Alphabet just picked up warrants worth up to 12.2 billion dollars in Marvell

[MegaPete71]

Alphabet quietly reshaped its custom AI chip strategy this week, and the mechanism it used is honestly more interesting than a typical supplier announcement. Google now holds warrants to buy as many as 58.97 million Marvell shares at $206.58 each, and if fully exercised down the line that stake would be worth roughly $12.18 billion. That is not pocket change even by Big Tech standards, and it signals Google wants a genuinely serious ownership level lever in the company helping build pieces of its AI hardware stack, not just a standard vendor relationship.

The structure of the deal is really the smart part once you look past the headline number. Google does not just wake up owning a giant Marvell stake overnight, most of these warrants only actually unlock as Google hits specific purchasing targets running through fiscal 2033. That means the size of Google's eventual ownership position is directly tied to how much actual chip business Google ends up doing with Marvell over the next several years, aligning the two companies' incentives in a way a flat one time equity purchase never could. The partnership itself spans AI inference accelerators, networking, storage controllers, memory interface controllers and near memory computing technology built specifically around Google's TPU ecosystem, so this isn't just about buying finished chips, it's a genuinely deep multi year engineering relationship.

What's worth being clear about is this doesn't push Broadcom out of the picture at all. Broadcom already has a long term agreement in place to develop and supply future generations of Google's custom AI chips running through 2031, and Marvell simply gives Google a second heavyweight partner, more engineering capacity, and a real hedge against leaning too heavily on any single supplier during a period when Big Tech's overall AI infrastructure spending is running at genuinely historic levels across the entire industry. Alphabet clearly isn't waiting around for a supply bottleneck to actually hit before acting, it's building redundancy directly into its hardware pipeline well ahead of time.

The demand justifying all of this is already showing up clearly in Google's own numbers. Second quarter Google Cloud sales surged 82% to $24.8 billion, with the backlog reaching a genuinely enormous $514 billion. Those figures change the entire conversation around why Alphabet needs custom silicon in the first place, this isn't about chasing an AI trend for its own sake, it's about a cloud business scaling fast enough that more computing firepower has become a genuinely urgent operational necessity rather than a nice to have.

Worth noting too that Wall Street is already pricing in a real premium for this whole growth story, with Alphabet trading around $342.50 against a GF Value estimate of roughly $249.72, putting shares more than 37% above that particular valuation benchmark according to the coverage. That gap makes execution the entire game from here, Alphabet has real cloud momentum and a mature TPU ecosystem, and now it's tying billions of dollars of potential equity upside directly to future Marvell chip purchases as it works to keep enough silicon supply flowing to actually match the demand it's currently seeing

Alphabet's $12.2 Billion Marvell Option Reshapes Its AI Chip Strategywww.tradingview.com

[Mbappe]

The warrant structure tying to purchasing targets through 2033 is honestly the smartest part of this whole deal and it doesn't get nearly enough attention relative to the flashy headline dollar figure. Aligning ownership upside directly with actual purchase volume means Marvell has real skin in the game to keep delivering on schedule, rather than just collecting a check upfront and having no ongoing incentive tied to actual performance over the following years.

[Matthew51]

Worth remembering Broadcom is still very much in this picture through 2031, so framing this purely as Google ditching one supplier for another misses what's actually happening here entirely. This reads much more like genuine diversification and risk hedging during a period of historic AI infrastructure spending than any kind of supplier replacement story.

[Sequence19]

37% above the GF Value estimate is a real premium that Wall Street is pricing in here, and it puts a lot of pressure on Alphabet to actually deliver on the growth story rather than just talk about it. Markets have a way of punishing companies hard when premium multiples like that eventually meet reality and the growth doesn't keep pace with what's already priced in.

[HarleyQuinn32]

That 82% cloud revenue growth number is honestly the real story buried underneath all this chip supplier drama. Alphabet isn't building custom silicon because it's trendy, it's building it because the underlying cloud business is scaling fast enough that raw computing capacity has become a genuine operational bottleneck, and $514 billion in backlog is a staggering number that demands serious infrastructure investment to actually fulfill over time.

[Paul73]

Google essentially building a TPU supply chain with two major independent partners now, Broadcom and Marvell, feels like a genuinely smart hedge against the kind of single supplier dependency that's bitten other companies hard during previous chip shortages. Redundancy in critical infrastructure like this is worth paying a real premium for even if it looks expensive on paper in the short term.

[Harry43]

Equity warrants tied to purchase volume is honestly a structure I'd like to see more of across this whole AI infrastructure boom generally, it forces genuine alignment between customer and supplier instead of the usual one sided vendor contract dynamic. Wonder if we'll see similar structures pop up in some of the other major chip supply deals currently being negotiated across the rest of the industry right now.

[ShadowPilot83]

The near memory computing and memory interface controller pieces mentioned here are honestly the more technically interesting parts of this deal to me personally, way more than the flashier inference accelerator headline. Memory bandwidth keeps being one of the genuine bottlenecks holding back real AI performance at scale, and a deeper partnership specifically targeting that exact problem could matter more long term than raw compute headline numbers ever really do.