This is investment research, not personal financial advice.
The fall was about timing, not the quarter
Marvell Technology, Inc. (NASDAQ:MRVL) closed at $216.62 on 28 August, down 10.28% after reporting fiscal second-quarter revenue of $2.739 billion and lifting its outlook for both fiscal 2027 and fiscal 2028. The $24.83 decline erased about $21.77 billion from the equity value, based on CNBC's 876.93 million-share snapshot. The same source put the post-result market value at $189.96 billion (CNBC 2026).
The quarter itself beat the midpoint of Marvell's earlier forecast by $39 million. Revenue rose 37% year on year, GAAP operating income rose 58% to $459.7 million, and data-center revenue rose 46% to $2.172 billion. Management now forecasts about $12 billion of fiscal 2027 revenue, up roughly 45%, followed by about $18 billion in fiscal 2028. The prior fiscal 2028 view was about $16.5 billion. Those are not the usual ingredients of a double-digit decline (Marvell 2026a; Marvell 2026c).
Timing supplied the missing ingredient. Marvell announced an expanded Google relationship ten days before the result, complete with a warrant linked to as much as $120 billion of Google revenue. The market had already carried that deal into the share price. On the earnings call, management indicated that some Google revenue sat inside the existing fiscal 2028 custom-silicon assumptions, with a much larger contribution weighted to fiscal 2029. An independent same-day report tied the decline to that timing and noted a forward earnings multiple around 58 times, versus roughly 32 times for Broadcom (Marvell 2026d; Reuters 2026).
The reaction looks roughly proportionate. Marvell delivered a strong quarter and a stronger two-year outlook. It also confirmed that the largest new customer programme contributes modestly before the period that matters most to a $190 billion valuation. A 10.3% reset removed one layer of anticipation. It did not make the remaining price undemanding.
That distinction sets the research question. The issue is no longer whether Marvell has exposure to artificial-intelligence infrastructure. It has plenty. The issue is whether custom chips, optical links and switching can turn a $12 billion fiscal 2027 business into enough owner cash by fiscal 2031 to support the value that remains after Friday's decline.
The warrant puts a number on Google's options
The Google agreement is unusually measurable. Marvell issued a warrant over 58.97 million common shares at an exercise price of $206.58. About 1.36 million shares vested at once. The remaining 57.61 million are performance shares divided into 240 tranches. Each tranche vests after $500 million of cumulative Google revenue is paid or, for a limited set of programmes, accepted for payment. Full performance vesting therefore corresponds to $120 billion of cumulative revenue (Marvell 2026d).
The mechanics matter in three ways. First, $120 billion is a ceiling for the vesting schedule, not a signed near-term sales backlog. The arrangement has no minimum purchase commitment. Google can terminate the commercial agreement on 30 days' notice for convenience, although vested warrant shares remain vested. Second, a portion of the revenue may involve services or other amounts with economics that differ from ordinary chip shipments. Third, the tranches run to 2040. A large headline spread over fourteen years says much less about fiscal 2028 cash than the headline alone suggests.
At Friday's close, all 58.97 million shares would equal 6.7% of CNBC's current share count. Exercise of the full warrant would send about $12.18 billion to Marvell, before expenses, and create the corresponding dilution. The warrant was only about $592 million in the money at $216.62. Most of its value is therefore time and commercial optionality rather than current intrinsic value.
The agreement still says something powerful about Marvell's position. Google is willing to tie Marvell's ownership economics to custom XPU and XPU-attach revenue. The warrant also covers future commercial agreements between the parties. That makes the relationship broader than a single accelerator design. Yet the absence of a minimum commitment leaves programme timing, unit volume and chip economics with Google.
This arrangement is a moat test as much as a revenue announcement. If Marvell's intellectual property shortens design time and connects Google's TPU ecosystem efficiently, the tranches can reinforce the relationship. If Google brings more work inside or changes suppliers, the warrant simply vests more slowly. The contract advertises the opportunity while preserving the customer's bargaining power.
Data center now supplies four dollars in five
Marvell designs data-infrastructure semiconductors and relies on outside manufacturers for fabrication, assembly and testing. Its products move, store, process and secure data. The important categories now include custom accelerators, custom networking processors, optical digital signal processors, Ethernet switches, retimers, storage controllers and security devices. Revenue is recognised when products or services transfer to customers, while the economic work begins years earlier through design engagement and qualification.
