Every Line of the Framework Moved Up and the Stock Had Already Bought It: Downgrading Marvell to Hold

Coverage Update Since the Q4 Print

We upgraded Marvell to Outperform at the December Q3 print at roughly $98 and maintained that rating at the March Q4 close, also at roughly $98, with a fair value range of $95 to $130 and a $130 twelve-month target. The thesis was that the multi-year framework (FY27 at $11B, FY28 at $15B, EPS above $5) was credible, under-modeled, and would re-rate the multiple toward the 25x to 30x that comparable AI-infrastructure silicon commands.

Both halves happened, and the second half happened much faster than we expected. Operationally, every pillar we listed as a watch item advanced: the FY27 outlook went up again, FY28 went up again, Celestial AI landed a named Tier-1 hyperscaler for scale-up, XConn produced an expanded PCIe and CXL switching position, and the AEC and retimer ramp added a third Tier-1 hyperscaler design win. Financially, the re-rate did not stop at 25x to 30x. The stock closed at $198.70 the day of the print, having more than doubled from the March level in under three months.

That leaves the rating question in a different place than it was in December. The upgrade was explicitly a valuation call on a business whose visibility the market had not yet paid for. That argument no longer exists at these levels, and the honest response is to say so rather than to re-anchor the target to the tape. This report grades the same thesis pillars we set in December and March, and changes the rating on price, not on execution.

Results vs. Consensus

Q1 FY2027 Scorecard

Year-over-Year Comparison

Sequential Comparison

Revenue Assessment

Revenue of $2,417.8M grew 27.6% year over year and 9.0% sequentially, both records. The mix was unchanged at 76% data center and 24% communications and other, but the composition inside data center shifted toward networking. Management attributed sequential and year-over-year growth to optical interconnect, custom silicon, and switching together, with interconnect described as the largest portion of the data center business and the one carrying the raise.

The important number is not the quarter, it is the shape of the year that management now describes. The prior framework, set in March, was high-single-digit sequential growth through FY27 exiting at roughly $3B in Q4. That framework is now obsolete on both counts.

"We are now guiding Q2 revenue to grow double digits sequentially, and we expect Q3 and Q4 revenue to also grow by at least 10% sequentially. As a result, we now expect $3 billion in quarterly revenue in Q3, one full quarter ahead of our prior outlook."
— Matt Murphy, Chairman and CEO

Building the year off the disclosed pieces gives $2,417.8M actual, $2,700M guided, roughly $3,000M in Q3, and roughly $3,330M in Q4 at the stated approximately 50% year-over-year exit rate, which sums to about $11.45B. That reconciles to management's "nearly $11.5 billion" and roughly 40% growth on the FY26 base of $8,195M.

Margin Assessment

Non-GAAP gross margin of 58.9% was 90 basis points below the year-ago 59.8% and 10 basis points below the prior quarter. This is the custom mix effect operating exactly as expected, and the Q2 guide of 58.25% to 59.25% points to more of it. The company is trading gross margin points for revenue dollars in custom silicon, which is the correct trade at this stage of the cycle, but it means the operating-leverage story has to come from the expense line rather than from the gross line.

It did not, this quarter. Non-GAAP operating expenses of $576.9M grew 11.6% sequentially against 9.0% revenue growth, so operating expenses as a share of revenue rose 60 basis points to 23.9% and non-GAAP operating margin fell 70 basis points to 35.0%. Two things drive that: a full quarter of Celestial AI and XConn cost structure with essentially no offsetting revenue (the 10-Q states XConn revenue and earnings since acquisition were not material), and deliberate research spending into scale-up. Year over year the leverage is intact, with operating margin up 80 basis points on 27.6% revenue growth.

The forward framing is where management put the leverage. FY27 non-GAAP operating expenses are set at approximately $2.45B, which after the $576.9M actual and the roughly $600M Q2 guide implies about $1.27B across the back half. For FY28 the CFO gave the clearest operating-margin commitment in the coverage window.

"For fiscal 2028, we expect non-GAAP operating expense to grow year-over-year approximately in the mid- to high teens on a percentage basis. This is significantly below the 45% revenue growth outlook Matt provided in his remarks for that year. As a result, we expect to achieve the upper end of our target operating margin model of 38% to 40% as we progress through fiscal 2028."
— Willem Meintjes, CFO

Assessment: Take the commitment at face value and it implies FY28 gross margin around 56% to 57%, because 45% revenue growth against mid-to-high-teens expense growth only lands at 39% operating margin if the gross line gives back roughly 200 to 300 basis points. Management has not said that, and was not asked. The 38% to 40% target is the constraint that matters for the earnings model, and it caps how much of the revenue raise reaches the bottom line.

EPS Assessment

Non-GAAP EPS of $0.80 grew 29.0% year over year on 27.6% revenue growth, which is barely any operating leverage at all at the per-share level, and was flat sequentially on 9.0% revenue growth. The bridge is arithmetic. Non-GAAP net income rose 4.8% sequentially to $718.0M while diluted shares rose 4.3% to 893.3M, and the non-GAAP tax rate moved from 10.0% to 11.0%. Guidance takes the share count to 915M in Q2, which is 6.9% above the Q4 FY26 level in two quarters.

"The increase from the prior quarter reflects the full impact of shares issued for the Celestial AI and XConn acquisitions as well as the shares issued as part of the NVIDIA investment."
— Willem Meintjes, CFO

GAAP EPS of $0.04, against $0.46 in the prior quarter and a $0.31 guidance midpoint, is a different and largely mechanical story. Interest and other loss, net was $256.1M versus $22.8M in Q4, driven by a $331.8M increase in the fair value of the Celestial contingent-consideration liability, offset by an $81.1M unrealized gain on a $300M-notional cash-settled forward stock purchase contract entered into in April to hedge that exposure. The $250.7M net charge is added back in the non-GAAP reconciliation. Because the contingent-consideration adjustments are not tax deductible, income tax expense of $48.8M landed on pre-tax income of $83.3M.

