Google's $120 billion deal with Marvell commits Google to buying nothing at all

Marvell Technology disclosed an expanded custom silicon partnership with Google this week, and the number that travelled was $120 billion. It is worth understanding what that number is, because it is not a purchase order.

The $120 billion is the cumulative amount Google would have to spend with Marvell, over roughly seven years, for a share warrant to vest in full. It is a threshold, not a commitment.

What was actually signed

The underlying commercial agreement was signed on 29 July and covers a range of custom silicon designed to work with Google's tensor processing unit ecosystem, the in-house accelerators that run much of the company's AI workload.

Alongside it, Marvell issued Google a warrant to buy up to 58.97 million Marvell shares at an exercise price of $206.58 each. Exercised in full, that is roughly $12.2 billion of stock and about 6.3 per cent of the company.

The vesting schedule is where the structure becomes clear. Around 1.36 million shares vest across the first four quarters. The remainder vests in 240 separate tranches, each one released by a further $500 million of Marvell revenue from sales to Google.

Two hundred and forty tranches at $500 million each is $120 billion. That is the entire derivation of the headline figure.

The warrant expires on 18 August 2033. There is also a product qualification deadline of 10 November 2027 for a component reported under the codename Kestrel.

What Google is obliged to do

On the evidence of the disclosed terms, very little.

Google is not required to exercise the warrant. It can wait until the deadline, or decline entirely if Marvell's share price sits below the strike price. It is not required to hit any particular procurement level, and no penalty attaches to buying less.

The arrangement gives Google an option and an incentive. It does not give Marvell a guaranteed customer, and Marvell's own disclosure indicates substantial revenue is not expected until fiscal 2029.

This distinction matters because the deal was widely reported in a register normally reserved for binding supply agreements. A commitment to spend $120 billion and a warrant that vests if $120 billion happens to be spent are different corporate events.

The circularity question

There is a further feature of the structure that has drawn scrutiny from analysts who follow the sector closely.

Marvell's market capitalisation rose by roughly $18.6 billion on the announcement day. That single-day increase exceeds the $12.2 billion the warrant would be worth if exercised in full.

Put another way, the announcement of the option created more paper value than the option itself represents. Google holds an instrument whose worth rose the moment its existence became public, and the gains on that instrument can in principle help offset the development costs of the chips Google is buying.

Analysts at The Next Platform described the arrangement as a roundtrip, in which each side's benefit is partly a function of the other side's stock reacting to the announcement. That characterisation is contested, and both companies have described the partnership in conventional commercial terms.

Why Google is doing this at all

The strategic logic is easier to read than the financial engineering.

Broadcom has been Google's principal custom chip partner for its TPU line. Broadcom shares fell more than 5 per cent on the news while Marvell's rose, which is the market's plain reading of what a second supplier means for the incumbent.

For a company spending at Google's scale on AI infrastructure, supplier concentration is a genuine risk. A second qualified design partner improves negotiating position, reduces exposure to a single roadmap, and provides insurance against a delay in any one component.

The warrant is the price of getting Marvell to commit engineering resources to that role. Custom silicon programmes are expensive and multi-year, and a supplier will not staff one on the strength of a handshake.

The counter-argument

The case for taking the $120 billion figure more seriously than this piece has is straightforward.

Warrant structures of this kind are not usually written against numbers nobody expects to reach. Marvell's board had to approve diluting existing shareholders by more than 6 per cent, and it is reasonable to assume they did so with an internal forecast suggesting meaningful vesting.

Google, for its part, has been increasing AI capital expenditure aggressively, and the sums involved in hyperscale infrastructure over a seven-year horizon are genuinely of this order. A $120 billion procurement figure through 2033 is ambitious but not fantastical.

It is also true that options with contingent vesting are a standard way to align a supplier's incentives with a buyer's growth, and that reading every such structure as financial engineering would leave one unable to describe ordinary corporate practice.

What to watch

The meaningful checkpoint is not the announcement but the Kestrel qualification deadline in November 2027. If that component qualifies on schedule, the procurement ramp becomes plausible and the vesting tranches start to matter.

Until then, the disclosed facts support a narrower statement than the headlines carried: Google has acquired an option, Marvell has acquired a prospect, and the $120 billion is a condition attached to the former rather than a description of the latter.