OneChronos CEO on Three Players Shaping the GPU Compute Market

SPEAKER_1 (0:00)

Do you want to stay with AI, though, and pivot a little bit here, turn our attention to the development of a financial market for GPU compute. Welcome in our next guest joining us, Kelly Littlepage, the CEO of OneChronos. Kelly, thanks for taking the time to be with us from the nice floor today. Now we have highly liquid markets for commodities, oil, financial assets, stocks, but there is not really an efficient market right now for GPU compute. So why is that such an important gap to solve as AI demand accelerates?

Kelly Littlepage (0:31)

If you look at the forecast for what we have in front of us to build out over the next five years, there's really not a more important gap to solve than the efficient pricing of compute and power and financing those assets, which is why we're very focused on it here at OneChronos.

SPEAKER_1 (0:45)

And so as you were looking at the need for this, for this concept of a financial market for GPU compute, what are the first steps here? I mean, how did this process begin and where are we in the process of building out this infrastructure?

Kelly Littlepage (1:01)

We're early innings for everything we see. And the first steps in our mind are really the development of a funding market for compute, which is how you bring these projects online. How do you get more capacity and how do you efficiently price the capacity and the cost of capital for what's getting built out? So we're starting with the funding markets first.

SPEAKER_1 (1:20)

So as you start with the funding markets, I mean, if we talk about compute, we have to talk about Nvidia here. When investors are looking at Nvidia, they're oftentimes really looking at the hardware, at the hardware play here. You argue that that underlying compute market is still relatively opaque. So what are we missing by not having this transparent market for compute?

Kelly Littlepage (1:44)

What we're missing is a lot of efficiency around capital formation and really the ability to underwrite larger and larger projects and recycle capital back into the underwriting ecosystem. So in the same way that crude oil futures were instrumental in allowing more people to lend to and finance energy projects, we need the equivalent of that for the AI build out.

SPEAKER_1 (2:05)

So if you have the equivalent of that, what does a functioning market for GPU compute actually look like then, Kelly? I mean, who are the participants that you expect to be on each side of that market?

Kelly Littlepage (2:18)

That is an excellent question. It's so core to how this works because there's really three key participants. You need off takers, meaning the people that are actually buying and using the compute. You need the people building the projects for compute and you need the people lending to the compute projects. The three are very interconnected in a way that we believe common trial options and what we do at OneChronos is going to be key to unlocking the value in this market.

SPEAKER_1 (2:44)

You mentioned crude oil as an example there, but compute is different, right? Because the use cases and the requirements are changing so rapidly as for crude, they're pretty set. So how does it make compute more difficult to price? How do you plan to approach that aspect?

Kelly Littlepage (3:03)

Right.

So it's the efficiency of the forward curve, I would say, that crude brought to the market that was really powerful for lending markets. But as far as an asset or a way to think about things, you would really be hard pressed to come up with something more different between oil and compute.

That is why the market structure for this matters a lot. In our view, the way that you handle this is not that you try to jam everything into a commodity mold, it's that you embrace that compute is actually different, and you price the single assets. You have a world with very specific claims and then financial products and securitization around things like individual data centers as opposed to saying that there is just one barrel of crude equivalent for compute.

SPEAKER_1 (3:47)

Right now, existing prices, a lot of the time as I was doing my research, looks like it comes out to the cost of a GPU hour. But you are making an argument that that is the wrong level to be looking at here in terms of understanding the economic value of compute. So what is?

Kelly Littlepage (4:05)

Yep.

So we believe that you cannot make it fungible. You cannot make computer GPUs fungible, really, at the hardware level right now. But what Jensen has gotten at, which is absolutely correct, is that you can make it fungible as an investable asset, as a durable asset. So that goes back to the point of securitization being the better model for this than crude oil, really. It's the claims around cash that are coming off of GPUs, as opposed to saying that one cluster of GPUs in the Pacific Northwest is equivalent to another in Virginia. They're just not.