The money printing is continuing. It will accelerate, and I think that AI is a bigger bubble than subprime in 2008. It will elicit a massive bailout, and that's what we are trading off of right now in crypto.
What do you mean by that, actually?
Think about SpaceX, OpenAI, Anthropic. These are the three major companies. They spend all the money. They've signed on the dotted line saying that they will buy hundreds of billions, trillions of dollars' worth of compute. Are they going to have money to do it? When the data centers get built — okay, cool, I built the data center, now pay me — where's the money? And these are all companies that don't make any money. In Anthropic's case, they just told us when you add in all of their costs like training and inference, it's actually $3 of cost to generate $1 of revenue. It's a terrible business.
If you're building in digital assets, you need infrastructure you can count on. That's BitGo — one platform for custody, wallets, staking, trading, financing, and more. Built for institutions and ready to scale with you. Visit bitgo.com to learn more.
All right, we're here with Arthur Hayes, the CIO at Maelstrom and also the founder at Flop Labs. Arthur, it's good to see you, man.
Yeah, good to see you in person.
It's good to do this in person, and honestly, on the ground it's very exciting right now because vibes are back. You talk a lot about why vibes are back, but why don't you just quickly jump into it with your strongest point about why crypto's in a good spot right now.
The money printing is continuing. It will accelerate, and I think that AI is a bigger bubble than subprime in 2008. It will elicit a massive bailout, and that's what we are trading off of right now in crypto.
What do you mean by that, actually? Because there's a lot of people who are talking about it being a bubble, but in terms of innings, where do you kind of place us in the build-out?
So I think we're, you know, two-thirds there. There's going to be a realization moment — let's call it late 2027, early 2028. Think about SpaceX, OpenAI, Anthropic. These are the three major companies. They spend all the money. They are the demand behind all the build-out. They've signed on the dotted line saying that they will, you know, buy hundreds of billions, trillions of dollars' worth of compute. Are they going to have the money to do it when the data centers get built? It was a free press release back then because they don't have to pay anything. They just say, 'Yeah, when you build the data center, I will buy it.' Okay, cool — I built the data center. Now pay me. Where's the money? And these are all companies that don't make any money.
— that have negative unit economics, and even in Anthropic's case, they just told us when you add in all of their costs like training and inference, it's essentially $3 of cost to generate $1 of revenue. It's a terrible business.
Well, it's interesting to think about too, because so much of this stems from Bitcoin miners as well. Bitcoin miners benefited from the idea of Bitcoin's price going up in dollar terms, which made all of this very interesting, but now I guess, how do you kind of connect the dots between what you've seen with the way Bitcoin was built out and what's happening now with data centers?
Well, the data centers are all built on three companies. There is no data center economy unless these three companies buy this compute.
Obviously in the beginning it probably was profitable to do so, like all things. And then we said, 'Oh great, these guys say they're going to build God. So let's just give them every ounce of capital we can.'
Yes, because God is a good TAM to own.
Yeah, exactly. You know, there's infinite TAM. Great.
Either you're going to die or you're a trillionaire — that's the bet. And so they're like, 'Cool, let's just get all the debt, let's get all the materials, let's get everything. We're just going to do a massive real estate project in the United States and Asia — please ship us all of these chips and memory. Australia, give us materials. China, give us other stuff. Let's all build these data centers.'
And these three companies are supposed to buy all this stuff based on economics that don't make any sense. They don't make any money. The customer doesn't pay the real cost of this AI thing, and if they did, they probably wouldn't use it as much. We'll see if demand and the desire of customers to spend money catches up with the ability to pay for these data centers in about a year.
Well, it's a good segue, and maybe what you're doing with Flop, because we keep hearing time and time again from almost everybody at any conference you go to in crypto: stablecoins are what agents are going to use, payments are going to be all stablecoins. You're kind of taking a counterpoint to that with what you're building at Flop and tying it to compute. So talk to me about why you're building that.
So my theory, from a fundamental level, is that the reason you hold any currency with a non-zero convenience yield is because that currency is directly convertible into what you need to survive. For humans, I can convert the fiat currency I receive from my job into food pretty easily — I walk down the street and buy some food with a Singapore dollar or a US dollar or whatever. For an agent, whatever the currency is — I think it's going to be Flop — they need to convert Flop directly into compute. The demand function and the supply function of Flop have to be directly related to compute and nothing else. So how is Flop created? Miners supply GPUs, they do useful inference for agents, and they get rewarded with Flop tokens. There's a direct linkage between Flop and the compute agents need to survive. A fiat currency, or the derivative which is a stablecoin, has nothing to do with compute — it has everything to do with government dictat. Government ministries decide, based on whatever economic theories, what the supply is and what the interest rates are. It has nothing to do with the productive capacity of an economy or compute.