The company simplified its end-market disclosure this year. Data center is one group. Enterprise networking, carrier infrastructure, consumer and the remaining industrial activities now sit in communications and other. Data center supplied $2.172 billion, or 79% of Q2 revenue. Communications and other supplied $567.8 million. The former rose 46%; the latter rose 10% and fell 3% sequentially (Marvell 2026c).
That mix explains both the operating momentum and the valuation risk. Connectivity is broadening from 800-gigabit optical products into 1.6-terabit links. Scale-out switching is forecast to more than double during fiscal 2027. Custom revenue is forecast to accelerate in the second half and more than double in fiscal 2028. Q3 group revenue is forecast at $3.15 billion, plus or minus 5%, with data-center revenue rising more than 20% sequentially (Marvell 2026a; Marvell 2026c).
The compounding engine has two loops. Design wins can produce several years of unit revenue because changing a custom chip or optical component late in a platform cycle is expensive and technically risky. Revenue then funds more research, adding reusable SerDes, packaging, memory and photonics blocks to the next design. Marvell spent $1.393 billion on research in the first half, equal to 27% of revenue. That scale allows it to spread engineering across several hyperscalers.
But custom silicon also concentrates the economics. A few customers can shift launch dates, alter volumes or integrate more design work themselves. Marvell's own filings identify dependence on major customers, purchase-order changes and customer vertical integration as material risks. Broadcom provides an imposing peer benchmark: its custom accelerators and networking portfolio operate at a much larger group scale, with deep software cash flow alongside semiconductors (Marvell 2026b; Broadcom 2026).
Five years of capital left returns thin
Marvell's five-year history is not a smooth organic series. The Inphi and Innovium acquisitions expanded data-center connectivity. The automotive Ethernet disposal produced a $1.8 billion pretax gain in fiscal 2026. Celestial AI and XConn entered the balance sheet in fiscal 2027. Amortisation, stock compensation and restructuring then create a wide gap between GAAP and management's adjusted presentation.
The table uses the filed annual numbers and the filed six-month Q2 FY2027 period. ROIC and interest cover are author calculations. ROIC uses GAAP operating income after a 15% normalised tax charge, divided by average invested capital. Invested capital is equity plus debt minus cash. Loss years receive no assumed tax benefit. The interim ratio doubles six-month NOPAT and compares it with average January-to-August capital, so it is a run-rate proxy rather than a filed annual measure (Marvell 2022; Marvell 2023; Marvell 2024; Marvell 2025; Marvell 2026b; Marvell 2026e).
| USD millions unless stated | FY2022 | FY2023 | FY2024 | FY2025 | FY2026 | Q2 FY2027, six months |
|---|---|---|---|---|---|---|
| Revenue | 4,462.0 | 5,920.2 | 5,508.0 | 5,767.3 | 8,195.3 | 5,157.1 |
| GAAP operating income | (196.5) | 240.5 | (820.2) | (575.6) | 639.0 | 799.1 |
| GAAP net income | (421.0) | (163.5) | (933.4) | (885.0) | 1,362.7 | 342.5 |
| Operating cash flow | 819.4 | 1,288.8 | 1,370.5 | 1,681.2 | 1,750.5 | 1,244.3 |
| Capital expenditure | 169.3 | 206.2 | 336.3 | 284.6 | 354.1 | 282.4 |
| Research and development | 1,508.5 | 1,814.4 | 1,956.6 | 1,948.0 | 2,178.7 | 1,393.4 |
| Author-computed ROIC | (2.1%) | 2.6% | (9.1%) | (6.0%) | 5.8% | 4.6% annualised |
| Author-computed interest cover | (1.7x) | 1.5x | (3.9x) | (3.1x) | 3.4x | 7.3x |
Revenue almost doubled between FY2022 and FY2026, but the capital record is less flattering. Reported operating profit was negative in three of the five years. The FY2026 return to positive operating income still produced only a 5.8% computed return on invested capital. The measure includes goodwill and acquired intangibles because those assets consumed shareholder capital. Removing them would make the operating franchise look better while erasing the price paid to build it.
The Q2 run rate is improving. First-half operating income reached $799.1 million, above the $639.0 million recorded in all of FY2026. Interest cover rose to 7.3 times. Yet invested capital jumped after the Celestial AI and XConn transactions. Goodwill reached $13.874 billion at August, 50% of total assets. The acquired technology may prove valuable; the filed return has not caught up with the purchase price.
