Nvidia is chucking money around again, this time with a $2 billion punt on Marvell Technology and a partnership pitch aimed at specialised AI chips.
Nvidia shares were up 5.59 per cent on the news, while Marvell jumped 12.80 per cent, suggesting traders like the smell of a tie-up.
The two outfits said the deal plugs Marvell into Nvidia’s AI ecosystem, giving customers building on Nvidia architectures “greater choice and flexibility in developing next-generation infrastructure”.
That wording is classic vendor fog, but the gist is simple enough: Marvell gets closer to the platform everyone is building around, and Nvidia gets more pieces made to fit.
A big chunk of the collaboration is silicon photonics, where light does the shuttling to move data around faster and with less heat drama than old-school electrical links.
In practice, that means more bandwidth between compute and networking, which is exactly where modern AI boxes start to choke once the models get silly.
Marvell will handle the custom chips, which is code for tailoring silicon to a customer’s workload and budget, rather than selling one-size-fits-all parts.
Nvidia will bring the supporting technologies and the data centre architecture, keeping its stack as the thing everyone has to design around, even when the silicon is bespoke.
Nvidia chief executive Jensen Huang said, “Together with Marvell, we are enabling customers to leverage NVIDIA’s AI infrastructure ecosystem and scale to build specialised AI compute,” and he is clearly not in the mood for modesty.
The interesting bit is how much of this ends up as genuine customer choice and how much is just another way to keep everyone living inside Nvidia’s walled garden.
