Nvidia is one of the few companies that were not Intel or AMD to ever ship an x86 core. They used this to great effect in their GPUs as their command processor,

Nvidia is one of the few companies that were not Intel or AMD to ever ship an x86 core. They used this to great effect in their GPUs as their command processor, they would offload parts of the driver binary into the GPU and execute locally for job setup; they later became an ARM licensee and changed that core out for an ARM one. Their hybrid CPU+GPU design they're trying to sell to customers (Ada Lovelace) is their full sized ARM core (a Tegra descendant) plugged into where the tiny ARM core (to be the command processor) goes. This is also how the Tegra X1 work; in a lot of ways, Ada Lovelace is just a really really big Nintendo Switch SoC, and not too dissimilar to how PASemi designed the M series SoCs, its a very common way to keep designs from ballooning silicon and power requirements in embedded land. Also, I'm not saying Nvidia's Mellanox acquisition was dumb; it was very smart. They bought the company that helped make AMD a household name in the supercomputer world: AMD famously used semi-stock Mellanox IB NICs, with custom firmwares, as their external facing hypertransport links combined with Mellanox IB switches; nothing else made could handle RDMA traffic as low latency AND has high bandwidth they needed, while also being flexible enough to be able to natively speak HTX directly on the bus. AMD had proven over a decade earlier that Mellanox knows how to be part of this. Nvidia could have used Mellanox parts at any point, btw. They just chose not to compete.