The incremental-return signal is therefore mixed. Revenue added $2.69 billion from FY2025 to FY2026 and operating income improved by $1.21 billion. That is strong operating leverage. End invested capital also rose as acquisition and preferred-stock funding entered the picture. The business is scaling. The acquired capital base still needs a longer cash record before the return can be called durable.
Cash flow improves once stock pay stays in the bridge
Operating cash flow gives Marvell's cleanest path through the acquisition accounting, but it is not owner cash by itself. Capital expenditure is a genuine reinvestment cost. Stock compensation is an economic cost that either dilutes existing ownership or requires cash repurchases to offset it. The owner-cash bridge therefore starts with operating cash flow, subtracts capital expenditure and then subtracts stock compensation. The final line is conservative because diluted share counts also capture some of the same economics, but it prevents adjusted profit from treating employee equity as free.
| USD millions | FY2022 | FY2023 | FY2024 | FY2025 | FY2026 | H1 FY2027 |
|---|---|---|---|---|---|---|
| Operating cash flow | 819.4 | 1,288.8 | 1,370.5 | 1,681.2 | 1,750.5 | 1,244.3 |
| Less: capital expenditure | (169.3) | (206.2) | (336.3) | (284.6) | (354.1) | (282.4) |
| Raw free cash, author-computed | 650.1 | 1,082.6 | 1,034.2 | 1,396.6 | 1,396.4 | 961.9 |
| Less: stock compensation | (460.7) | (552.4) | (609.8) | (597.4) | (590.8) | (533.8) |
| Owner cash after stock pay, author-computed | 189.4 | 530.2 | 424.4 | 799.2 | 805.6 | 428.1 |
First-half raw free cash reached $961.9 million. After $533.8 million of stock compensation, owner cash was $428.1 million. The Q2 presentation makes the pressure visible: stock compensation was $326.2 million in the quarter, up from $153.6 million a year earlier, partly because Celestial AI and XConn entered the group. Management's non-GAAP operating income of $1.003 billion excluded stock compensation, intangible amortisation and other items; GAAP operating income was $459.7 million (Marvell 2026b; Marvell 2026c).
The balance sheet can carry the current programme. Cash was $3.933 billion and debt $4.963 billion in August, leaving author-computed net debt of about $1.03 billion. First-half operating cash flow exceeded first-half interest expense by more than eleven times. Liquidity is not the immediate problem.
Capital allocation is more complicated. Marvell completed Celestial AI in February for cash, stock and contingent consideration, then acquired XConn. It also issued $2.0 billion of convertible preferred stock to Nvidia and kept repurchasing common shares. Q2 repurchases were $200 million and dividends were $54 million. The quarter's $254 million return did not offset the $326 million stock-compensation expense on an economic-cost basis (Marvell 2026b; Marvell 2026c).
This is the anti-thesis to management's margin story. Non-GAAP operating margin reached 36.6%, up 180 basis points. GAAP operating margin reached 16.8%, up 230 basis points. Both improved. The gap between them was still 19.8 percentage points. A valuation built on adjusted earnings assumes a meaningful portion of that gap either fades, converts into growth worth more than its dilution, or is fully reflected in the future share count.
The moat is design intimacy, with concentration attached
Marvell's strongest defence is not ownership of fabrication. It is integration into a customer's architecture. A custom accelerator or attach chip combines customer specifications with Marvell's interface, memory, packaging and networking intellectual property. Qualification can take years. Once the device enters production, replacing it may disturb software, board design, thermal planning and supply arrangements.
Connectivity broadens the relationship. Marvell can participate in the accelerator, the link between processors, the link across a rack and the optical connection across data centres. Celestial AI adds a photonic fabric aimed at bandwidth and power limits. XConn adds CXL and PCIe switching. If those parts work together, a design relationship can lead to more content around each accelerator.
The financial evidence supports a widening position in data-center connectivity. Data-center revenue rose from $1.491 billion to $2.172 billion in one year. GAAP gross margin improved from 50.4% to 53.1%. Operating leverage appeared in both GAAP and adjusted results. The Google warrant adds direct evidence that a major customer expects a long commercial relationship (Marvell 2026c; Marvell 2026d).
Counter-evidence is equally concrete. One customer now has an instrument that can create 58.97 million shares. The vesting design rewards revenue, not Marvell's margin or return on capital. Google can terminate without a long notice period and has no minimum purchase commitment. A hyperscaler can also develop more of its own silicon, use multiple design partners or change the timing of a platform.