"For the first quarter, GAAP earnings per diluted share was $0.04 lower than our guidance, reflecting the impact of purchase accounting for the Celestial AI and XConn acquisitions and the related earn-out obligation. We expect this to normalize in the second quarter as reflected in our strong GAAP net income guidance."
— Willem Meintjes, CFO

Assessment: The GAAP miss is not a quality-of-earnings red flag in the usual sense, and the Q2 GAAP guide of $0.37 supports the normalization claim. The share count is the real issue. Marvell has bought a photonics platform, a switching team, and a partnership with the dominant AI compute vendor, and it has paid for all three primarily in equity. That is defensible capital allocation. It also means investors should model per-share growth well below revenue growth for the next several quarters, and it is why the flat sequential EPS deserves more attention than it received.

Segment Performance

Data Center

Data center revenue of $1,832.7M grew 27.2% year over year and 11.0% sequentially, with growth across optical interconnect, custom silicon, and switching. The segment held 76% of the mix, up from 74% in Q4. The Q2 guide steps this up sharply: mid- to high-teens sequential growth into the mid-40% range year over year, which implies roughly $2.1B to $2.2B and would put data center above 78% of revenue.

Management's framing of the multi-year shape is the disclosure to hold onto, because it describes acceleration rather than the deceleration that normally follows a base of this size.

"Our data center revenue grew 46% year-over-year in fiscal 2026, and we are now projecting growth to accelerate to approximately 50% in fiscal 2027 and accelerate again to 55% in fiscal 2028."
— Matt Murphy, Chairman and CEO

Applied to the FY26 data center base of more than $6B, roughly 50% growth points to approximately $9.1B in FY27 and roughly 55% on that points to approximately $14.2B in FY28. Adding communications at roughly $2.3B rising at low-single digits reconciles cleanly to the $16.5B FY28 total. The framework is internally consistent, which is more than can be said for many multi-year targets.

Assessment: Three consecutive years of data center growth at or above 46%, with the growth rate rising each year, is the single most demanding claim in this report. It is supported by named product ramps rather than by market-share assertions, which is why we treat it as credible. It is also why any single program slip is a thesis event rather than a quarter event.

Interconnect

Interconnect is the largest piece of data center and the source of most of the raise. The FY27 growth expectation moved from more than 50% at the March call to more than 70% now, having started the year at roughly 30%.

"Demand for our interconnect products continues to accelerate. And as a result, we have increased our fiscal 2027 revenue growth expectations for this business to more than 70% year-over-year."
— Matt Murphy, Chairman and CEO

The building blocks disclosed: 800G demand strengthening while 200G-per-lane 1.6T products ramp quickly following their second-half FY26 production launch, with another substantial step up expected in FY28; leadership carried into the next PAM4 generation at 400G per lane, first demonstrated at the Optical Fiber Conference in April 2025; and a broadband analog business in TIAs and drivers that management expects to exceed a $1B annualized run rate within the next few quarters.

Data center interconnect is undergoing an architectural change management calls scale-across, where the back-end AI network extends between physically separate data centers because power and space constrain single sites. Marvell pioneered pluggable DCI and ships to all five major U.S. hyperscalers. Aggregate bandwidth needs for scale-across are projected at more than ten times current front-end DCI networks, and the transition skips a generation, moving from 400G and 800G directly to 1.6T. The company's first secure 1.6T ZR and ZR+ modules, on a 2nm coherent DSP, are expected to sample this year.

"Our leadership position here is translating into strong revenue momentum for our DCI module business, giving us line of sight to a $1 billion annualized revenue during fiscal 2028."
— Matt Murphy, Chairman and CEO

Assessment: DCI doubling from roughly $500M in FY26 to a $1B annualized rate in FY28 is a concrete, verifiable milestone attached to a disclosed customer set, which makes it the most auditable line in the framework. The interconnect raise from 30% to 70% inside twelve months is also the clearest evidence that Marvell's content per accelerator is rising faster than accelerator units, which is the strongest version of the bull case.

Scale-Up Optics

Scale-up interconnect is the newest and, in management's framing, the most strategically important emerging opportunity. Marvell claims coverage of all three mainstream modulator technologies (MZM, EAM and MRM) for both near-package and co-packaged optics implementations, backed by more than one million DCI modules shipped on its silicon photonics over the past decade and more than fifteen billion hours of field data across four generations. A third-generation 6.4T light engine is in engagement with multiple Tier-1 customers, and MRM device demonstrations are complete with TSMC collaboration ongoing.

The Celestial AI contribution moved from a promise to a named win.

"The solution has already been selected by a Tier 1 hyperscaler for its next generation of XPU scale-up networks."
— Matt Murphy, Chairman and CEO

The revenue framing moved with it. In March, scale-up optics for FY28 was approximately $150M and was entirely Celestial.

"And as a result, we expect our scale-up optics business to ramp significantly next fiscal year, with revenue forecasted to more than double our prior outlook of approximately $150 million, which was based at that time solely on Celestial AI."
— Matt Murphy, Chairman and CEO

In Q&A, that was sized at approximately $300M for FY28 across genuine near-package and co-packaged optics solutions. Separately, the company added Polariton, a developer of plasmonic-based silicon photonics devices, whose modulators management says have demonstrated bandwidth "exceeding 1 terahertz up to 10x higher than current silicon photonics and thin-film lithium niobate-based solutions", destined for the DCI and coherent-light roadmaps at 3.2T and beyond.

Assessment: Doubling a $150M FY28 line to $300M is a small absolute number against a $16.5B total, and that is the point: scale-up optics is a FY29-and-beyond earnings driver being validated in FY27. The named Tier-1 hyperscaler selection is the single most important de-risking event for the Celestial acquisition since it was announced in December, because it converts a photonics thesis into a design win. It also raises the probability of the earnout milestones, which is part of why the contingent-consideration liability marked up.

