officer and are entirely mechanical: restricted-stock units vesting at a $0 exercise price with shares withheld in kind at $206.26 to cover the tax liability. These carry no signal in either direction and should not be characterised as selling.

The eighth does carry signal, and the timing is uncomfortable. The president and chief operating officer sold 10,000 shares at $180.50 on 2026-08-03 — an outright sale at a market price, filed the same day, one session before the stock closed at $218.59. That is a 21.1% move away from him in a single day.

We do not read this as a bearish signal about the business, and we say why: the residual holding of 227,941 shares (worth $49.8M at today's price) is 22.8x the size of the sale, the disposition is recorded as indirect, and a 4.2% trim by an executive whose compensation is overwhelmingly equity is ordinary diversification. What it does establish is that no insider was positioned for today, which is a mild caution against reading the move as informed anticipation of anything. There is no insider buying anywhere in this file.

11. Verdict, kill-criteria and flip conditions

Buy — Tactical.

The case for. The growth is real and the filings say so in the company's own words: revenue +42.1% in fiscal 2026 to $8.195B, the data-centre end market +46% to $6.100B and now 74.4% of the company, the April-2026 quarter +28% with gross margin up 1.8 points, and consensus modelling a 41.5% compound rate to $23.233B by January 2029 with unusually tight agreement through January 2028. Two well-informed parties have put capital behind the interconnect thesis in the last six months — $3.5B paid for Celestial's photonic-fabric platform, and $2.0B of convertible preferred taken by the dominant GPU vendor at a $91.84 conversion price. The knowledge base is 30-of-33 bullish. And uniquely in this batch, today's 12.81% gap left the stock 30.9% below its high rather than at a record, which is what creates the 2.0:1 asymmetry the position rests on.

What you are accepting. 22.6x trailing revenue. Three customers at 75% of gross accounts receivable, and rising. Tangible book value per share of $1.69 against a $218.59 price, because 58.8% of the assets are goodwill and intangibles. A share count that went UP 4.3% in one quarter despite $2.040B of buybacks — buybacks funded by selling a business rather than by operating one. A GAAP earnings line that is now mechanically inverse to the share price. An earnings print in 23 days that management has already told you will show $0.37 of GAAP earnings. A knowledge-base lane that is 17-of-33 one voice, whose registered test — that Marvell would outperform Micron — is currently failing badly.

Why Tactical and not Core. Three reasons, each independently sufficient: the base case of $250 is 0.5% below the street consensus of $251.21, so we have essentially no edge on level; the conviction lane lacks breadth; and there is a known, dated, adverse-looking event 23 days out. A Core rating requires an edge, breadth, or a clean window. This has none of the three. What it has is an asymmetry and a favourable entry structure, which is exactly what Tactical is for.

Pre-registered KILL criteria — what would break the thesis:

• Any disclosure that a 10%-plus customer is insourcing, has moved a programme, or has cut a commitment. This is the one event that turns a growth-rate question into a step change. It would most likely appear as a guidance cut without a demand explanation.
• The number of 10%-plus customers falling, or accounts-receivable concentration rising above 75% in the next 10-Q.
• GAAP gross margin falling below 50% for two consecutive quarters, indicating that custom-silicon mix is compressing structural profitability rather than that a single quarter absorbed a charge.
• Data-centre revenue growth decelerating below ~20% year on year — the April quarter was 27%, and the consensus path needs it materially higher, not lower.
• A primary equity offering following the 2026-07-09 prospectus supplement, particularly one priced at a discount, given that the share count is already rising.
• Goodwill impairment on Celestial or the earlier acquisitions. With $13.1B of goodwill and intangibles against $14.3B of equity, an impairment is a book-value event, not a footnote.
• A decisive break below the recent lows on company-specific news rather than sector beta.

Pre-registered FLIP TO CORE (upsize):

• A reclaim of the 50-day moving average at $240.15 that holds, confirming today was a trend change.
• A disclosed fourth or fifth 10%-plus customer — the single most valuable de-risking event available to this company.
• Celestial revenue becoming material and disclosed as such, validating the $3.5B.
• The Series A preferred converting to common, which would signal the strategic holder prefers equity exposure to a preferred claim.
• The FY2029 estimate range narrowing upward from its current $5.28–$11.55 span as coverage on that year broadens beyond 13 analysts.
• A weak GAAP print on 2026-08-27 that sells the stock off toward $180-190 while non-GAAP EPS and guidance hold — this is the specific, pre-registered scenario in which we would buy the remainder immediately rather than gradually.

