AI Advisory Services

Video Transcript: Solving big honking AI problems? The key to success is starting small!

I like big honking problems. I think just starting small and helping to innovate and build out small use cases—I mean, that's what we have learned internally over the last two years—is you got to just tackle all of these individual use cases. We've deployed over 300 use cases internally, and sharing that knowledge with our customers over the last couple of days has been awesome. I think they're really starting to kind of lean into how we can help them understand that we've been there, we've gone through it, we've got a ton of learnings, and now we're sharing them with them in an event like this.

You talk about that experience of being able to actually—we've gone through this journey, we know what it takes to get to where we are, and it's really, I think, that's the best part of the partners looking at us because of our capabilities. I think from the partners, everybody needs help, and there's very few companies that can give them additional scale certified across all those portfolios that you're talking about—workforce transformation, infrastructure transformation. And yeah, everybody—that collaboration of we need to go help customers together has been the key thing I've talked to partners about.

At our demo booth in the lab yesterday, we had all the AIPC bar to show people you don't always need that NPU or GPU to show these use cases. You can run some of the stuff on laptops as well. I think it's really important that's an aspect of generative AI. You hear these big energy stories and how are we going to get all this power, and I think it is important to recognize that by what we're bringing to the market with this capability is you can do this on a PC or you can do this on an H200 that fires up and sucks a bunch of power down, but you don't know which one's going to solve your use case for you. So I think that's why having that end-to-end silicon conversation and ability, I think lessens the fear. I think there's a lot of people that are afraid to have these energy conversations because they think the price tag is going to be too big or they can't get the data center space, but we can solve their problems really quickly.

Cloud-based solutions, laptop-based solutions—I mean, it's even moving towards phone-based solutions. This is a great time to be involved in technology, and SHI is really leading the way. And you said what our experts were telling all the customers today too—it's all about the use cases. So it's not just cloud, it's not just big data center factories, it's understand the use case and then leverage expertise to apply that right technology to the use case.

Well, when you look at the fact—what is it, 85% of projects in generative AI right now fail before they actually hit the production state? There's a lot to fear on the messing up part, and I think this is where, as we help the individuals that are bringing this into their companies to become the heroes rather than the zeros that ultimately fail, I think that getting them over that reticence and being able to use our experience and use the things that we have actually experienced and kind of walk them through is, I think, the most exciting part of the way we can solve their FOMO. But that's why the Mindspark example is so powerful, and I know it was talked about with our customers. It's a cultural transformation, it's not a technology issue necessarily—that's almost secondary. It's like, how are we going to get your employees on board to make this happen?

Well, I think getting your employees on board, that cultural element, is probably the biggest aspect of a successful generative AI project. Is your company ready to adopt this scale of technology? And you know, it's been forever the challenge of that change management of adoption, but the pace in which you mentioned—I mean, this pace of which it's going, what are we at, 300 use cases now in a year? But that's, I think, part of that ability to small use cases, build on the successes, and then ultimately that adoption becomes like wildfire. It's great to see it.

I don't know if it's FOMU or FOMO, but I do think there's kind of some mental blockers that people have been: like my data is not ready, my infrastructure is not ready. And there are ways to get started, whether it starts with a laptop and running some of these use cases on a laptop, or whether that starts with the cloud, or whether that starts with just experimenting with the NVIDIA services online. You can do some of these things just through a web interface. It's just experimenting with it. We've heard from a lot of customers, my data is not ready, but there are ways to approach your use cases that don't rely on really, really high-quality data or a whole data lake. A lot of customers, as you go down market, they don't even have data teams, they don't have data scientists, they don't even have a data lake. So it's like, how do you begin to try to even think about tackling this problem when everybody's telling you AI relies on data? But there really are ways to tackle these use cases. If it's as simple as writing an email better, automating a task, there are a ton of use cases that we found in that 300 that don't rely on quality data. So there's just ways to get started without addressing too many of the blockers that you might have.

Well, being the big data nerd at the table and recognizing that data comes in all forms, I think this is the other part of just opening and exposing people to what can be turned into a form of generative AI. Looking at whether they be PDFs or videos or images, those are all that multimodal ingestion for retrieval, augmentation, generation, implementation—those can be really easy wins. That's a repository on somebody's desktop in some cases. I heard, like, we're having dinner with a customer last night—100,000 employees across the globe, and he himself, the IT leader, has converted all of his training material for onboarding into videos in all the different languages. And he says it takes him now just less than 30 minutes. And all his peers are asking, did you do all those videos in all these different languages yourself? And of course, he says yes. But just thinking about those use cases of how that type of complex problem across the globe and how you can apply that and then learn and keep doing more to generate those 300 types of use cases, it's a good one.

We've always had a cloud-based AI lab sandbox for our customers. Today we announced we're going to be massively investing in our on-prem AI lab with the leading OEM manufacturers to sort of give customers the option of virtual versus on-prem. Did you hear much of that from customers over the last day or two?

Yeah, I mean, I think customers are always looking at how to get better time to value on any technology they make. And if we can help them in that selection process, POC process, whether it be cloud, on-prem, what's it going to look like, make me feel comfortable with that investment so I can utilize it right away—100%. We see that across, especially with the release with AI and having that ability in our new AI capability labs, but the ability in our other labs around cyber and other things that can just do spin up on demand to educate the teams. It's really that skill shortage that we can help people get in there and test those use cases out before they make a commitment, a purchase, etc., and really get that time to value.

I think it's eye-opening when we have customers come to an event like this, take them through facilities, take them through capabilities, hear from our presenters and talk about their experience. I just don't think customers even realize some of those things were available to them or that we had expertise in those areas. And then certainly being able to let them put their hands on it or let them try something on it without them having to buy it—some of this infrastructure is not even readily accessible to customers right now. So I think customers are seeing it as like, oh, I didn't even know SHI can help me do that, and that's fantastic.