Switching

Scale-out switching runs on sustained 12.8T demand and a 51.2T ramp, with engagement from both existing and new customers and a roadmap already executing toward 200T Ethernet.

"Given this momentum, we expect scale-out switch revenue in fiscal 2027 to exceed $600 million, doubling from fiscal 2026, and we currently see the business tracking to more than $1 billion in annualized revenue in fiscal 2028."
— Matt Murphy, Chairman and CEO

Scale-up switching is the greenfield. XConn substantially expanded the team and capability set. Management's read is that PCIe switches now serving current-generation scale-up networks will give way to purpose-built high-radix UALink, ESUN and NVLink solutions, and Marvell claims a position across all three protocols, the third through the NVIDIA partnership. Multiple Tier-1 engagements are in flight, each described as a multibillion-dollar lifetime revenue opportunity.

Assessment: Scale-out switching going from roughly $300M in FY26 to more than $1B annualized in FY28 is a genuine franchise emerging from what was a sub-scale acquisition six years ago. Scale-up switching is more interesting for what management said about where it sits in the model, which is covered in the Q&A section below. Both are credible; only one is in the numbers.

Custom Silicon and XPU Attach

Custom remains on track for more than 20% growth in FY27, led by the flagship XPU program, with CXL and NIC attach programs also ramping. The FY28 statement is the change.

"Looking ahead to fiscal 2028, we now expect custom revenue to more than double year-over-year, which is higher than our prior outlook."
— Matt Murphy, Chairman and CEO

Three drivers, unchanged in composition from March but each larger: continued growth from existing programs including the flagship XPU; more than ten XPU attach programs reaching higher production volumes, with demand exceeding prior forecasts particularly in NIC and CXL memory attach driven by inference key-value caching; and the new Tier-1 XPU program entering volume production.

"This program continues to progress very well through development, and we already have firm requirements in place for all of next fiscal year."
— Matt Murphy, Chairman and CEO

Several new designs were won since the last call, with the usual roughly two-year development cycle before revenue. The long-term target was reaffirmed: "We remain confident in achieving our target model for our custom business to deliver on over $10 billion in revenue in fiscal 2029."

Also inside data center: the AEC golden cable program has design wins with three Tier-1 U.S. hyperscalers plus several other customers, and AEC combined with retimers is expected to more than double year over year in FY27 and keep growing rapidly in FY28. XConn adds PCIe Gen 6 and CXL 3.1 switching, paired with Marvell retimers and memory expanders as end-to-end reference designs.

Assessment: The custom line is where execution risk concentrates and where the FY29 target lives. More than doubling FY28 while holding the FY29 $10B goal means custom has to roughly double again from FY28 to FY29, which requires the new Tier-1 program to scale on schedule. Management describes that program as having firm requirements for all of next fiscal year, which is meaningful but is a one-year commitment against a three-year target.

Communications and Other

Communications and other delivered $585.1M, up 28.7% year over year and 3.1% sequentially. Management called the recovery complete: "The communications and other end market has now largely recovered from inventory corrections at our customers." Going forward, the segment is expected to reflect underlying enterprise networking, carrier and consumer trends rather than a restocking cycle. FY27 growth is held at approximately 10% and FY28 at low-single digits.

Q2 is guided down mid-single digits sequentially and up high single digits year over year, so roughly $556M. That is the first sequential decline in this segment in the coverage window.

Assessment: This is the correct way to normalize a recovery: say it is over and reset the growth rate rather than extrapolate the rebound. The sequential decline was neither explained nor questioned, which is a small gap given the segment is still 24% of revenue. At low-single-digit FY28 growth, communications shrinks to roughly 14% of the FY28 mix, and Marvell becomes, in practice, a data center company with a legacy attachment.

The Multi-Year Framework: What Moved

The single most useful way to read this print is as a delta against the framework set eleven weeks earlier.

Two features of this table matter more than the individual rows. First, the FY28 revenue raise is built on an explicitly conservative cloud CapEx assumption: management is planning for the rate of cloud CapEx growth to moderate into the 30%-plus range and still expects data center to accelerate to roughly 55%. That is a content-per-dollar claim, not a spending claim. Second, management stopped restating the FY28 EPS figure. The ">$5" from March was never withdrawn, but with revenue up 10% and an explicit 38% to 40% operating-margin target now on the record, the arithmetic supports a number closer to $6, and management chose to let the reader do that work.

Guidance and Outlook

Implied Q2 operating margin. At the $2,700M revenue midpoint, a 58.75% non-GAAP gross margin midpoint and approximately $600M of non-GAAP operating expenses give roughly $986M of non-GAAP operating income, or about 36.5%. That is a 150 basis point sequential improvement and is the first quarter in which the FY28 operating-margin path becomes visible.

Implied back-half ramp. To reach nearly $11.5B for the year, the second half must deliver roughly $6.33B, or about 55% of annual revenue, with Q3 near $3.0B and Q4 near $3.33B. Management has committed to at least 10% sequential growth in each of those quarters, which makes the back half a commitment rather than a hope. The corresponding non-GAAP operating expense budget is roughly $1.27B across two quarters against the approximately $2.45B annual figure.

Guidance style. Marvell has now raised the FY27 outlook at four consecutive updates, from roughly $9.5B at a September investor call to $10B in December to $11B in March to nearly $11.5B now. That is a company guiding to what it can see rather than to what it hopes, and the pattern argues the current number is more likely to be exceeded than missed. It is also, at this point, fully anticipated by the market, which is a separate problem covered in the rating.

Key Topics and Management Commentary

Overall Management Tone: Management was the most assured it has been in the coverage window, presenting a raise to both fiscal years and a partnership with the dominant AI compute vendor as routine progress reports rather than headline events. The posture on demand was declarative and on supply was operational, with the only hedging appearing around the forward gross-margin path, which was never framed. Analyst pushback was narrow and directed almost entirely at whether the forward numbers are too conservative, not whether they are achievable, which is itself a signal about how the framework is now received.

