Nvidia is building an entire data center in a single box.
NVIDIA Corp is about to take a page from Apple and transform into something bigger than it already is.
Only a few companies have successfully merged the worlds of hardware and software and become true tech platforms. Apple is the standout, of course. Apple was first a hardware company, and for many years built boring old desktop computers, before moving on to the iPod. But when Apple built the iPhone and the iOS platform that powered it, everything changed.
Apple became a platform that allowed your computer, phone, tablet, TV and wristwatch to be seamlessly integrated. In conjunction with iOS, Apple built the App Store, which opened the door to the app revolution. The App Store was extremely valuable to app developers and users alike, and Apple profited immensely. The iOS platform transformed Apple from a low-margin, cyclical hardware company to a high-margin tech platform with tens of billions of dollars in recurring services revenue.
Here's where Nvidia is looking at enormous growth potential. For decades, Nvidia has sold individual graphics processing units (GPUs) that are merely one component of larger systems, whether those be PCs, servers, cars or robots. Along with GPUs, those systems included central processing units (CPUs) from, for example, Intel and Advanced Micro Devices , and networking switches from Juniper Networks and Cisco.
The internal hardware in turn would include application-specific integrated circuits (ASICs) from companies such as Broadcom and Marvell Technology, cables from companies like Amphenol and TE Connectivity, and software from Microsoft and IBM, all packaged together and sold to the end user by companies such as Dell Technologies Inc., HP Inc and SUPER MICRO COMPUTER INC.
With the Blackwell platform (and its future generations of products), Nvidia is designing and selling all of the above components (and more) in one rack, directly to the end customer. In essence, Nvidia will be selling an entire data center in a single box. The only thing Nvidia doesn't do is build and assemble the physical goods. Just like Apple, Nvidia uses Taiwan Semiconductor Manufacturing to manufacture its chips, and Foxconn (TW:2354) (and similar companies) to assemble the final product.
There has been a lot of concern of late about the durability of Nvidia's continued dominance given the company's $2.6 trillion valuation and the potential for hyperscalers like Microsoft, Amazon.com, Alphabet's Google and Meta Platforms to decrease their capex budgets if they cannot build AI's "killer app" soon.
Nvidia CEO Jensen Huang himself is reportedly focused on ensuring Nvidia does not become to the AI revolution what Cisco and Sun Microsystems were to the internet revolution in the early 2000s. Those companies were essentially pure hardware companies that shriveled once the capex budgets of their customers declined and never recovered.
What most people don't understand about the AI revolution is that chip development leads AI development.
Why is Nvidia in a position to build an Apple-like platform and cut out so many of its competitors, and even its customers?
The reason is that Nvidia, like Apple, builds a vastly superior product to those of its competitors. Remember, Apple wasn't the first company to build a smartphone; recall that Research in Motion, or RIM, with its BlackBerry device dominated the market when Apple introduced the iPhone in 2007. But the BlackBerry was simply an inferior product compared with the iPhone. As the iPhone enabled the app revolution, so too is Nvidia enabling the AI and robotics revolutions.
You say you want a revolution?
Here's how Nvidia is enabling the next multitrillion-dollar revolutions:
First, and most importantly, what most people don't understand about the AI revolution is that chip development leads AI development, not the other way around. The ChatGPT "moment" occurred on Nov. 30, 2022, when OpenAI released its GPT-3.5 model and showed the true power of AI for the first time. But what made GPT-3.5 so much better than all of the other AI models that preceded it? It was the first one trained on Nvidia's cutting-edge (at the time) DGX A100 platform. Using Nvidia's latest chip, the H100, OpenAI researcher Andrej Karpathy was able to recreate GPT-2 in just 24 hours (something that took months with older chips). If you want to build the best AI, you have to use the Nvidia platform.
Second, Nvidia's products achieve optimum performance only when they are used as part of the greater platform. One of the keys to building the best AI is being able to capture the full capabilities of the GPUs used to train and run the AI. In order to do that, you need the most bandwidth and least latency possible. That's why Nvidia built its own networking architectures, NVLink and Infiniband; its own switches; and its own network interface cards (NICs) to allow maximum communication between individual GPUs.
Third, Nvidia's software stack is crucial to building the best AI. Nvidia's CUDA software allows developers to talk directly to Nvidia's GPUs. Almost all AI engineers at this point are familiar with the CUDA software and don't want to learn another software package from a vendor like Intel or AMD to instruct their GPUs. Moreover, Nvidia's software is optimized to help the end user get the most out of the hardware, and thus, by using any other provider, you are sacrificing performance.
These reasons, among others, make the Nvidia platform the absolute best product to use to build and run AI applications. They are what enabled the ChatGPT moment, and they will enable future AI innovations like Meta's Llama-4, Tesla Motors's FSD and Optimus Robot, and much, much more.
As a platform, Nvidia will slowly increase the services it offers to its customers in addition to its hardware. As of this time last year, Nvidia's services operation was already on pace to generate more than $1 billion in annual recurring revenue (services revenue was a minimal amount only a couple of years ago). Nvidia recently introduced its "Nvidia AI Enterprise" software, which Huang described as an "operating system for artificial intelligence."
Nvidia built the hardware that enabled the AI revolution. It is in the driver's seat to build the software to propel its continued expansion.