Marvell's $120B Google Deal Is an Earnout, Not an Order — What to Check Tonight

The most useful number in Marvell's Google announcement wasn't the $120 billion. It was $500 million.

Here is the setup you can reproduce tonight. On August 19, Marvell told the market it had expanded a custom-chip partnership with Google's Tensor Processing Unit (TPU) ecosystem, and handed Google a warrant to buy up to roughly 58.97 million Marvell shares — about $12.2 billion at a $206.58 exercise price. The warrant does not vest all at once. It vests in 240 equal tranches, and one tranche releases for every $500 million of custom-product revenue Google actually buys from Marvell through fiscal 2033.

That makes the frame of the whole story easy to state: $120 billion is a ceiling, not a contract. Reading it as booked order flow is where retail usually gets hurt. Run it through the screen, and the surging stock price reads differently than the headline.

What the Google seat is worth

The weight of the announcement is real, so it earns a real read. Marvell already designs custom silicon for Amazon's Trainium and Microsoft's Maia; Google's TPUs were Broadcom's turf for a decade. This agreement hands Marvell a foot in that door: AI inference accelerators, storage and network interface controllers, memory interface controllers, and near-memory computing.

That is the substance behind the word Piper Sandler used on September 10 when it started coverage at Overweight with a $270 price target — it called the $120 billion agreement "transformative". Piper sees Marvell winning custom-share and riding the next wave of optical networking, and models non-GAAP EPS growing at a 45% CAGR to about $19 by calendar 2030, the base of an 18x multiple. The stock moved up in the session after, and other firms reached higher: KeyBanc has a $400 target, Oppenheimer $325, Needham $300.

The vesting line is the wallet

Now the part that survives being written down. This warrant, on crypto terms, is a revenue-gated unlock. It doesn't matter how much hopium the analyst deck contains; each tranche only exists against money Google actually spends.

Warrant input	Value
Shares covered	~58.97 million (≈7% of Marvell)
Exercise price	$206.58
Gross exercise value	~$12.2 billion
Vesting	240 tranches, 1 tranche per $500M of custom-product revenue
Full-vesting revenue	~$120 billion
Exercisable until	August 18, 2033

Put that table next to the company's run rate and the size of the bet becomes legible. Marvell guided roughly $12 billion in revenue for the current fiscal year and about $18 billion for next; data center alone — its fastest engine, 79% of revenue — printed about $2.17 billion in one quarter, up 46% year over year. Full Google vesting therefore implies on the order of $15 billion a year of new custom-chip revenue from a single customer, back-loaded toward fiscal 2029, when Piper expects TPU attach volume to actually start. That is the difference between a company growing and a company becoming roughly two of itself.

Two readings, one deciding line

That is the observation. Here is the judgment, with the two readings kept honest.

The bullish read: Google is de-concentrating its chip supply the way Amazon and Microsoft already did, and Marvell's proven custom pipeline plus its near-monopoly in optical DSP makes it the credible second supplier. An investor-allocated warrant that lines up Google's incentive with Marvell's vesting milestones is a governance feature, not a gimmick.

The bearish read: the $120 billion is aspirational. Google keeps a separate long-term agreement with Broadcom through 2031, and nothing in the filing says how orders split between the two suppliers. The rev-gated vesting can stall out at a fraction of the ceiling if Google rebalances, dilutes through only part of the schedule, or simply buys less than the hype prices in. And you are paying for the ceiling: the stock has climbed more than 200% in a year, trades near its 52-week high, and carries the widest end of a trailing multiple — Piper's own $270 target sits roughly where the shares already trade, and Cantor stays neutral on valuation against Nvidia.

The data that separates the two readings is the same input every quarter: actual custom silicon revenue tied to Google, disclosed through filings and earnings — not the deck's adjectives. Watch whether the tranche schedule starts accumulating with the fiscal 2029 attach ramp, and whether Marvell holds the 38%–40% non-GAAP operating-margin target it said it expects to reach in the fourth quarter. The October 6 investor day, where Marvell lays out its long-term AI model, is the first real checkpoint.

When this playbook expires

This is a forecast with an expiry date, and the exit gets written before the entry. The Google-as-transformative thesis stops working when either of two lines breaks: the TPU-attach revenue does not show up on the vesting path by fiscal 2029, or margin guidance slips below the 38–40% target — the margin-line miss already knocked the shares down about 10% once this year. Re-verify at each earnings print by checking Google-related revenue against the vesting schedule before you treat the $120 billion as real. If the tranches are vesting, you are holding a growth stock with a whale-sized customer. If they are not, the $120 billion was always display material — a nice line for a headline, and nothing you could trade on.