1. The NVIDIA Partnership and the $2 Billion Preferred Placement

The most consequential disclosure of the quarter is not in the revenue lines at all. On March 31, 2026, Marvell completed the private placement of 2.0 million shares of Series A Convertible Preferred Stock to NVIDIA for an aggregate purchase price of $2.0 billion. The shares carry a $1,000 stated value and an initial conversion price of approximately $91.84, converting into a maximum of roughly 21.8 million common shares. The preferred participates in dividends and liquidation on an as-converted basis, votes with common except on director elections, and carries no redemption or preemptive rights. For accounting purposes it is treated as a second class of common stock under the two-class method, which is why the earnings release carries a note explaining the EPS computation.

The commercial content has three pillars: extending an existing DSP, TIA and driver supply relationship into silicon-photonics collaboration for scale-up; NVLink Fusion integration allowing Marvell custom chips and networking silicon to interface with NVIDIA infrastructure; and AI-RAN, enhancing OCTEON base-station processors to work directly with NVIDIA GPUs so operators can run 5G and 6G radio workloads alongside AI on one software-defined platform.

"The collaboration connects Marvell's custom silicon and optical networking capabilities directly into the massive NVIDIA ecosystem to help build scalable, highly efficient AI data centers and telecommunications networks."
— Matt Murphy, Chairman and CEO

Management's framing of the equity component was brief and warm: "We deeply appreciate the partnership and the investment from NVIDIA."

Assessment: Strategically this is the strongest possible endorsement of Marvell's SerDes, optics and custom-silicon platform, and NVLink Fusion in particular converts a competitive relationship into an interoperability position that hyperscalers building mixed estates will value. Financially it deserves a colder read. NVIDIA acquired an option on roughly 21.8 million shares at $91.84 at the end of March; those shares closed at $204.83 on the reaction day. Whatever the placement was worth as validation, the capital came in at less than half of where the stock trades eight weeks later, and existing holders funded the difference. The partnership is a genuine asset. It was not free.

2. The Raise Is About Acceleration, Not Recovery

The pattern in the outlook revisions has changed character. The FY27 number went from roughly $9.5B to $10B to $11B to nearly $11.5B across four updates, which could be read as a company catching up to a demand environment it under-forecast. The FY28 revision is different, because it accompanies an explicit statement that data center growth accelerates from roughly 50% to roughly 55%, on top of 46% in FY26, while cloud CapEx growth is assumed to moderate.

That combination only works if Marvell's content per unit of customer spending is rising, which is exactly the claim management makes through the agentic-AI argument: more complex inference routes more queries across more of the cluster, which raises traffic that must be switched at low latency over longer reaches, and it drives more CPUs, which in turn drive more NICs, PCIe switches and retimers.

"As a result, we believe agentic AI can provide another significant tailwind for our interconnect switching and XPU attach franchises."
— Matt Murphy, Chairman and CEO

Assessment: This is the intellectual core of the bull case and it is well constructed. Networking content scaling faster than compute content is a defensible structural claim, and the interconnect growth path from 30% to 70% inside one fiscal year is empirical support for it. The vulnerability is that the same logic runs in reverse if inference workloads consolidate onto fewer, denser domains rather than spreading across clusters and sites.

3. Interconnect Is Carrying the Company

Interconnect is the largest portion of data center, the largest contributor to the FY27 raise, and a major driver of the higher FY28 outlook. The portfolio spans scale-out PAM4 (800G ramping, 1.6T ramping quickly, 400G-per-lane demonstrated), coherent light for 2 to 20 kilometer campus reaches at low power with integrated MACsec now being introduced, pluggable DCI transitioning to 1.6T for scale-across, and broadband analog TIAs and drivers approaching a $1B annualized run rate.

The scale-across transition is worth isolating. Building AI clusters larger than a single site can power means the back-end network extends between data centers, with continuous XPU-to-XPU traffic rather than the front-end traffic that drove the original pluggable DCI market. Management puts aggregate scale-across bandwidth requirements at more than ten times current front-end and DCI networks, and expects those architectures to adopt 1.6T directly while traditional DCI is still moving from 400G to 800G.

Assessment: Marvell shipping pluggable DCI to all five major U.S. hyperscalers is a durable incumbency, and skipping a speed generation is exactly the kind of discontinuity that lets an incumbent extend a lead. The $1B annualized DCI milestone in FY28 against roughly $500M in FY26 is the cleanest single test of whether scale-across is real, and it arrives on a schedule investors can check.

4. Celestial AI Converts From Thesis to Design Win

Six months ago Celestial AI was an announced acquisition with a photonic-fabric technology story and a single implied lead customer. This quarter it is a closed $3.5B acquisition whose photonic fabric, contributing EAM modulators and low-power analog SerDes, has been selected by a Tier-1 hyperscaler for its next generation of XPU scale-up networks, with Marvell's full engineering and operations organization focused on bringing the first-generation chiplet to high-volume manufacturing.

Management's characterization of the combination is that the Celestial team plus Marvell's optics organization can now present customers with end-to-end solutions spanning XPU to switch with any interface and any optical or copper connection between them, and that the forward scale-up optics ramp includes Marvell products alongside Celestial products rather than Celestial alone.

Assessment: The purchase price allocation tells you how much of this is still promise: of $3,537.4M in total consideration, $2,404.4M went to goodwill and $951.0M to intangibles. That is an acquisition bought almost entirely on future cash flows. The named Tier-1 selection materially improves the odds, and the $300M FY28 scale-up optics framing is a first revenue datapoint. It is also why the earnout liability marked up $331.8M in one quarter, and why up to roughly 22.4 million additional shares remain issuable through FY29.

5. Scale-Up Switching Is Deliberately Outside the Numbers

Marvell claims a position across all three emerging scale-up protocols: UALink and ESUN through internally developed switches, and NVLink through the NVIDIA partnership. XConn substantially expanded the team. Multiple Tier-1 engagements are in flight, each characterized as a multibillion-dollar lifetime revenue opportunity, in a market where management's argument is that no incumbent share position exists yet.

