NVDA and the launch of NIMs

Investment Analysis Clubs Saul’s Investing Discussions
1

I wanted to expand on something I mentioned in passing on an earlier post that makes me very excited about NVDA – their inference microservices, called NIMs. I’m posting this because I haven’t seen a lot of mentions on the board of this, but I find it quite relevant to mention.

NIMs are software AI snippets that a developer can use to “drop” AI into their code. They are like old-timey software library modules (such as “print” etc.) but for AI. In NVDA’s words: “It expands the developer pool by abstracting away the complexities of AI model development…”

What excites me about this is if NVDA gets this right, these NIMs will take off. They will take off because why would any software developer reinvent the wheel when they can focus on the rest of the car?

If a developer uses the NIMs, this ties the solution/product to the NIMs, bringing an additional revenue source for NVDA – possibly massive and ongoing - that is software-based.

In addition, NVDA is trying to create an AI “marketplace” where companies/developers can add their own NIM-like services for sale (likely new development on top of existing NIMs) and offer them too. Remains to be seen if it actually takes off, as NIMs are very new, just announced in March.

NIMs are free to use for development and research but cost once you deploy: $4500 per GPU per year, or $1 per GPU per hour.

So here’s an example – remember NIMs were just announced in March (although I imagine some companies had early access):

  • NVIDIA Morpheus … enables developers to create optimized applications for filtering, processing, and classifying large volumes of streaming cybersecurity data. Morpheus incorporates AI to reduce the time and cost associated with identifying, capturing, and acting on threats, bringing a new level of security to the data center, cloud, and edge.

And this mention on Seeking Alpha about a Zscaler and NVDA partnership:

https://seekingalpha.com/news/4114816-zscaler-nvidia-partner-to-deliver-ai-powered-security-solutions?source=content_type%3Aall|first_level_url%3Aportfolio|section%3Aportfolio_content_unit|section_asset%3Anews|line%3A3

(For those who don’t subscribe, here is one relevant paragraph:

Zscaler (ZS) will also leverage Nvidia Morpheus framework and Nvidia NIM inference microservices to deliver additional predictive and generative AI solutions to market, including the Zscaler ZDX Copilot with NVIDIA Morpheus and the Zscaler ZDX Copilot with Nvidia NIM.)

So this enables ZScaler to deliver additional AI capabilities in Cybersecurity quickly. And it locks ZS onto NVDA GPUs and additionally paying NVDA for the use of these software AI capabilities.

There’s also NIMs for healthcare (Clara), audio and video (Maxine), recommenders (Merlin), LLM-related stuff like text to speech and speech recognition (Riva) and I’m sure more to come.

If NVDA puts as much energy into this as they put into their chips and they get it right, this software will be instrumental to AI applications running everywhere. Equally important, it’s solidifying the moat and adding an additional annual revenue stream from software.

78 Likes
2

How is Snowflakes use of Nvidia NIMO ‘retrieval’ “game Changing”, per Jensen Huang, CEO of Nvidia, here at ~21:31

The whole of this video is exceptional, IMO.

Best

Jason

23 Likes
3

It ‘does’ sound impressive. But at this point I am a snowflake sceptic. They have been promising incredibility of one sort of another for the past 5 years. No doubt the new (3rd!) CEO is very smart and capable, but these corporate cheerleading fests are getting old. As a side note, image how Huang feels – he must do these videos appearances weekly at least.

7 Likes
4

It ‘does’ sound impressive. But at this point I am a snowflake sceptic. They have been promising incredibility of one sort of another for the past 5 years. No doubt the new (3rd!) CEO is very smart and capable, but these corporate cheerleading fests are getting old. As a side note, image how Huang feels – he must do these videos appearances weekly at least.

My post above is more snowflake related, rather than nvidia, so maybe not so pertinent. I’m with Mizz in thinking the NIMs are a really potent feature, and I hope they get a lot of traction. I even hope snowflake leads the charge on this. Really!

9 Likes
5

That announcement was this past March 18.

Not to be outdone, Databricks announced a quite similar-sounding “partnership” with $NVDA this past June 12:

My take: if you want to be (…or remain) a material force in the tech industry, you HAVE TO have a $NVDA “partnership”, because at the moment, $NVDA GPUs are the only available mechanism to achieve HPC at scale, and “time is of the essence” is an extreme understatement.

I speculate that going forward, a “partnership” may even be the ante you have to pay in order to even get in line to buy $NVDA GPUs in quantity, given the sky-high demand.

In any case, because of the demand, I imagine that $NVDA is in a powerful negotiating position when the terms of these “partnerships” are set.

Even the mighty $MSFT has a partnership with $NVDA:

26 Likes