Where MRVL fits in the Synthos Framework Portfolio. The AI-infrastructure / next-exponential sleeve, as a 1.5-2% position, built in three tranches over four to six weeks with the 2026-08-27 print explicitly inside the build window. It is a forward next-exponential rather than a trailing compounder — 41.5% modelled revenue growth with acceleration, which is the profile the selection framework is built to find — but the customer concentration and the 22.6x revenue multiple cap the size. On overlap within the batch: Marvell, AMD and Intel are three different bets on the same buildout and should not all be held at full size; Marvell is the interconnect/custom-silicon expression, AMD the merchant-accelerator expression, Intel the manufacturing-turnaround expression. Marvell is also the most direct beneficiary among them of the optical transition, and the least dependent on winning a head-to-head accelerator contest. Logged as a tracked Synthos call (Buy — Tactical) as of 2026-08-04 at $218.59.

Single biggest risk: customer concentration, and the fact that it is worsening. Three customers represented 75% of gross accounts receivable at 2026-05-02, against five customers at 72% a year earlier. The company's own risk factors state that large customers "may begin developing and making their own semiconductor solutions which could result in a loss of business for Marvell" — the threat is not a competitor, it is the customer. The 10-K and 10-Q both present a table of revenue-level 10%-plus customers whose values did not survive text extraction, so we cannot size the revenue concentration precisely, and we decline to infer it from receivables. At 22.6x revenue, a market that learns one of the three is leaving does not re-rate gently.

Most fragile assumption in the price: that 41.5% annual revenue compounding through January 2029 is achievable without the customer base broadening. The consensus path requires revenue to nearly triple in three years from a customer set of roughly three. Either those three triple their purchases from Marvell — which requires their own capital expenditure to keep compounding and their in-house alternatives to keep failing — or new customers arrive. The filings give evidence for the first and none for the second. If that assumption breaks, it breaks the earnings and the multiple simultaneously, which is what produces a −28.6% bear case rather than a −10% one.

