It really boils down to what one player decides to do, and that's Nvidia, right? If you think about the forward curve of an asset, theoretically you'd want to see a downward trending forward curve with compute futures, right? This is logical. As more chips make it into the marketplace, it becomes cheaper and easier for people to actually get their hands on compute. Of course, this isn't actually a free marketplace. It's entirely up to Nvidia, right, the maker of these chips, to actually determine the prices. And they are, one could argue, disincented from creating a world where there's price transparency, right? They don't actually want people to know how much this costs, and b, from actually making things cheaper, because theoretically as the cost of compute goes down, their chips become less valuable. There are some sort of competing incentives here.
Welcome to Prof G Markets. I'm Ed. It is August 13th. Let's check in on yesterday's market vitals. The S&P 500 climbed toward a fresh record following the latest inflation report. More on that in a moment. On CME, the odds of a rate hike this year fell to 54%. Meanwhile, the Dow was roughly flat. Brent crude was relatively stable, as was the yield on 10-year Treasuries. CoreWeave shares rallied 19% after doubling its second-quarter revenue. And finally, SpaceX shares rose 10% after Elon Musk posted a recording of a company all-hands on X. In the meeting, Elon told employees that AI revenue will exceed all other SpaceX revenues by next month. Okay, what else is happening?
Wall Street will soon be trading AI computing power like a commodity. The CME, one of the world's largest futures and options exchanges, announced yesterday that it will start launching compute futures in October. Each contract will represent one month's rent on an Nvidia chip. The idea, in the CME's words, is to turn computing power into a quote standardized tradable commodity. This could give data center providers and AI companies both price transparency and the ability to hedge against swings in the cost of compute. But it all rests on two big questions. Number one, is compute actually a commodity? And number two, should we actually be trading it? Here to break this down, we're speaking with Rohan Goswami, business reporter at Semaphore. Rohan, this is the new future of AI. We're financializing it. We're turning it into a commodity. That's the new plan.
From the CME, what is your reaction to this? Does this make sense to you?
Look, we wrote a story a few months ago, right? This is not a new thing. Actually, this has been in the works for some time. Larry Fink made an oblique reference to it at Milken, right, earlier this year, that compute was going to become a financialized resource. And you quickly saw a couple of data providers actually step up to try and create the infrastructure for the CME and for ICE to actually offer these as products. I think they like to say that it's like oil or it's like electricity. It's scarce, it's hard to transport, it's sort of diffuse. I don't know that I fully agree with that, if only because oil is not something — oil is actually finite, right? And so the idea that this is one-to-one with oil falls apart there a little bit. And it's also, again, I know you and Ed have talked about this a lot, I know you've written about this a lot. It really boils down to what one player decides to do, and that's Nvidia, right? If you think about the forward curve of an asset, you can basically — and for those who don't really understand that, that's the idea of like what direction month by month a commodity will trade in. So they can go up, they can go down, right? And if they go down, that's basically suggesting that in the future things will become cheaper. So theoretically, you'd want to see a downward trending forward curve with compute futures, right? This is logical. As more chips make it into the marketplace, it becomes cheaper and easier for people to actually get their hands on compute. Of course, this isn't actually a free marketplace. It's entirely up to Nvidia, right? The maker of these chips to actually determine the prices. And they are, one could argue, disincented from creating a world where there's price transparency, right? They don't actually want people to know how much this costs, and b, from actually making things cheaper, because theoretically as the cost of compute goes down, their chips become less valuable. Of course, there's more nuance to that, and I'm certainly not at you or Ed's level versed in this, but there are some sort of competing incentives here.
And it also, I think, as you alluded to, raises a broader question, which is, do we want to financialize this stuff, right? I know you talked about this yesterday, but this isn't the only way that Wall Street is starting to dip its toes really aggressively into a space that had kind of before been a closed loop, right? So, we'll get to the Nvidia part in a moment because it's true. Nvidia is such an important piece of this because this first contract that CME is going to release here is going to be on the H100 chip, which is Nvidia's chip. But I just want to linger on the commodity piece of this for a moment. It seems like there is a lot of debate right now as to whether we can actually call AI compute a commodity like oil, like gold, like wheat, all of these commodities that trade on the CME. And it seems like one of the big issues is this idea that it isn't fungible. It isn't an identical commodity wherever you go because there are different types of chips. Chips can depreciate over time. You know, one generation of chip might be worse and older and less sophisticated than another. To what extent do you think that is a problem, or is that something that the exchanges can just figure out?
The exchanges to a degree have kind of figured it out. If you're thinking about units of measurement for whatever it is, hot-rolled steel is probably a good example, right? Not all steel is created equal. Some steel is crappy, some steel is good. But generally, we've agreed to come together and say, okay, as an index and as a forward curve of a commodity, it is essentially a monolithic thing, and very few if any CME customers are taking custody of physical commodities, right? So, I think the fungibility or the tangibility of it is less of an issue. And I think, look, there's actually real utility, as you pointed out, for hyperscalers or even for frontier labs if they know that demand is going to ramp up over time or they want to manage their downside. There are actual end users here. But I think that, broadly speaking, the comparison — again, I hate to keep coming back to this, but it is not ever, in my opinion, going to be a real market, because it would be analogous to Standard Oil, right? If there was one company that basically controlled in its entirety the flow of oil. Now you could argue that, of course, the OPEC countries kind of functionally do that, but there are a lot of countries that are not part of OPEC and don't engage in that, right? But here there's no other game in town except for Nvidia, right? So if Nvidia's stated mission is to make chips less scarce, to bring more chips into the ecosystem, you're depending on them basically to set the price, not just now but down the road, of this asset.
