Why Marvell and Memory Stocks Are Down After Nvidia Guided FY28 Growth to 70%
Nvidia’s quarter was not the problem. Revenue of $96.2 billion, up 106% year over year, earnings of $2.22 per share against $1.05, gross margin at 75.0% versus 72.7%, and a third-quarter guide of $108 billion plus or minus 2% against $57.01 billion a year ago. Colette Kress then told analysts to expect fiscal 2028 revenue growth of 70% while the street was carrying 44%, and added that customer forecasts point to demand roughly doubling, with the guidance reflecting supply constraints rather than orders. The stock rose about 6%. Marvell, Micron, SanDisk and Western Digital did not.
The most useful number in the release is the gap between those two figures. Customer forecasts imply something near a double; management will commit to 70%. Nvidia has said separately that it can currently serve roughly 70% of demand, and Jensen Huang was explicit that the constraint is the supply chain rather than any single vendor. That spread — call it thirty points of Nvidia’s own growth rate, withheld — is the memory shortage quantified by the largest buyer of memory in the world. Inventory went from $25.8 billion to $31.6 billion into the Vera Rubin ramp, and the company disclosed $145 billion committed in the quarter to lock down components. Server DRAM rose 64% in the second half of last year and TrendForce now models a 260% move across 2026. On the fundamentals, the print is a pricing-power disclosure for Micron, SanDisk, SK Hynix and everyone else selling bits into the rack.
And the memory names sold off anyway, which tells you the trade is no longer being priced on bit pricing.
Memory equities that are up two to six hundred percent this year have relocated substantially all of their value into 2028–2030 cash flows. That is a duration position wearing a semiconductor ticker. With the 30-year Treasury sitting around 5.3%, near levels last seen in 2007, an additional confirmation that DRAM is tight in 2027 is worth less to the multiple than another basis point of term premium costs it. The same mechanism ran on 18 August, when Micron, SanDisk and Western Digital fell 6% to 9% on no memory-specific news at all. What moved was the discount rate, and after that the coefficient on the discount rate. A shortage headline that arrives into a 5.3% long bond does not clear the bar.
There is a second reason the memory reaction was muted that is specific to the names in question. SanDisk spent the last two quarters converting spot exposure into contracted revenue — NBM agreements with eight customers covering roughly half of fiscal 2027 bits and about two-thirds of fiscal 2028 at floors near $0.29 per gigabyte. That structure is the reason the company survives the next downcycle. It is also the reason a confirmed shortage does less for it than the headline suggests, because a contracted bit does not reprice when Nvidia says it cannot get enough of them. The market is currently paying its highest multiples to the most cyclical exposure, not the most protected. Investors reading last night’s memory commentary as an upgrade to the whole complex should check which portion of a given company’s fiscal 2027 revenue is still open to spot before assuming it participates.
Marvell is a different problem entirely, and it is almost purely a positioning problem. Consensus is $0.93 in earnings on roughly $2.72 billion of revenue, up 39% and 35%. The stock is up about 183% year to date and trades near 55 times forward adjusted earnings against a sector median around 24 and its own five-year average near 37. Matt Murphy pre-spent the fiscal 2027 and fiscal 2028 outlooks in May when he raised both, which means a headline beat tonight moves nothing that has not already been paid for. The street targets bracket the disagreement rather than resolve it: Wells Fargo at $310, Susquehanna at $265, Morgan Stanley at $224.
The revealing detail is where today’s de-risking is happening. Options are pricing roughly a 10.3% move, which is below the 12% average of the last four post-earnings reactions and well inside a historical range that runs from 3.09% to 18.60%. The put-to-call ratio into the print is around 0.53, skewed bullish. Nobody is buying protection. Shares opened higher on Nvidia’s numbers — Marvell was up about 4% pre-market — and gave it back. That combination is holders trimming into strength, not hedgers positioning for a drop, and it is why the tape can look bearish while the derivatives look complacent.
What actually matters after the close is the Google line. The 18 August warrant covers 58,970,907 shares at $206.58, and 57.6 million of those vest in 240 tranches of about 240,042 shares, one for every $500 million of Google custom-products revenue recognised through January 2033. The strike was in the money on the day it was signed. The dilution schedule and the revenue schedule are now the same schedule, which means every half-billion that validates the custom silicon thesis issues stock against the holders of it. Shareholders cannot compute their own dilution without that revenue series, and tonight is the first opportunity for management to disclose it.
The moat argument is unchanged and it sits below the accelerator socket rather than in it. SerDes and optical DSP attach — 800G and 1.6T optics, 51.2T and 102.4Tbps switching, storage and memory-interface controllers — survives a hyperscaler deciding to design its own compute die. The XPU business may not. Nvidia’s roughly 2.5% convertible preferred stake is best read as buying exposure to Marvell wins that contain no Nvidia silicon, which is a statement about where the durable content is. Base case holds $220 to $270. Bull case retests the $329.88 record on a third-quarter guide above $3 billion or a named socket. Bear case is a cohort derating toward the five-year multiple, which puts the stock in the $140s to $160s without anything going wrong operationally.
Watch whether Google custom-products revenue appears tonight as its own reported line. The contract requires tracking it to the half-billion; the disclosure is now a shareholder entitlement rather than a courtesy.