($) Software is Why Nvidia Will Not Become Cisco
NIM is the tip of the spear
This is premium member only content. Thank you for being a paid subscriber of Interconnected Premium! If you aren’t yet part of our premium edition, I hope you join us!
When Nvidia eclipsed Microsoft earlier this week to become the most valuable public company in the world, my social media feeds were inundated by comparison charts and admiring words of Jensen Huang. Just as ferocious were posts of screenshots of near-identical headlines printed for Cisco in March of 2000 (like this one below), right before the DotCom bubble burst.
The pattern looks too conveniently similar. But lazy pattern matching belies Nvidia’s clear awareness of this history and active efforts to avoid Cisco’s fate.
What is Nvidia’s plan to not become the next Cisco? Software.
Yes, Nvidia is slowly but surely becoming a software company, even though it is known to most people as the GPU hardware company. It was an evolution that Cisco of the 90s, content with selling its routers and switches as the picks and shovels of the Internet, never made successfully. This lack of paranoia cost Cisco’s market leadership, allowing newcomers with better software offerings, like Arista Networks, to challenge its core networking business. Meanwhile, companies like Zoom beat its other assortment of software offerings, e.g. Webex.
Luckily for Nvidia shareholders, Jensen is as paranoid as ever in the classic Andy Grove sense, as most recently illustrated in this The Information profile. CUDA, Nvidia’s free software framework that’s tightly-coupled with its GPUs, has shown that the real moat is in developer-friendly software. Without owning that layer, even the most powerful processors will simply sit in the data centers collecting dust. Nvidia is now doubling-down on a grander software-hardware strategy.
Within this decade, I would not be surprised if Nvidia’s earnings reports break out a separate line item for its software revenue. (It is already charging customers $4,500 per GPU per year to use its software stack.) In fact, its software master plan is already unfolding. The tip of the spear of this plan is NIM (Nvidia Inference Microservices).