In Q&A, management stated plainly that this business is not in the $16.5B FY28 outlook, and has never appeared in an analyst-day framework. It is being presented as free optionality on top of a number that is already 45% above the current year.

Assessment: This is either genuine conservatism or a very well-constructed way to keep an upgrade path in reserve, and the distinction matters less than the fact that it is disclosed. Investors should treat it the way management does: excluded from FY28, possible in FY29, unquantified. The bull case that gets you above our fair value range runs through this line converting.

6. Custom Is the Load-Bearing Wall

Custom grows more than 20% in FY27 and more than doubles in FY28, on three roughly equal drivers, and management reaffirmed the target of more than $10B of custom revenue in FY2029. That last figure has a history: it derives from long-term targets set in April 2024 through calendar 2028, then revised upward at the June 2025 custom silicon event when the addressable market was resized to roughly $55B against a 20% share ambition.

The near-term evidence is stronger than the framework alone: firm requirements are in place for all of next fiscal year on the new Tier-1 XPU program, more than ten XPU attach programs are reaching higher production volumes, and new designs continue to be won on a roughly two-year development cycle.

Assessment: Custom carries the FY28 acceleration and the FY29 target, and the arithmetic is demanding. If custom is a little over $4B in FY28 and more than $10B in FY29, that is $5B to $6B of incremental revenue in a single year from a small number of programs. Management confirmed that reading without qualification. This is the pillar to monitor most closely and the one where a single hyperscaler roadmap change does the most damage.

7. Supply Is Being Bought With Cash Two Years Ahead

The company is locking capacity aggressively, using what management describes as the playbook from the last supply crunch: share a five-year demand outlook with a small set of key suppliers and back it with prepayments.

"We are forecasting approximately $1 billion in prepayments during this fiscal year with the first payments beginning in the second quarter. These prepayments will be applied against future material purchases."
— Willem Meintjes, CFO

The filings corroborate the scale. Unconditional purchase commitments to foundries and test and assembly partners stood at $2,756.8M as of May 2, 2026, of which $2,265.2M falls due in the remainder of FY27 alone, alongside a separate $709.3M of technology services and license fee commitments. After quarter end, Marvell signed agreements securing wafer and substrate capacity through FY2030 and FY2033 respectively, committing to $870.0M of deposits payable quarterly from Q2 FY27 through Q2 FY28.

Assessment: This is the strongest non-verbal confirmation of the demand outlook in the entire release, because prepayments are a cash commitment against a forecast rather than a statement about one. It is also a real free-cash-flow claim. Operating cash flow was a record $638.8M in the quarter and capital expenditures were $155.7M, giving roughly $483M of free cash flow; roughly $1B of prepayments plus $870M of deposits will absorb a meaningful share of FY27 and FY28 generation, which is part of why buybacks have slowed.

8. Operating Leverage Was Promised for Next Year, Not This One

The FY27 non-GAAP operating expense budget of approximately $2.45B against nearly $11.5B of revenue implies roughly 21.4% of revenue, versus 23.9% in Q1, so the leverage is back-half weighted. The FY28 commitment (mid-to-high-teens expense growth against 45% revenue growth, reaching the upper end of a 38% to 40% operating margin model) is the specific, checkable version of the leverage claim, and it is the first time management has committed to a point inside that range.

Assessment: The commitment is credible and it is also the ceiling. A 38% to 40% target model, restated as a target rather than raised, means the incremental margin on the FY28 revenue raise is bounded. Investors modeling operating margin above 40% on the strength of the revenue acceleration are modeling something management has explicitly declined to promise.

9. The Share Count Is Now a Thesis Variable

Diluted shares went from 856.2M in Q4 FY26 to 893.3M in Q1 FY27 to a guided 915M in Q2. The sources are disclosed and additive: 26.8 million common shares issued for Celestial AI and XConn, roughly 21.8 million shares underlying the NVIDIA preferred, and up to roughly 22.4 million more issuable under the Celestial earnout through FY29. Against that, the company repurchased $200.0M of stock in the quarter, down from $340.0M in the year-ago quarter, and paid $53.8M of dividends, while $227.2M went to tax withholding on employee net share settlements.

"In parallel, we plan to continue to repurchase shares to manage dilution."
— Willem Meintjes, CFO

Assessment: That sentence is a change in the capital-return posture, not a continuation of it. FY26 returned $2.245B and was framed as returning capital; FY27 opens at $200.0M of repurchases in the quarter, framed as offsetting issuance. Given the competing claims on cash from $1B of prepayments and $870M of deposits, that is a defensible allocation. It should still be modeled honestly: per-share growth will trail operating growth for at least the next several quarters, and the Q1 result, with 9% sequential revenue growth producing no sequential EPS growth at all, is the demonstration.

10. Photonics Roadmap Buying Continues

Polariton adds plasmonic modulators to a silicon-photonics platform that already spans MZM, EAM and MRM, with micro-LED and micro-pixel approaches also under investment. Management's stated rationale is modulator bandwidth: plasmonics enable substantially higher bandwidth than conventional silicon photonics, which becomes the constraint as optical transmission speeds rise toward 3.2T and beyond.

"Polariton has already demonstrated Plasmonic modulator bandwidth exceeding 1 terahertz up to 10x higher than current silicon photonics and thin-film lithium niobate-based solutions."
— Matt Murphy, Chairman and CEO

Assessment: Marvell is now deliberately funding multiple competing photonic technologies rather than picking one, which management characterized as not wanting to miss out and being prepared to pivot with customers. That is a rational hedge in a market where the winning modulator architecture is genuinely undetermined, and it is a real expense commitment against the operating margin target. The cost of optionality is showing up in the operating expense line before the revenue does.

