Well let's get to the big story breakdown this morning.
Bloomberg reporting that Chinese AI start Moonshot building its own official Kim model using 20,000 Nvidia chips that gained access to the hardware through Alibaba's data center.
And as traditional hyper scaler data centers face power and capacity constraints a new class of new cloud infra providers we are detecting how AI compute is delivered.
AI has announced a 3 year $71.9 million dollar agreement to supply and video Blackwell B 300 and 200 capacity to a leading AI platform.
Now this does mark.
The latest in a surge of multi-year contracts as enterprises moved from model training to low latency production infer workloads.
Well here at the New York Stock Exchange this morning to weigh in on what we're seeing in this space as Michael Manni Calco CEO of Cumulus AI.
Michael, good morning.
Thank you so much for joining us.
Well, it has been quite a busy week when it comes to the AI.
Trade from what we're seeing with the technology to partnership announcements, but first and foremost, when it comes to enterprises, tell us about the AI challenges right now.
I think the AI challenges for enterprises are really thinking about how they want to deploy at scale, how they need to budget at scale, and then how they need to plan for the killer abuse cases that are providing true returns on their investment and the infrastructure.
Yes, and when it comes to cumulus AI, what are you doing right now?
So we're providing the infrastructure that powers the AI applications and the AI models.
So when we talk to the enterprises, they're typically coming to us and saying, hey, we found some killer app use case that has a good return.
Now we're trying to figure out how to.
Optimize our spend on AI and our infrastructure is able to give them an option to do that through various levels of service.
They can host their own models.
They can host and host their own inference.
They can fine tune on our infrastructure, and it's all about the compute layer and that's what we're providing for them.
Yes, so Michael, for the layperson out there who is watching this right now and who is trying to understand what's beneath the surface, then what is the competitive advantage of cumulus AI.
To bring the AI to life, there's a full stack that has to happen.
It's everything from the energy source to the energy production to the land and the power and the data centers, and then it goes into the computes and the networking and the storage and everything behind the scenes.
It's a heavy infrastructure build that has to happen to bring these AI models to life, and that's our value proposition.
We package that infrastructure up in a nice consumable form so that when somebody is ready to deploy, they don't have to do all the heavy lifting.
Yes, and speaking of which, we know with the earnings coming out from the hyper scalers recently, there has been so much focus on the CAPX side of the business for good reason.
So walk us through what it actually means when we're talking about infrastructure here.
Yes, so infrastructure is in a lot of cases a heavy build, so it requires physical construction.
It requires physical cabling, physical infrastructure, and all of this is cap intensive.
So we operationalize all of that for them.
So they can consume it as a service, and that's one of our big selling points.
We also make it easier for them to deploy quickly and speed is one of our key tenets.
So that's how I tend to think of it.
And it's not just speed because we're talking about limitations and when we're talking about timeline in terms of building, there are a lot of challenges here.
So what do you expect to see in reality here within the US borders in the next several years?
In reality, what we're seeing is a massive amount of demand and supply that's been built.
In a way that we've done this since the beginning of the Internet, right, so that supply is oftentimes 34 years out from coming online, and the pace that AI is moving is, is incredible and that that that traditional way of building infrastructure just isn't.
Isn't enough and you're seeing that in all the earning assessments.
We continue to hear the hyperscalers saying we're capacity constrained.
We're going out and subcontracting capacity to keep up the demand from our internal teams and our own customers, and that's the type of capacity we're looking at new ways to bring online faster.
And a great way to think about that is we're not trying to build a monolithic Stargate 1 gigawatt type campus.
The things that you're Those are extremely important, but those take years and years and years to bring online.
So we're thinking about other ways to paralyze that and build smaller pockets of compute faster to try to keep up with the demand and provide a new way to bring demand online.
Yes, and I do want to expand on this by talking about your recent announcements.
So tell us why this makes a difference and what the plan is as we move forward.
Yes, I think.
The easiest way to say this is our customers are coming to us and asking for more compute yesterday.
Literally that's the request, and unfortunately that's not reality, and that's what's driving a lot of our growth and it's a great space to be in when you look at the demand for tokens which are powering all the applications which are powering all the user requests and the use cases.
We see that demand continue to rise because I believe that we're still in the early innings, that there's still a small percentage of the market that really has adopted the AI tools at scale, and as that continues to grow, the demand will just continue to mount, and we need to bring more capacity online quickly.
Yes, and finally, Michael, before I let you go, because you speak to so many stakeholders and you're entrenched in the space, we know that there has been a comparison of artificial intelligence to the beginning of the internet, but where do you think we are in the timeline when it comes to.
Artificial intelligence and what do you expect to see not just in the near term but as we head out a decade or two decades from now? and you're absolutely right, this is one of the largest infrastructure builds in history, and I believe, and I used to say a year and a half I'd say we're still in the parking lot waiting for the game to start.
I think we're in the game.
I think we're in the early innings, and I think as we find more and more use cases we've got a long way to go.
Well, Michael, it was a pleasure having you here.
Thank you so much for joining me and thank you so much for sharing all of your insights.
Appreciate it.
Thank you.