Yes, and it seems that this all relates to Nvidia and more specifically to this $500 billion memorandum of understanding financing package that they announced with the Avengers of Wall Street — KKR, Blackstone, BlackRock, Goldman, etc. It seems as though this is the first step to legitimize AI compute, specifically Nvidia compute, as a commodity that can be used as collateral to raise hundreds of billions, perhaps trillions of dollars of debt. Is that the gist of what's going on here?
I think there's a fair read there, but I also think it's kind of a risk-spreading mechanism, right? If you think about the deals that Nvidia has cut or said it will cut, right? Recall the $100 billion memorandum of understanding they signed with OpenAI. I don't even remember when that was, right? This has all been concentrated in this multi-trillion-dollar juggernaut that is basically lending its balance sheet to companies that don't have credit ratings. Anthropic, OpenAI, frontier labs, hyperscalers — they're stepping in and they're saying, 'Look, we are good for the money. Go ahead and buy our stuff.' Now, if you are an Nvidia shareholder, if you are an Nvidia bondholder, if you are the US government, frankly, that at some point starts to make you really uncomfortable that all this risk is lying and is concentrated with one entity that is incented to behave like a monopoly, right? If you think about what Nvidia is doing, they're subsidizing their chips at the expense of any potential new entrant and trapping these labs inside of their ecosystem. Now, this is functionally doing the same thing because, of course, Nvidia isn't bringing all these guys together to go buy chips from a competitor, but it is at least spreading the financial risk around, but also the financial upside, right? If you think about what it does for a Goldman or a KKR or a Blackstone, it provides them exposure to this play without actually having to invest directly in this play. You are essentially wrapping the risk around Nvidia's diligence, around their ability to design these fabs. If you are bullish about this space, it's great. You are partnering with someone who knows what they're doing, and all you're doing is being good for the money.
Looking at this memorandum of understanding, whatever we want to call it, the $500 billion —
Sure.
Memorandums of understanding, memorandum — as several have pointed out, it is not one cohesive vehicle. These are one-off vehicles that we don't really yet have much clarity on, but yes, sorry.
Yes. No, that is helpful clarification. Jensen Huang has said that this should assuage concerns about circular financing. The circular financing concerns, of course, being that these companies are kind of investing money into each other and then those companies pay the money back to them, which then the original company books as revenue, and then we think that the whole economy is growing, but really the money is being recycled. This has been discussed a lot on this podcast. Jensen Huang says that this should make those concerns less of a concern or put them to bed. Does it? Should this? What do you think?
I mean, I think, look, as the public markets and the private markets have started to look increasingly similar, so too have the risks associated in the public markets. And if you look at the S&P 500, I don't know off the top of my head what percentage gains have been propelled by Nvidia and any sort of AI-adjacent trade. But it is completely fair to say that you are either an AI have or have-not, and the have-nots go like this right now, and the haves, generally lumpy, go like this, right? And the same is true in the private markets. If you look at the sales that are getting the real sales, not the private-to-private stuff, it is stuff in and around AI. It is cooling systems. It is energy. It is land. It is actual data centers, right? Everything is an AI trade. And so at the end of the day, just because more people are financing it, it doesn't change the key-man risk. I think about like there was a company in the late '90s, right, that very famously went bankrupt later on, or was rather acquired — Lucent, right, which had a massive backlog of orders, totally into the tens of billions of dollars, and they said there's no need for anyone to be worried here. We are financing these things. We are helping people with these things. We are going to help our customers get here. Of course there's demand, right? Well, of course the dot-com bubble burst. All their customers went bankrupt, and Lucent was left with this theoretical backlog of orders that actually amounted to nothing. Shareholders were wiped. It was acquired. It is one of those situations where I look at this and I go, 'Well, there's no really avoiding this risk. Just because there are more players involved doesn't mean the actual risk has gone down.'
Yes. Yeah. Our research associate Dash Lone pointed me to another late '90s anecdote. He pointed out that Enron actually tried to turn internet bandwidth into a tradeable commodity as well.
Seemed like a good idea. Then we overbuilt the fiber-optic cable, and then suddenly the capacity flooded the market. The prices collapsed, and then suddenly it looked like a very bad idea. Is that a good analogy?
In fairness to Enron, Enron actually did build a very pioneering energy trading desk that was acquired — I can't remember by whom, maybe it was Citadel or another firm — but an incredibly lucrative desk that actually formed the bedrock of a lot of sort of commodities trading in that space generally. But to return to compute futures, I honestly, I don't know. I think at least with internet bandwidth, everyone was using this. This was something that everyone touched. You could see a market emerging for it at some point, right? But if you are fundamentally bearish on this space and think everything is going to zero, it's not like the internet where at some point there will be enough human beings to satiate this demand. If you think this is a dud, if you think this is going to zero, which some people do, and you think it's very circular, then yes, the comparison is apt in that it's fraudulent, but it's not an apt comparison in that it's not going to go anywhere.
