⚠ Selling the BallastModerate threat

Selling the automotive Ethernet business to Infineon for $2.5 billion in August 2025¹ was, on its own terms, a good trade: a strong price for an asset that did not fit the data-center thesis, and cash for a company that had spent years carrying acquisition debt.

The second-order effect deserves attention. Automotive was slow-growing and unglamorous, and it was also uncorrelated with hyperscaler capital spending. Every such disposal raises the company's exposure to a single end market, and Marvell has now concentrated to the point where data center is roughly 74% of revenue, up from 40% two years earlier². Focus and fragility are the same decision viewed from different distances.

A diversified semiconductor company gives up upside in a boom and survives a bust on the businesses that did not boom. Marvell has traded that structure away at the top of the largest capital-spending cycle its main customers have ever run.

There is no number that proves this wrong in good times — concentration looks like conviction while the cycle runs. The test comes on the way down. Watch what proportion of revenue sits outside data center when hyperscaler capex growth first decelerates, because that remainder is the entire cushion.