Custom AI chips: the upside is real — the revenue path may not be

The investment narrative around artificial intelligence has become overly simplified over the past two years. Markets often assume that rising compute demand automatically translates into linear revenue growth for every infrastructure supplier. In reality, growth visibility varies sharply across the value chain. The closer a company sits to core compute, the more predictable demand tends to be. The further it moves into optimization, customization and integration, the more results depend on the capital allocation decisions of a handful of hyperscale customers.

Marvell Technology sits firmly in that second category. It is not the face of AI hype, but embedded deep inside hyperscale infrastructure—where the focus is on moving data faster, cheaper and more efficiently between compute clusters. The long-term opportunity is substantial, yet near-term visibility is inherently limited. That tension explains why analysts can simultaneously point to significant structural potential while…