Ask anyone at the gym which companies own artificial intelligence, and you'll hear the same three names: Nvidia, Microsoft, Google. Marvell Technology almost never makes the cut. That's a mistake. Marvell designs the chips the hyperscalers build for themselves, plus the optical wiring that keeps their clusters talking to each other.
| Price LIVE | $272.29 |
| Market Cap | $244.7B |
| Forward P/E | 40.2x |
| Total Revenue (TTM) | $9.45B |
| 52-Week Low | $70.69 |
| 52-Week High | $329.88 |
| Analyst Consensus | Strong Buy (43 analysts) |
| Analyst Target Mean | $290.97 |
Strip the jargon and Marvell does two things. It designs custom AI chips, called XPUs, alongside one hyperscaler that pays for the design and then buys the finished silicon. It also sells the interconnect that ties those chips together: optical modules, PAM4 digital signal processors, coherent chips and Ethernet switches.
The second half is the part most people miss. A custom accelerator is only as fast as the network feeding it, and that network is the bottleneck Marvell sits on. Together, the two of them control roughly 95% of the custom AI ASIC co-design market, with Broadcom taking the larger share.
Demand is not the debate. In July, Google signed a custom-silicon agreement attached to its TPU ecosystem and received a warrant covering up to 58.97 million shares, worth roughly $12 billion at grant. Google becomes one of Marvell's largest shareholders. Amazon and Microsoft already buy custom work from the same bench.
Here's where it gets interesting for anyone holding the stock. Management has said the custom business ran near $1.5 billion in fiscal 2026, about 18% of revenue. It should grow more than 20% this year, then more than double in fiscal 2028. Custom silicon, though, carries thinner margins than the legacy chips it displaces.
Marvell already showed the trade-off. For the quarter ending in October, management guided non-GAAP gross margin to 57.5% to 58.5%. That midpoint sits about 90 basis points below the 58.9% just reported, a step down management blamed on the custom mix. The growth is real, and so is the drag on the margin line.
Think of Broadcom as the bespoke tailor with the grander atelier. Marvell is the tailor who also owns the mill that makes the cloth. The accelerator is the suit; the optical interconnect is the fabric it cannot be cut without. That vertical grip is the whole argument, and it is why no hyperscaler treats Marvell as a spare part.
Which brings us to the real question. Demand is showing up in the numbers. What the stock is waiting on is the margin bridge: can Marvell double custom revenue and still hold gross margin near 58%? Management has never broken out custom-silicon gross margin, so nobody outside the building actually knows.
The market is not asleep. Shares have more than tripled this year and trade near 40x forward earnings, about 17% below their June high. Cantor Fitzgerald lifted its target into the event and kept a hold rating. Morgan Stanley has raised its own while staying at equal weight. The Street believes the growth but has not yet decided about the margin.
Now the other side, from a short-seller's chair. Marvell's own 10-Q shows one distributor took 44% of net revenue last quarter, up from 34% a year earlier, and a single direct customer took another 16%. Two counterparties, 60% of the top line. Four customers were 72% of gross receivables, and the ten biggest were 82% of fiscal 2026 revenue.
That concentration hands the pricing power to buyers, not the seller. Broadcom clears 67% gross margin; Marvell's GAAP margin is 53%. The ramp is arriving at a lower gross margin, not a higher one, and guidance for the October quarter already bakes that in. If custom doubles again in fiscal 2028 while the margin keeps sliding, the bears were right all along.
Every AI cluster is a group chat between thousands of chips, and somebody has to keep the messages flowing. Marvell sells the plumbing, and bigger stakes mean more traffic. Marvell has proven the demand. What it has not proven is that a design shop can price like a platform.
That is the test waiting on the Investor Day on Oct. 6. Watch the fiscal 2029 revenue framework, which management has parked at a placeholder above $10 billion with an upside bias. Then watch the gross-margin bridge, the math that shows custom revenue doubling without gutting the margin line.
The demand is settled. The margin is the scoreboard, and in a matter of days Marvell has to put a number on it.
Disclosure: The Signal holds no position in MRVL. Positions may change. This is not financial advice.



