Marvell's $12.2B grant to Google is a warrant, not a stake: it takes about $120B of chip spend to fully vest

Marvell issued Google a stock warrant, not a $12.2 billion equity stake, according to the company's 8-K filed 19 August 2026 for an earliest event of 18 August 2026 (Items 1.01 and 3.02, Exhibit 4.1 the Warrant Agreement). Reuters and CNBC both reported the deal. Any headline that says Google "gets a $12.2 billion stake" has the instrument wrong. Google gets the right to buy shares, and most of that right is contingent on how much it spends.

The instrument: a warrant priced at $206.58, exercisable to 2033

The commercial agreement is dated 29 July 2026; the warrant itself was issued 18 August 2026. It covers up to an aggregate of 58,970,907 shares of common stock at an exercise price of $206.58 per share, exercisable until 18 August 2033, issued unregistered under Section 4(a)(2) of the Securities Act.

That is where the "$12.2 billion" comes from. Multiply 58,970,907 shares by $206.58 and you get roughly $12.18 billion, the full notional value if every share vests and Google exercises the lot. It is a ceiling on the equity Google could acquire, not cash that changed hands and not a block Google already holds.

The vesting is the story: about $120 billion of spend to fully vest

Here is the number that is not in the "$12.2 billion" headline. Only 1,360,867 of the warrant shares are time-based, vesting in equal quarterly installments during the first year. Everything else, the overwhelming majority, vests in 240 equal tranches, one tranche for each $500 million of Custom Products revenue through fiscal 2033.

Do the arithmetic the filing invites: 240 tranches times $500 million is $120 billion of Custom Products purchases before the warrant is fully vested. So the honest way to size this is not $12.2 billion of equity in. It is a performance option that only converts in full if Google routes something on the order of $120 billion of custom-silicon spend through Marvell over roughly seven years. The equity is the reward for the spend, and it trails the spend by design.

What "Custom Products" actually covers

The tranches are tied to a defined revenue line, not to Google's total chip bill. The filing describes Custom Products as parts that "attach to the TPU ecosystem," and lists AI inference accelerators, storage controllers, network interface controllers, memory interface controllers, and near-memory compute. That framing matters: the warrant rewards Marvell for selling the connective silicon around Google's tensor processing units, the pieces that move and hold data next to the accelerators.

On the paper itself, the terms are lender-friendly to Marvell's cap table. The warrant may not be transferred other than to controlled affiliates without Marvell's consent, the time-based shares carry a lock-up, and Google receives customary registration rights so the shares it does earn are eventually saleable subject to securities laws and volume limits. This is structured as a long commercial partnership with equity attached, not a financial position Google can flip.

The Broadcom read is competitive, not a confirmed loss

The obvious market question is what this means for Broadcom, Google's long-standing TPU partner. The correct answer is: it is a competitive signal, not a confirmed socket loss. Nothing in the 8-K says Google is moving TPU business away from Broadcom, and the warrant is scoped to the "attach" silicon around the ecosystem rather than the TPU compute die itself. Read it as Marvell buying its way deeper into Google's custom roadmap with equity incentives, and treat any story about Google dropping Broadcom as unproven until a filing or a shipping product says so. For the broader antitrust and platform backdrop these deals sit in, see our coverage of Apple's services antitrust exposure.

The takeaway

Model this as a volume-linked option, not as $12.2 billion of equity in. The warrant is fully vested only if Google buys about $120 billion of Custom Products through fiscal 2033, and until then the equity accrues $500 million of spend at a time. Two things are worth watching on the actual filings rather than the headline. First, the four quarterly time-based vests over the next year, which are the only shares that convert regardless of spend. Second, whether any single $500 million revenue tranche actually prints in Marvell's Custom Products line. If those tranches start hitting, the partnership is real and the dilution is earned. If they do not, the $12.2 billion was always a ceiling, and the market priced a spend that had not happened yet.