Marvell Beat, Raised and Fell 10%. Read the Gross Margin Line

Marvell Technology reported record quarterly revenue of $2.739bn, up 37% year on year, beat on earnings with non-GAAP EPS of $0.94, guided the next quarter up 15% sequentially, and raised its full-year outlook. The shares closed at $216.62 on Friday 28 August, down 10.28%.

The number that explains the session is not in the headline results. It is in the guidance table: non-GAAP gross margin for the third quarter is guided to 57.5% to 58.5%, against the 58.9% just delivered. That is the sound of a company buying revenue growth with margin, and after a run that had already priced in a great deal of good news, the market charged it for the trade.

What actually happened

Marvell reported its second quarter of fiscal 2027 after the US close on Thursday 27 August. Net revenue of $2.739bn was a record and grew 37% year on year. GAAP gross margin was 53.1%; non-GAAP gross margin was 58.9%. Non-GAAP diluted EPS came in at $0.94, up around 50% year on year. On any reading of the quarter that has closed, this was a strong print.

The forward guide is where it turns. For the third quarter of fiscal 2027, Marvell guided to net revenue of $3.150bn plus or minus 5%, non-GAAP diluted EPS of $1.10 plus or minus $0.05, GAAP gross margin of 52.9% to 53.9%, and non-GAAP gross margin of 57.5% to 58.5%. Non-GAAP operating expenses are guided to approximately $655m, against approximately $1.015bn on a GAAP basis. The revenue figure implies roughly 15% sequential growth, which is a very large step for a semiconductor company. The margin figure implies compression at every point of the range against the quarter just reported.

Management was explicit about why: mix is the primary driver of the sequential gross margin headwind, with a strong ramp in custom silicon. That is not a disclaimer. It is the business model being described accurately.

Alongside the quarter, Marvell raised its longer-dated outlook — fiscal 2028 revenue of about $18bn, up from a prior target of about $16.5bn, against roughly $9.5bn over the last twelve months. Read cold, raising a two-year revenue target by around 9% is a good outcome. It was not enough, for reasons that have a date on them.

Why the market reacted the way it did

On 19 August, eight days before the print, Marvell disclosed a substantially expanded custom-silicon agreement with Google covering products attached to the TPU ecosystem — custom AI inference accelerators, storage controllers, network interface controllers, memory interface controllers and near-memory compute. The commercial terms reported alongside it were extraordinary: potential revenue of up to $120bn through fiscal 2033, and warrants for Google over up to 58.97m Marvell shares, an equity position that could reach around $12.2bn and make the customer one of the company's largest shareholders.

That announcement reset expectations for what the August quarter's guidance would contain. Chief executive Matt Murphy described the deal as game-changing and said custom AI chip revenue could prove a lot larger than currently modelled. But on the call he also said the quiet part: some Google revenue is already embedded in the forecasts through fiscal 2028, and the more meaningful contribution begins in fiscal 2029. He declined to put a new number on it, deferring to the company's Investor Day in October.

So the market got a raised fiscal 2028 target that largely predates the deal's real economics, an explicit statement that the payoff sits a year beyond the guidance horizon, and no revised model to anchor it. Against a share price that had run hard into the print on precisely that expectation, the arithmetic did not work. Volume of 47.7m against a three-month average of 40.3m is only about 1.2 times normal — this was a repricing by holders reconsidering the timeline, not a stampede.

One point of housekeeping on the move itself, because the figures in circulation differ. CNBC and 24/7 Wall St reported declines of roughly 6% and 7% respectively, written off after-hours and intraday prints. Sources anchored to the closing price of $216.62 — the Globe and Mail, Motley Fool, MarketBeat and TheStreet — agree at 10.3%. The close is the figure used here.

The Openbook read

The wire copy settled on "expectations were too high", which is true and is not an argument. The argument is that the guide contains a specific, quantified deterioration that the headline growth rate conceals, and that this is a Growth-versus-Profitability trade-off rather than a disappointment.

Growth is the strongest of the five factors and got stronger on Thursday. Thirty-seven per cent year-on-year revenue growth, a 15% sequential guide, a fiscal 2028 target raised to about $18bn against roughly $9.5bn trailing, and a customer agreement with a headline value of up to $120bn through fiscal 2033. Nothing in the print weakens this factor.