11. Communications Has Finished Recovering

The segment is up 28.7% year over year and has been reset to approximately 10% growth for FY27 and low-single digits for FY28, with Q2 guided down mid-single digits sequentially. Management explicitly closed out the inventory-correction narrative that has run through this business for two years.

Assessment: Getting this out of the way is helpful, because it means every remaining growth question is a data center question. It also removes a cushion. Communications provided a counter-cyclical recovery tailwind through FY26 that no longer exists, so from here the company's revenue line is close to a pure function of AI infrastructure spending and Marvell's content within it.

Analyst Q&A Highlights

The $10 Billion Fiscal 2029 Custom Target

The opening exchange went straight to the largest number in the framework, testing whether management really intended to reaffirm a target that implies $5B to $6B of incremental custom revenue in a single year. Management confirmed it without hedging and walked through the derivation: long-term targets set in April 2024 through calendar 2028, resized at the June 2025 custom silicon event when the addressable market was put at roughly $55B against a 20% share ambition. The follow-on question, about when investors should expect disclosure on the new large customer, drew a milestone-level answer rather than a timeline.

Q: "Matt, towards the end of your presentation, you mentioned something along the lines of your custom XPU is on target to hit $10 billion in fiscal '29. I just wanted to make sure that we heard that correctly because that would mean that fiscal '28 if your XPU is a little over $4 billion and then it gets to $10 billion. So that's an increase of at least $5 billion to $6 billion year-on-year in sales."
— Vivek Arya, Bank of America Securities

A: "Two years ago, it looked like a very steep hill to climb. We're clearly making progress there. And yes, between our existing programs, our new ramp and just a plethora of XPU attached programs, all of which have sized up significantly since we won them. We definitely see line of sight to hit those targets."
— Matt Murphy, Chairman and CEO

Assessment: Management reaffirmed the most aggressive figure it has ever put on record and grounded it in share-of-TAM arithmetic rather than in a bottoms-up program list. The new Tier-1 ramp was sized at roughly one-third of next year's custom growth, which usefully bounds how much of FY28 depends on it. It leaves FY29 dependent on programs that have not been named.

Whether the Customer Base Is Broadening Beyond the Named Programs

A recurring line of questioning probed whether Marvell is moving into the general-purpose custom compute wallet, which would be additive to everything disclosed, and how many hyperscalers are actually in the pipeline. Management confirmed engagements across all U.S. hyperscalers, some involving both XPU and XPU attach and some attach only, and made the more useful point that the disclosed growth requires nothing new.

Q: "I wanted to ask about the breadth of the customer base. I know we've talked about the existing XPU you have. There's this new customer on the XPU and then you do have some XPU attached with a third customer as well. But there's some speculation that you actually might be moving into the compute TAM, and that's by far the biggest custom compute wallet out there. So -- so is that included in the forecast? Or would that be incremental to that?"
— Timothy Arcuri, UBS

A: "So everything we've laid out, which is the greater than 20% growth this year, more than doubling next year and then still having line of sight to our long-term targets. That's all based on the designs we've already won and locked and even going back really the last summer."
— Matt Murphy, Chairman and CEO

Assessment: The most reassuring answer on the call. Newer wins were characterized as an insurance policy against slippage in existing programs rather than as a requirement, which is the right structure for a multi-year framework. It also implicitly confirms that the framework is concentrated in a small number of already-won sockets, which is the same fact viewed from the risk side.

Why Interconnect Decelerates in the Fiscal 2028 Framework

The sharpest question of the call inverted the usual skepticism: given interconnect growth has been revised from roughly 30% to 50% to more than 70% within a single fiscal year, why does the FY28 framework show it slowing to merely above cloud CapEx growth? Management did not defend the deceleration as a forecast so much as decline to raise it yet, and then enumerated the reasons it might prove low.

Q: "Matt, you talked about the growth rate going to, I think, over 70% this year. If I recall right, it was at the beginning of this year, 30% and then 50% and now 70%. So it's clearly accelerating. I guess the question is, why would you think other than just conservatism that, that would slow to kind of closer to, but still above the hyperscaler CapEx rate next year?"
— Ross Seymore, Deutsche Bank

A: "So I think there's a lot of optionality is what I would say at the moment, Ross. But I think today, net-net, you look overall, 16.5 where we're comfortable overall, but I think there's upward bias for sure."
— Matt Murphy, Chairman and CEO

Assessment: Management effectively pre-announced that the $16.5B is a floor. That is useful information and it is also the reason the print did not move the stock further: if the guide is understood as conservative by construction, the next raise is already partially discounted. The tell to watch is whether the FY28 interconnect growth rate gets revised the way the FY27 rate did, three times in twelve months.

What Scale-Up Optics Actually Ships Today

With copper, near-package optics and co-packaged optics all in play, a question pressed on which specific products are generating the upside rather than the category as a whole. The answer emphasized breadth of technology coverage and, more pointedly, the difference between a demonstrated silicon record and a roadmap presentation.

Q: "Matt, I was hoping to zoom in a little bit more on the interconnect business and specifically scale up. Obviously, a lot of that is in front of you. But there's also a lot of different dynamics there, right, whether it's copper, optical, MPO, CPO, copper. What has been surprising to you the last few months as far as what 2 or 3 products are you seeing the bigger upside from?"
— Tore Svanberg, Stifel

A: "And the fact is we have the proven technology. It's not a PowerPoint. It's not a concept. I mean, we have 15 billion hours as an example, of device data on 4 generations of silicon photonics already."
— Matt Murphy, Chairman and CEO

Assessment: The substantive content was that the Celestial team combined with Marvell's optics organization now sells an end-to-end XPU-to-switch story, and that FY28 scale-up optics revenue includes Marvell products and not only Celestial products. That is a broader base than the acquisition case assumed. The reliability data is a real competitive argument in a market where first-generation co-packaged optics deployments carry field-failure risk.

