Marvell Is Now a One-Segment Growth Story — and That Changes the Risk

When Marvell Technology reported its fiscal second quarter ended Aug 1, the headline was easy to skim over: total revenue hit a record $2.739 billion, up 37% from a year earlier, and management raised its revenue outlook for both this fiscal year and next. The number that actually describes this company sits one level further down in the release, and it is the one worth holding onto. The Data Center segment pulled in $2.1715 billion, up 46% year over year — which means a single segment now supplies 79% of total company revenue, up from 74% a year ago.

Data Center segment share of Marvell total revenue% of total net revenue

The Data Center segment rose to roughly four-fifths (79%) of Marvell total revenue in Q2 FY2027, up from 74% a year earlier.

Period	Data Center % of revenue
Q2 FY2026 (year ago)	74
Q2 FY2027	79

Do the division yourself and it verifies: $2.1715 billion divided by the $2.739 billion total is 79.3%. The published 79% is not a rounding trick; it is the arithmetic.

Diversification is no longer doing the work

That five-point mix shift is the whole story in miniature. A segment growing 46% while the company grows 37% is a segment outrunning the total — the whole rises more slowly than this one part, which is another way of saying the part is pulling the whole. However well Marvell's other businesses are doing, this is a data-center story now, carried by a single engine rather than a broad portfolio. The mechanics run through the hyperscaler buildout that powers the AI cycle. Big cloud companies are spending heavily on AI infrastructure, and that spend lands in two places inside Marvell: custom AI silicon — chips designed to a specific customer's spec, sometimes called XPUs or ASICs — and the connectivity that stitches them into clusters, the optics and high-speed Ethernet that move data between chips and racks. Marvell even spent last December buying XConn Technologies and Celestial AI to widen that electrical-and-optical interconnect layer.

That is the causal chain beneath the mix shift: hyperscaler AI capex feeds custom silicon and interconnect, the segment climbs toward four-fifths of revenue, and the segment's fast growth underpins the raised outlook. Everything the bull case is buying reduces to whether that chain keeps delivering.

A raised plan that rests on one segment

Management did not just report the quarter; it argued the future through the same lens. Marvell raised its combined revenue outlook for fiscal 2027 and fiscal 2028 to about $30 billion, from $20 billion a year ago. Chief executive Matt Murphy described AI-related bookings as "exceptionally robust" and pointed to a significant acceleration in the Custom business beginning in the second half of fiscal 2027 — the custom-silicon portion of the data-center portfolio, named as the second-half driver. For the current quarter it guided revenue to $3.15 billion, a step up from the $2.739 billion just delivered. That is guidance, not delivered results. The ~$30 billion two-year number is a plan, and the specific product-cycle details under it — the optical DSPs, the co-packaged optics, the 1.6T Ethernet — partly rest on acquisition press releases and secondary coverage rather than a published segment revenue series. Only one disclosed data-center dollar figure exists for this period, so the 74%-to-79% climb cannot be trended beyond this single quarter; the acceleration is asserted, not yet traced through successive reports. None of that discredits the claim. It does sharpen where the exposure sits. The Google custom-silicon partnership announced in mid-August, in which Google received a warrant to buy up to about 58.97 million Marvell shares tied to revenue milestones, is a telling detail — but it is one more proof that Marvell's AI growth runs through a handful of large custom programs rather than across a diverse customer base. Growth concentrated in one segment, and inside that segment in a few hyperscaler relationships, is growth that rises and falls with data-center capex.

The variable the thesis turns on

So the read for an investor is not whether Marvell matters in AI — it clearly does — but where the exposure sits. The driver to watch is the Data Center segment and whether the Custom business actually accelerates as guided in the second half, because the entire two-year plan is built on that one segment's continued outgrowth. If data-center growth slows below roughly 35% year over year, or its share of revenue slips back under 79%, the conversation changes from "broadening AI revenue base" to a single-engine story losing its throttle. That is the checkpoint. Everything else in the quarterly release — the margin line, the beat, the raised guide — is downstream of it.