Acquisitions create another caveat. The company has assembled Inphi, Innovium, Celestial AI and XConn around its original portfolio. That has produced broader capabilities, but $13.874 billion of goodwill and negative or low filed ROIC show the cost. A widening product moat does not automatically mean a high capital return when much of the product set was acquired.
Management's next test is execution rather than another announcement. The October investor day can disclose how much of the $18 billion fiscal 2028 outlook comes from existing connectivity, existing custom programmes, Celestial AI and Google. Segment economics are more valuable than another large cumulative-revenue ceiling.
A $190 billion value requires FY2031 cash near the top of the grid
A conventional near-term earnings multiple is a poor fit for Marvell today. GAAP profit contains acquisition amortisation and contingent-consideration marks; adjusted profit excludes recurring stock compensation. An equity cash model, with stock compensation kept in the owner-cash bridge and dilution carried in the share count, better matches the economics.
The starting equity value is $189.96 billion. Net debt adds about $1.03 billion, although the preferred stock and contingent acquisition liabilities require separate attention. Fiscal 2028 revenue is forecast at about $18 billion. The market therefore values the common equity at 10.6 times that revenue two years before it arrives. Friday's Reuters report also placed the shares near 58 times forward earnings, well above Broadcom's comparison multiple (Reuters 2026).
The 10-year US Treasury yield was 4.67% on 27 August. The 10% discount rate used below therefore contains about 5.3 percentage points above that risk-free benchmark. A lower discount rate increases every range, but it would also give a long-duration fiscal 2031 cash assumption more influence over today's value (Federal Reserve 2026).
The reverse model is demanding. Assume owner cash starts around $2.0 billion in fiscal 2027, grows in a straight line through fiscal 2031, is discounted at 10%, and receives a 25-times terminal multiple. Reproducing the current $189.96 billion equity value requires about $10.8 billion of fiscal 2031 owner cash. At $30 billion of revenue, that is a 36% owner-cash margin after capital expenditure and stock compensation. A 30-times terminal multiple lowers the required cash to about $9.1 billion, or 30% of the same revenue base.
The sensitivity shows how heavily the present value rests on the endpoint. Values below use 950 million diluted shares, a 10% discount rate, a $2.0 billion fiscal 2027 owner-cash starting point and a straight path to the stated fiscal 2031 cash. They are author calculations, not company forecasts.
| FY2031 owner cash | 20x terminal | 25x terminal | 30x terminal |
|---|---|---|---|
| $6B | $94 | $113 | $133 |
| $8B | $123 | $150 | $176 |
| $10B | $153 | $186 | $219 |
| $12B | $183 | $222 | $261 |
Two variables dominate: owner cash and the terminal multiple. A two-point shift in revenue growth matters mainly through the cash endpoint. A five-turn shift in the multiple can move value by tens of dollars per share. This is why the timing of Google revenue caused a large reaction even though quarterly revenue beat expectations. Later cash is discounted more heavily and leaves less evidence before the market must choose a terminal assumption.
Four paths from the post-result close
The severe-downside range of $55 to $80 assumes fiscal 2031 owner cash of $5 billion to $6 billion, nearly one billion diluted shares and an 18-to-20-times terminal multiple. This path includes a custom-programme delay, gross margin pressure and continuing stock compensation. The balance sheet remains solvent; the valuation loses most of its AI premium.
The bear range of $90 to $125 assumes $7 billion to $8 billion of fiscal 2031 owner cash, 955 million to 975 million diluted shares and a 20-to-23-times terminal multiple. Marvell reaches much of its fiscal 2028 revenue plan, but customer concentration and acquisition costs prevent adjusted margins from converting fully into owner cash.
The base range of $145 to $190 assumes $9 billion to $10 billion of fiscal 2031 owner cash, 940 million to 960 million diluted shares and a 25-to-28-times terminal multiple. Custom silicon more than doubles in fiscal 2028, connectivity keeps growing, Google becomes larger in fiscal 2029, and GAAP margin closes part of the gap with the adjusted presentation. This range remains below Friday's close at its upper end.
The bull range of $210 to $270 assumes $11 billion to $12 billion of fiscal 2031 owner cash, 920 million to 950 million diluted shares and a 28-to-32-times terminal multiple. Marvell secures several hyperscaler programmes, Celestial AI adds profitable photonics revenue, gross margin stays above 52% GAAP and dilution remains controlled. Friday's $216.62 close sits near the bottom of this range.