Scale-Up Switching Sits Outside the $16.5 Billion

Questioning turned to whether the scale-up switch opportunity, spanning UALink, ESUN and NVLink, could ultimately exceed the scale-out switch business that is doubling this year. The answer contained the single most important modelling disclosure of the call.

Q: "I know it's a large market and you have relatively small share in that market. But as we go from scale out to scale up, you mentioned 3 different, I guess, ESUN, NVLINK and also UALink. So it seems to me that, that could be even larger market. And obviously, it's all greenfield market for you. So I'm just curious as to how you're thinking about that opportunity in '28 and beyond."
— Srinivas Pajjuri, RBC Capital Markets

A: "But on the scale up and on the switching side, that is not really in any numbers I'm talking about right now. I mean that's not very little or nothing in the $16.5 billion. And then the year after, we'll probably get some contribution, but that's never showed up in an Analyst Day we've done or any kind of discussion. So that's all in front of us, and that's all upside."
— Matt Murphy, Chairman and CEO

Assessment: An explicit statement that a multibillion-dollar lifetime opportunity carries no FY28 revenue in the framework. This is the cleanest upside optionality in the story and the most likely source of the next framework raise. It should be valued as an option, not as an estimate, and it is the specific development that would move us back toward a positive rating.

Whether the Raise Reflects Demand or Newly Available Supply

One line of questioning tried to separate two very different explanations for the guidance increase: did Marvell secure more capacity, or did it become more confident in forecasts customers had already given it? The answer came from the operations side and reframed the question as one about supplier relationships rather than spot availability, noting that nothing touching AI has been unconstrained since 2020 or 2021.

Q: "Question is about some more color on how you're addressing capacity constraints right now. You spoke about some of the prepayments. And so I mean, first part of the question is, how you're managing to get that additional capacity."
— Christopher Caso, Wolfe Research

A: "And ultimately, the way we've been able to manage this is by building very tight relationships with a small number of key suppliers, giving them a 5-year forecast of what we need and hitting what we need each time along the way and doing what we said we were going to do."
— Christopher Koopmans, President and COO

Assessment: The question was not fully answered. Management described how it gets capacity without saying whether the raise was supply-unlocked or demand-driven, which matters because the two imply different things about how much more upside exists. The corroborating evidence sits in the filings rather than the transcript: unconditional purchase commitments to foundry and assembly partners stand at $2,756.8M, with $2,265.2M due in the remainder of FY27, and $870.0M of new capacity deposits were committed after quarter end. A company does not commit that cash against a forecast it has not been given.

The Composition of the Fiscal 2028 Custom Double

A closing question asked which of the three custom growth drivers dominates, testing whether the doubling depends disproportionately on the unnamed new Tier-1 program. Management held the split unchanged from the prior call and noted that each bucket has been revised upward rather than one carrying the increase.

Q: "I'm wondering if we look at the fiscal '28 outlook for the custom to double and the 3 drivers that you outlined, can you give us a little bit more color as to which of these is the biggest or how to think about the contributions between existing customer expansion, the XPU attach and then the new Tier 1? Just trying to get a sense of relative size of each."
— Simon Leopold, Raymond James

A: "So last quarter, when we talked about it, it was about 1/3, 1/3, 1/3 in those different buckets, the existing programs, XPU attach and then in our new program. And it's about the same."
— Matt Murphy, Chairman and CEO

Assessment: The most useful risk-sizing answer of the call. A balanced split means no single program failure breaks the FY28 custom double, and that all three buckets were revised up together implies a broad demand signal rather than one customer's roadmap. The offsetting observation is that "everything wants to be more" is a description of a supply-constrained allocation decision, not of a demand ceiling, which makes the FY28 custom number a company choice as much as a market outcome.

What They're NOT Saying

1. Customer concentration, which got materially worse. The 10-Q shows one distributor at 45% of net revenue, up from 36% a year ago, alongside a direct customer steady at 16%. Three customers now account for 75% of gross accounts receivable, against five customers at 72% a year earlier. Neither the release nor the call mentioned concentration in any form, and no analyst raised it. Two relationships now carry 61% of revenue in a business whose entire thesis is multi-customer diversification.
2. The forward gross margin path. Management gave FY28 operating expense growth and an operating margin target but said nothing about gross margin, despite custom more than doubling in the same year and custom carrying below-corporate margins. Non-GAAP gross margin has now declined 90 basis points year over year and is guided lower again at the Q2 midpoint. The 38% to 40% operating margin commitment implies FY28 gross margin somewhere near 56% to 57%, which is a meaningful step down that was never stated.
3. Any detail on the new custom sockets. Asked directly to characterize the new design wins by type of accelerator or attach product, management declined: "No additional details at this time." Investors are being asked to credit wins whose category, customer and size are all undisclosed, on the promise of a comprehensive look-back at some future point.
4. Free cash flow, and what the prepayments do to it. The call cited record operating cash flow of $638.8M but gave no free cash flow figure and no FY27 capital expenditure framing, while committing to roughly $1B of prepayments this year plus $870.0M of capacity deposits disclosed only in the filing as a post-quarter event. Unconditional purchase commitments to foundry and assembly partners stand at $2,756.8M, with $2,265.2M of that falling due before the fiscal year ends. The cash cost of securing growth is real and was described only as a positive.
5. Why the buyback slowed by 41%. Repurchases were $200.0M against $340.0M in the year-ago quarter, and FY26 in total returned $2.245B. The reframing to repurchasing "to manage dilution" is a meaningful change in capital-return posture that was stated in one clause and not questioned.
6. The geographic shift. China destination shipments rose to 44% of revenue from 37%, while U.S. destination fell to 7% from 16%, and Asia overall reached roughly 83% of revenue from roughly 74%. The 10-Q offers the standard explanation that destination reflects contract manufacturing rather than end-customer location, but a seven-point year-over-year shift toward China in the current export-control environment merited a sentence and did not get one.
7. The FY28 EPS number. The March framework carried non-GAAP EPS above $5. With FY28 revenue up roughly 10% and an explicit operating margin target now on record, that figure is stale, and management neither restated nor withdrew it. Leaving it standing while raising revenue is a quiet way to let the sell-side do the raising.
8. The sequential decline in communications and other. Q2 is guided down mid-single digits sequentially in a segment that just grew 28.7% year over year, with no explanation offered and no question asked. It is 24% of revenue.