Provenance & disclosures

• Traceability: 38 raw knowledge-base hits on MRVL and Marvell; 33 used; 5 discarded (1 text-only match where the entity is a different company; 2 carrying our own quarantine_misattributed thinker label, excluded on principle; 2 demoted as generic multi-name sector lists). Breadth 5 distinct lanes, net conviction positive-high but narrow. 17 of 33 usable claims come from a single thinker (skill 2.0); 6 are the company's own earnings release (MRVL_mgmt, skill 0.5); 5 more are Marvell's chief executive speaking as a guest and filed in our base under the host's channel rather than the speaker — corrected here and half-weighted as company voice. All quotes verbatim from stored claim text. One claim's figure ("down ~62% in the correction") could not be reconciled with a computed maximum drawdown of −30.9% and was not used. One registered test in the dominant lane — "Marvell outperforms Micron going forward" (2026-05-09) — is currently failing and is reported as such.
• Data as-of: fundamentals — income statement 2026-05-02 (FQ1 FY2027, 10-Q filed 2026-05-28); balance sheet and cash flow 2026-01-31 (FY2026 10-K filed 2026-03-11), with cash updated to $3.8B at 2026-05-02 from the 10-Q; estimates 2026-08-04; prices 2026-08-04, quote timestamp 1785873601 = 2026-08-04T20:00:01Z (16:00:01 ET close), price $218.59, 50-DMA $240.15, 200-DMA $136.25, RSI 54.3; knowledge-base claims 2026-08-04. No figure comes from memory, recall or external retrieval.
• Filing-versus-vendor contradictions resolved in favour of the filing (Section 8): (i) seg_prod reports a single fiscal-2026 end market of $6.100B against $8.195B of revenue — the 10-K and 10-Q report two end markets, and the derived communications-and-other figure of $2.095B is corroborated by the 10-K's stated growth rates to within 0.7 percentage points; (ii) totalDebt of $4.790B includes $319.7M of lease obligations that the 10-K excludes from its own "$4.5 billion… consisting of senior notes outstanding" — we use $4.471B, with the $499.8M current portion matching the filing exactly.
• Capitalisation correction: the vendor's market capitalisation of $191.434B is common-only and excludes 2.0 million Series A convertible preferred shares issued to a strategic investor on 2026-03-31 for $2.0B, convertible at ~$91.84 into ~21.8 million common shares — 138% in the money, worth $4.765B. Fully converted market capitalisation ≈$196.20B; fully converted enterprise value ≈$196.87B, both used throughout. priceToBookRatioTTM, evToSalesTTM and enterpriseValueMultipleTTM as supplied are computed on the understated count.
• Data rejections: trailing GAAP price-to-earnings (74.4x / 75.2x) rejected as a lens — trailing GAAP net income of $2.526B contains a $1.8B pre-tax divestiture gain and a $331.8M contingent-consideration charge; our derived underlying figure is ~$1.25B, ~153x. All est.ebitAvg and est.ebitdaAvg rows rejected as internally impossible (FY2027: EBIT $324.8M against net income $3.532B), and ratios_ttm.ebitMarginTTM of 38.28% rejected against operatingProfitMarginTTM of 16.20%; no EV/EBITDA multiple appears anywhere in this dive, per the data contract. The FY2031 estimate row (revenue −22.8% from FY2030 on more analysts) flagged and not used. The 52-week high is reported as $329.88 in quote and $316.43 in tech; we use $316.43, which reconciles to the computed −30.92% drawdown.
• Basis note: this dive values Marvell on forward non-GAAP EPS, the basis on which the est block and company guidance are stated, while recording that stock-based compensation of $590.8M (7.5% of fiscal-2026 revenue, 42% of that year's free cash flow) is excluded from that basis and is a real cost paid in dilution — evidenced by the diluted share count rising from 856.2M to 893.3M in a single quarter despite $2.040B of buybacks.
• Revenue-quality note: reported revenue is stated net of a customer warrant (up to 4.2M shares at $87.77, grant-date fair value $227.6M, 0.9M vested at 2026-05-02) "recognized as a reduction to revenue as qualifying revenues are recognized." The period charge is not quantified in the filing and is therefore stated rather than adjusted for.
• Cash-flow note: fiscal-2026 free cash flow of $1.396B was less than the $2.245B returned to shareholders; the difference was funded by the $2.5B of proceeds from the Infineon divestiture. The record buyback is not a run-rate.
• Disclosure limitation: both the 10-K and 10-Q introduce a table of customers representing 10% or more of net revenue, but the table values did not survive text extraction. We therefore report the accounts-receivable concentration (three customers, 75%, at 2026-05-02) and decline to infer the revenue concentration from it.
• Estimate coverage: 29 analysts on FY2027 and FY2028 revenue, 24-25 on the corresponding EPS — good. 13 analysts on FY2029 EPS with a $5.28–$11.55 range, and 6 on FY2030. The FY2028 row is used as the primary valuation anchor; no conclusion rests on FY2030 or FY2031.
• Peer note: the vendor peer set (Fortinet, Corning, Infosys, Monolithic Power, Strategy Inc, Cloudflare, NXP, Roblox, Synopsys, TE Connectivity) is not a usable comparable set for a data-centre interconnect supplier, and no peer multiples are supplied. No peer-multiple comparison is drawn.
• Non-equity tripwire: checked and passed. MRVL is common equity — price $218.59, beta 2.246, a variable $0.24 dividend yielding 0.11%, 33.1M shares traded, and a 52-week range of $62.31–$316.43. It is not a preferred, a baby bond or any par-like instrument. The company does, however, have $2.0B of convertible preferred outstanding, treated in Section 4.
• Fair-value caveat: the $156 / $250 / $345 anchors are 25x / 40x / 55x the January-2028 consensus non-GAAP EPS of $6.245 (25 analysts), cross-checked against January-2027 EPS of $4.051 and enterprise value to January-2029 revenue. Stated arithmetic, not a discounted cash flow. The base case assumes multiple compression from 54.0x to 40.0x and derives its entire return from earnings growth; it does not require re-rating. The bull case requires the multiple to hold near current levels on a much larger earnings base and is flagged as the fragile leg.
• Timing: FQ1 FY2027 was reported 2026-05-27 and the 10-Q filed 2026-05-28, sixty-eight days before this dive. The next print is 2026-08-27, twenty-three days away, with company guidance of $2.7B ±5% revenue, non-GAAP EPS $0.93 ±$0.05 and GAAP EPS of only $0.37 ±$0.05. That timing is an explicit input to the Tactical rating and to the staged construction in Section 9.
• Half-weighted sources: all company guidance, the six MRVL_mgmt earnings-release claims, and the five claims whose speaker is Marvell's chief executive, are treated as company voice and discounted accordingly wherever used.
• Not investment advice. Independent research, educational and informational only, never personalised. No recommendation to buy, sell or hold any security is made to any person.
• Version: 2026-08-04-full.