Marvell (MRVL) Jumps 13% on Google TPU Deal: The $120 Billion Number Buried in the Warrant
Marvell disclosed a commercial agreement with Google on Wednesday and the stock opened 13% higher at $243.66, extending a year-to-date advance that already ran above 150%. The headline number circulating is $12.2 billion — the value of a warrant Marvell issued to Google covering 58,970,907 shares at an exercise price of $206.58. That figure is the least informative thing in the filing. The number that matters is $120 billion, and it is not stated anywhere in the announcement. It has to be derived from the vesting schedule, and once derived it tells you exactly how large Marvell believes this relationship becomes.
Here is the mechanism. A warrant is a contractual right to buy stock at a fixed price for a fixed period. Google is not a shareholder today; it holds an option that expires August 18, 2033. Of the 58.97 million shares, only 1,360,867 vest on time — in equal quarterly installments over the first year following execution. That is 2.3% of the total. The remaining 57,610,040 shares vest in 240 equal tranches, and each tranche unlocks only when Google and its affiliates have purchased another $500 million of Marvell custom silicon, measured from Marvell’s fiscal third quarter of 2027 through the end of fiscal 2033. Two hundred and forty tranches at half a billion dollars each is $120 billion of cumulative revenue from a single customer over roughly six and a half years. Marvell’s total revenue across all customers over the trailing twelve months is under $9 billion. Full vesting would require Google alone to buy more than twice Marvell’s entire current annual revenue, every year, for the life of the agreement.
That is not a forecast and should not be read as one. Nobody at Marvell expects the last tranche to vest. What the schedule does is publish an upper bound that the company was willing to write into a binding instrument, and the granularity is the useful part: at 240,042 shares per tranche, the market now has a revenue counter attached to Marvell’s largest new customer relationship. Every $500 million of Google business converts into a disclosable vesting event. Marvell has effectively agreed to report the thing investors most want to know and least often get, which is the actual ramp rate of a hyperscaler socket rather than the announcement of one.
The economics of the rebate are more interesting than the dilution. At the current spread between the $206.58 strike and $243.66 in the market, each vested tranche hands Google about $8.9 million of intrinsic value against $500 million of purchases — an effective discount of roughly 1.8%. Cheap. But the discount is not fixed in percentage terms; it is fixed in shares. If Marvell trades at $400 by the time the middle tranches vest, the same tranche is worth $46 million against the same $500 million, and the rebate has grown to over 9%. Google negotiated a discount that expands precisely in the scenario where the partnership works. That is the alignment the structure is designed to create, and it is why the deal is better read as a customer-incentive arrangement than as a financing event. The accounting will follow that logic: consideration granted to a customer generally reduces revenue as it vests, so the tranches will show up as a non-cash drag on reported top line rather than as an expense line, and the drag scales with the share price. Anyone modelling gross margin off the headline revenue ramp needs to carry that.
Marvell has run this play before. In December 2024 it issued Amazon Web Services a warrant for 4.18 million shares at $87.7706, with the bulk vesting against revenue through January 2030. The Google warrant is more than fourteen times larger in share count at more than double the strike. The escalation is the signal. Marvell is now systematically buying hyperscaler commitment with equity, and the size of the bid tells you what the company thinks the seat is worth.
The competitive read is where the significance concentrates. Broadcom has been Google’s custom silicon partner across multiple TPU generations and signed a long-term agreement in April running through 2031; it fell about 3% on the disclosure while Alphabet was unmoved, which is the market pricing this as share transfer rather than market expansion. That reading is premature. The scope Marvell disclosed covers AI inference accelerators, storage controllers, network interface controllers, memory interface controllers and near-memory compute — the silicon that surrounds the TPU rather than the TPU compute die itself. Those programs coexist with Broadcom’s role. The zero-sum reaction is a reflex, not an analysis.
It also understates why the deal is good for Marvell specifically. The durable asset here is not a compute socket. Compute sockets are exactly what a hyperscaler eventually in-sources; Google has been designing its own accelerators for a decade and has no structural reason to stop. Marvell’s moat sits in the attach layer — SerDes, optical DSP, memory and network interface silicon — where the content is required regardless of who designs the accelerator, and where the engineering is hard enough that the vertical-integration argument keeps failing. A memory interface controller ships whether the TPU is Broadcom’s or Marvell’s or Google’s. Near-memory compute attaches to the bandwidth constraint that is now the binding limit on inference economics. Marvell has taken the position that survives the customer’s own ambitions, and taken it inside the largest custom silicon program outside Nvidia. Nvidia’s own $2 billion convertible preferred stake, roughly 2.5%, was placed on the same logic: exposure to Marvell wins that contain no Nvidia silicon.
Base case holds $220–$270 into the Aug 27 fiscal Q2 print, which cannot move the model much because guidance was pre-spent in May. Bull case retests the $329.88 record if the October 6 investor day breaks out custom-silicon revenue by program and the Google ramp is visible in the FY28 frame. Bear case is a cohort derating to $140–$160 on multiple compression, which is what the $126-to-$400 street target spread across 38 analysts is actually arguing about — the terminal number, not the quarter.
The decision-relevant disclosure is the first vesting event. Watch for the quarter in which Marvell reports time-based and performance-based warrant shares separately. The gap between 1,360,867 and anything above it is the first honest measurement of how fast $120 billion is being approached.