Nvidia (NVDA 0.29%), the company traditionally known for designing chips for high-end video game graphics on PCs, has ushered in a new era of AI. The generative AI chatbot ChatGPT quickly picked up tens of millions of users starting in late 2022, creating ballooning interest in other generative AI services -- nearly all of which are "trained" using Nvidia semiconductor systems. 

What followed was an epic financial report unlike anything seen before. Cloud infrastructure companies are all scrambling to buy as many AI training chips as they can get their hands on. But Nvidia has a new way to access its hardware that makes AI easier and more affordable to train than ever before: DGX Cloud. 

Wait, what is generative AI?

Let's get up to speed on what generative AI services like ChatGPT are. AI-powered software has been around for decades, quietly working behind the scenes to do things like organize and present information on the internet or auto-fill words while users type in a word document. 

Generative AI is a newer sub-field of artificial intelligence in which computer software is used to "create" new content. There are lots of iterations of it. It could be a user typing in text to prompt the AI to create more text (known as text-to-text), a text-to-image or text-to-video prompt that creates new visuals, image-to-video, video-to-text, or really any iteration that can be imagined. Generative AI is even used to translate speech in real time, power self-driving cars, and even to create software code, with the user explaining what they want the software to do using simple everyday language. 

Nvidia opens the door to new possibilities

The software algorithms that can accomplish this content creation (known as large language models, or LLMs) first need to be "trained." This is where Nvidia comes in. Using massive amounts of data, like that found on the internet or from a company's own private digital information, the LLM algorithms are trained in how to respond to user prompts by tracking relationships in sequential data, be it text in a sentence or sequential frames of video.  

Nvidia helped pioneer this work with over a decade and a half of research and development into data center supercomputing. Its video game chips (GPUs, or graphics processing units) can shred through massive amounts of data, so the company figured out how to apply these GPUs to data centers (a data center becomes part of "the cloud" when data and software stored on it is accessed via an internet connection).

However, up to this point, a company or start-up that wanted to train generative AI would need to purchase and install a great deal of data center infrastructure. This can get ridiculously expensive. Nvidia's latest flagship system alone, called the H100, can cost upward of $40,000 -- and that's just one piece of the infrastructure needed for AI training! A DGX SuperPod (which utilizes many H100s along with other chips) reportedly starts at about $7 million and can cost as much as $60 million.

A picture of an Nvidia H100 system for use in a data center.

An Nvidia H100 system. Image source: Nvidia.

DGX Cloud aims to solve this problem. Rather than a company needing to purchase and set up its own expensive data center, it can now rent access to Nvidia's AI training infrastructure and related software on a monthly basis. After testing early access with some early adopter businesses, DGX Cloud is now broadly available via subscription to Oracle's Cloud and through Nvidia's own infrastructure in the U.S. and U.K.  

A game changer for Nvidia

Generative AI can have transformative use cases across all industries of the global economy. Some research points toward AI soon adding $4 trillion in economic value every year. It's difficult to quantify what exactly that means, but the early results are quite tangible for Nvidia. The company reported revenue of $7.2 billion in its last quarter, but anticipates revenue to be some $4 billion more in the next quarter alone (to about $11 billion in total quarterly sales) due mostly to generative AI infrastructure demand.

While management didn't provide specific guidance for the second half of the current fiscal year, CFO Colette Kress said Nvidia has "procured substantially higher supply for the second half of the year" from its manufacturing partners. Suffice to say Nvidia's revenue is poised to rocket to new all-time highs this year and next. Even Tesla CEO Elon Musk recently doled out praise for Nvidia, saying the electric vehicle leader would take as many Nvidia GPUs as it could get, but that Nvidia has many other customers it has to prioritize, too.

This explains the wild optimism surrounding Nvidia stock right now. Shares trade for well over 40 times Wall Street analysts' expectations for next year's earnings. It's a premium price tag to be sure, so investors should tread carefully. There's limited visibility on just how much money Nvidia is going to haul in from its generative AI infrastructure, let alone its brand new DGX Cloud subscription service. But one thing's for sure: Nvidia has ushered in a new era of AI, and it's well positioned to continue carrying the torch for this movement for years to come.