Market Reaction

• Pre-print setup: MRVL closed at $198.70 on May 27, entering the print up 133.8% year to date against the S&P 500's 9.9%, up 211.3% over the trailing twelve months, and up 25.6% over the trailing thirty days. The 52-week closing range into the print was $60.19 to $208.26, so the shares came in within 5% of their highest close of the year.
• Reaction session (May 28): Opened at $198.75, effectively unchanged, traded a $194.70 to $207.40 range, and closed at $204.83, up 3.1% or $6.13. The S&P 500 rose 0.6% on the same session.
• Volume: 52.6 million shares against a 30-day average of 26.4 million, or 2.0 times normal.

A 3.1% gain on a beat, a Q2 guide 12% above the prior quarter, a FY27 raise of more than $0.5B and a FY28 raise of roughly $1.5B is a thin reward, and it is the most informative number in this report. The stock did not fail to react because the print was ambiguous. It reacted modestly because the framework it delivered was already substantially in the price after a 25.6% run in the preceding thirty days and a doubling since the March print.

The intraday pattern reinforces the reading. The shares opened flat, traded 2.0% below the prior close at the session low, and finished near the upper end of the range on twice-normal volume. That is a market working out whether an outlook it had already anticipated justified paying more, and concluding modestly yes. Positioning was heavy going in, and the print validated rather than surprised.

Sell-side price targets moved higher after the print, which is the expected mechanical response to a framework raise of this size. It is worth separating that from the stock's own behavior: targets followed the price rather than leading it, and the 3.1% session tells you what marginal capital was willing to pay for the raise on the day it arrived.

Street Perspective

Debate: Is the fiscal 2028 outlook conservative or fully valued?

Bull view: Management effectively said the $16.5B carries upward bias, excluded scale-up switching entirely, and built the number on an assumption that cloud CapEx growth moderates to the 30%-plus range. Every one of the last four FY27 revisions has been upward, which argues the pattern continues and that $16.5B becomes $18B before FY28 begins.

Bear view: If the framework is understood as conservative by construction, the market has already priced the next raise. The stock trades at a mid-30s multiple on FY28 earnings that require the operating-margin target to be hit and the share count to stop expanding, neither of which is guaranteed. Buying a company on the expectation that its own guidance is too low works until the quarter the guidance is merely met.

Our take: The bulls are right about the number and the bears are right about the price. We model FY28 revenue near $16.5B with genuine upside toward $18B, and we still cannot make the current multiple attractive, because the earnings that reach shareholders are constrained by the 38% to 40% margin ceiling management set and by a share count that has grown 6.9% in two quarters.

Debate: How much should the NVIDIA relationship be worth?

Bull view: A $2B strategic investment plus NVLink Fusion integration plus a silicon-photonics collaboration from the vendor that defines AI infrastructure is the strongest possible validation of Marvell's SerDes and optics platform, and it opens semi-custom architectures that neither company could serve alone. None of it is in any revenue line yet.

Bear view: The economics ran the other way. NVIDIA converted $2.0B into an option on roughly 21.8 million shares at $91.84 while the stock traded at more than double that level weeks later, and the partnership deepens Marvell's dependence on an ecosystem it does not control. Strategic investments from dominant platform vendors have historically been better for the investor than for the investee.

Our take: The strategic value is real and we would underwrite it, but not at any entry price. The right way to hold both facts is that Marvell gained a durable interoperability position and paid for it with equity struck at less than half of today's price. That is a fair trade at $92 and a much less obvious one at $205.

Debate: Does custom concentration break the thesis?

Bull view: More than ten XPU attach programs, engagements at every U.S. hyperscaler, a balanced one-third split across the FY28 custom growth drivers, and continued new design wins mean the concentration risk is being actively diluted. The FY29 target does not depend on any single socket.

Bear view: The filings say otherwise about today. One distributor is 45% of revenue, three customers are 75% of receivables, and the FY29 custom target requires $5B to $6B of incremental revenue in one year from programs that remain unnamed. Competitive pressure in custom silicon is intense and hyperscalers have repeatedly shown willingness to multi-source.

Our take: The bear case is currently better supported by disclosure than the bull case is, and the gap between them is the most under-discussed risk in the story. Diversification is a stated direction rather than a present condition, and the concentration figures moved the wrong way this quarter without a single question about it.

Model Update

Valuation impact: Raising the target from $130 to $215 on a materially higher earnings path, and lowering the rating. The two moves are consistent: our estimate of the business improved by roughly 3% on the FY28 earnings line and roughly 14% on FY29, and the share price improved by 109% since the March print. Fair value range $175 to $240, against $95 to $130 previously. The $215 target is roughly 36x FY28E and roughly 27x FY29E, which is a premium multiple appropriate to a company compounding earnings above 40%, applied to estimates we are comfortable defending.

Thesis Scorecard Post-Earnings

Overall: The operating thesis is stronger than it was in December on every bull pillar, and the investment thesis is weaker, because the price has moved further and faster than the earnings. Two new bear pillars open this quarter, both sourced from the filings rather than the call: equity issuance and customer concentration.

Action: Trim. This is a downgrade on valuation and capital structure, not on execution. We would revisit a positive rating on a pullback toward the high $160s to low $170s, on evidence that scale-up switching is converting to disclosed revenue inside the FY28 window, or on a gross margin path for FY28 that lands materially above the 56% to 57% our operating margin arithmetic implies. We would move negative if the diluted share count passes 950 million without a corresponding step-up in the earnings path, or if the distributor concentration climbs further without a diversification disclosure.