Profitability is where the pressure sits, and it needs to be read on the right basis. The comparison that matters is non-GAAP gross margin of 58.9% delivered against 57.5% to 58.5% guided — like for like, both non-GAAP. The GAAP figures tell the same directional story at a lower level, 53.1% delivered against 52.9% to 53.9% guided. Confusing the two bases would put several points of imaginary margin into the analysis, and the gap between them is roughly six points, so it matters.

But the dollars run the other way, and this is the part worth doing properly. Non-GAAP gross profit in the reported quarter was around $1.61bn. At the guided revenue midpoint of $3.150bn and a gross margin midpoint of 58.0%, third-quarter non-GAAP gross profit would be around $1.83bn — up roughly 13% sequentially. Even at the bottom of the guided margin range, gross profit dollars still grow around 12%. Marvell is not shrinking its profit pool to win custom sockets. It is growing the pool more slowly than the revenue line, which is a different and much more manageable problem.

The question that matters two years out is whether that ratio holds. Each 100 basis points of gross margin is worth roughly $32m a quarter at the guided revenue run rate, and roughly $180m a year against an $18bn fiscal 2028 revenue base. If custom mix keeps grinding the percentage down while volumes compound, the revenue line dominates and the model works. If margin compresses faster than volumes grow, it does not. The company has not yet given the market enough to distinguish between those two paths, which is precisely why October matters.

Momentum is the factor carrying the most risk and the least information. A large share of the move into this print was anticipation of the Google agreement being converted into raised numbers. It was not converted — it was deferred. A momentum score built on an expected disclosure rather than a delivered one is fragile by construction, and Friday is what that fragility looks like when it resolves.

Solvency is not materially changed by the quarter. The relevant forward consideration is the Google warrant structure: an equity instrument of that scale vesting against commercial milestones has dilution implications that will need modelling as it vests, and that is a capital-structure question rather than a balance-sheet one.

Reward/Risk is the honest place to land this. The reward is a company with a named hyperscale anchor customer, a multi-year revenue ramp with a credible upper bound, and gross profit dollars still compounding. The risk is concentration — a single customer relationship now carries a material share of the growth narrative — combined with a valuation that had already discounted the good version of the story and an admission that the largest contribution arrives in fiscal 2029. Absence of disclosure is what caused the derating, and that is a repeatable read for the next custom-silicon print: when a company defers the numbers to an investor day, the market prices the deferral, not the deal.

The read-across

The immediate comparison is Broadcom, which sits in the same custom-accelerator business at far greater scale and reports its fiscal third quarter on 2 September. Broadcom has guided third-quarter revenue to approximately $29.4bn, with AI semiconductor revenue guided to $16.0bn — growth of more than 200% year on year, and more than half of total quarterly revenue, against $10.8bn in the prior quarter. Broadcom designs custom accelerators for a roster of hyperscale customers including Google. Its gross margin commentary this week will be the cleanest available test of whether the custom-silicon mix effect Marvell just described is company-specific or structural to the category. If Broadcom guides its own margin down on the same mix argument, Marvell's Friday looks like the market learning something about the whole business model rather than about one company's quarter.

For Nvidia and the merchant GPU model, the read is the mirror image. Custom ASICs win on cost per unit of performance for a specific workload and lose on margin structure for the supplier; merchant silicon carries the pricing power. The more custom sockets shift to Marvell and Broadcom, the more the industry's aggregate gross margin migrates from the supplier to the hyperscaler — which is, from a hyperscaler's point of view, the entire purpose of the exercise.

The wider point for anyone screening the AI semiconductor complex — and our screener shows how wide the factor dispersion inside it has become — is that revenue growth and margin quality have started to separate within a group that has been traded as one thing. Marvell's quarter is the first clean example of a name where those two factors point in opposite directions at the same time.

What to watch next

• Broadcom's results on 2 September. Specifically the gross margin guidance and any commentary on custom accelerator mix. This is the category-level control experiment, and it is two days away.
• Marvell's Investor Day in October. Murphy explicitly deferred the revised Google modelling to this event. The specific things to look for are a fiscal 2029 revenue framework, a long-term gross margin target, and the vesting schedule and dilution profile of the Google warrants.
• Third-quarter results, expected in early December. The single most informative line will be where non-GAAP gross margin actually lands within the 57.5% to 58.5% guided range, and whether fourth-quarter guidance compresses it further. Two consecutive quarters of guided-down margin would turn a mix effect into a trend.
• Any customer concentration disclosure. As the Google ramp builds, the proportion of revenue attributable to the largest customer becomes the key risk figure in the filings, and it will move faster than the revenue line does.