The scenario placement is the central valuation finding. The post-fall price does not straddle the base range. It sits where fiscal 2029 Google scale, strong connectivity economics and a high terminal multiple all need to coexist. The anti-thesis is that those conditions can coexist: Q2 data-center growth was 46%, custom revenue is due to more than double in fiscal 2028, and the Google contract may extend well beyond currently named products. The present valuation already asks for much of that evidence before it appears in cash.
The next two disclosures must turn bookings into cash
The first catalyst is Marvell's investor day on 6 October. It can clarify the bridge from roughly $12 billion of fiscal 2027 revenue to roughly $18 billion in fiscal 2028, the role of Google inside that bridge, and the economics of Celestial AI. A useful disclosure would separate custom accelerator, custom attach and connectivity contributions rather than combine them under one AI label.
Q3 results are the second catalyst. The current midpoint is $3.15 billion of revenue, 53.4% GAAP gross margin and $1.10 of adjusted diluted earnings per share. Data-center revenue is forecast to rise more than 20% sequentially. A miss on data-center growth or a gross margin below 52% would challenge the path before Google becomes large. A result near the midpoint with stronger GAAP conversion would support it (Marvell 2026a; Marvell 2026c).
The FY2027 10-K will settle the first owner-cash test. The threshold here is $2.5 billion of annualised owner cash after capex and stock compensation. The filed share count also matters. A move above 950 million diluted shares before material Google warrant vesting would indicate that employee and acquisition dilution is arriving ahead of the customer-linked dilution.
Through fiscal 2028 and fiscal 2029, the crux narrows to three observations: custom revenue more than doubles before Google becomes much larger; GAAP gross margin stays above 52%; and owner cash moves toward the $9 billion to $12 billion fiscal 2031 range embedded in the upper scenarios. Google warrant vesting will provide a rare outside check on cumulative customer revenue, even if the filing does not disclose programme margin.
Source notes: confidence and missing evidence
Verification is full for identity, price, share count, the result, the five-year history and the warrant mechanics. The CNBC regular-close snapshot and SEC filings resolve the market and legal facts. The Finance API sidecar correctly resolved Marvell's identity and surfaced the annual filing history, but its latest daily price stopped on 21 August and its filing list stopped before the 27-28 August result documents. CNBC and the newly filed SEC documents replaced those stale fields; no stale API price enters the article.
One material gap remains. Marvell does not disclose revenue or margin by named hyperscaler. The earnings call comments reported independently establish the Google timing, but they do not give a clean fiscal 2028 dollar contribution. The $120 billion warrant ceiling cannot fill that gap. Scenario values therefore treat Google as one driver inside a wider custom and connectivity path, not as a contracted $120 billion backlog.
The 10.3% fall was justified as a timing reset. Record Q2 revenue and higher forecasts strengthen the business case, while the warrant and valuation show that the market had travelled further than near-term Google cash. At $216.62, the remaining price still observes Marvell through a fiscal 2029 lens. October's bridge and the next two cash-flow filings will show whether the operating evidence can catch it.
References
- Broadcom 2026. Broadcom Inc. Q2 FY2026 Form 10-Q, 9 June 2026.
- Marvell 2022. Marvell Technology, Inc. FY2022 Form 10-K, 10 March 2022.
- Marvell 2023. Marvell Technology, Inc. FY2023 Form 10-K, 9 March 2023.
- Marvell 2024. Marvell Technology, Inc. FY2024 Form 10-K, 13 March 2024.
- Marvell 2025. Marvell Technology, Inc. FY2025 Form 10-K, 12 March 2025.
- Marvell 2026a. Marvell Q2 FY2027 financial-results release, 27 August 2026.
- Marvell 2026b. Marvell Q2 FY2027 Form 10-Q, 28 August 2026.
- Marvell 2026c. Marvell Q2 FY2027 financial and business results presentation, 27 August 2026.
- Marvell 2026d. Marvell Form 8-K covering the Google warrant, 18 August 2026.
- Marvell 2026e. Marvell FY2026 Form 10-K, 11 March 2026.
- CNBC 2026. CNBC regular-close quote snapshot for MRVL, 28 August 2026.
- Reuters 2026. Reuters report on Marvell, Google timing and the market reaction, 28 August 2026.
- SEC 2026. SEC company filings record for Marvell Technology, Inc.
- Federal Reserve 2026. Federal Reserve Bank of St. Louis 10-year US Treasury rate, 27 August 2026.