Marvell's $115B Lumentum Deal Bets AI's Future Runs on Light

Marvell's $115B Lumentum Deal Bets AI's Future Runs on Light

For years the AI boom has been told as a story about GPUs. Marvell just put $115 billion on the idea that the real bottleneck is the light carrying data between them.

Marvell Goes Shopping for Photons

Marvell Technology announced Monday that it will buy optical-components maker Lumentum in an all-stock deal valued at $115 billion, one of the biggest tech deals of 2026. Lumentum, roughly half Marvell's size, makes the optical parts that move data through networks and inside data centers, and it competes with Coherent and China's Zhongji Innolight.

It's Marvell's largest acquisition since its $10 billion buyout of Inphi in 2021, and it follows the $3.5 billion Celestial AI purchase the company closed in February. Nvidia, never one to miss a party, invested $2 billion each in Marvell and Lumentum earlier this year.

A Big Bet in the Middle of a Chip Selloff

Marvell is betting that Big Tech's AI spending spree is here to stay and that demand for optical components will keep outstripping supply. The company already helps firms like Amazon build custom AI chips, and in August it signed a deal to supply chips alongside Google's AI processors worth up to $120 billion in sales through fiscal 2033. Adding Lumentum gives it more firepower against its larger rival, Broadcom.

Investors are less sure. Since talks were first reported late last month, Marvell shares are down about 22% and Lumentum about 8%, caught in a broad chip selloff that sped up after Intel's weak forecast. That's a lot of skepticism to absorb on a deal paid entirely in stock. There's a policy angle too: Lumentum stands to benefit from a proposed US ban on new Chinese optical transceivers, which could tilt the supply race further in its favor.

The bigger picture is that AI infrastructure is turning into a full-stack game. Custom chips, interconnects and optics used to be separate businesses with separate winners. Marvell is wagering that the company owning more of that pipeline gets the biggest seat at the hyperscalers' table.

AI's next big constraint may not be how fast chips think, but how fast they can talk to each other. Marvell just paid $115 billion to own the phone lines.

Most of us won't be buying an optics company this quarter, but if you're working out which AI capabilities actually belong in your own products, our team can help you map it out.

Source: CNBC