Compute Is A Trillion-Dollar Market Trading In Group Chats | Brett Harrison & Andrawes Bahou

Andrawes Bahou (0:00)

Welcome to Token 2049

Jason Yanowitz (0:02)

Token 2049 is back October 7th and 8th, bringing together 25,000 attendees, 300 speakers and 500 exhibitors for the world's largest crypto event. Token 2049 is happening alongside and in partnership with our own digital asset Summit Asia. So you can experience both conferences in Singapore during the same week. Across Token 2049 week, there'll be more than 1,000 side events culminating with AFTR, 2049 and Formula One weekend. And the speaker line up is stacked. Shane Copland of Polymarket, Jeff Yon of Hyperliquid, Arthur Hayes, NASDAQ CEO, Adina Friedman and many more. Join us in Singapore October 7th and 8th for Token 2049 and Digital Asset Summit Asia.

Nothing said on Empire is a recommendation to buy or sell any investments or products.

All right, everyone, very excited about this. I've been wanting to do this episode for a while. I am feeling like I do not understand the compute markets well enough. And so I wanted to bring on two of the smartest people in the compute markets that I know, Brett Harrison, who is the founder of Architect and Andrawes, who is the founder and CEO of Compute Desk. So, gents, welcome.

Brett Harrison (1:18)

Yeah, good to be here. Thanks so much.

Andrawes Bahou (1:19)

Yeah, good to see you. Good to be here.

Jason Yanowitz (1:22)

So I also, for anyone watching on YouTube, I screwed up and we do not have two mikes today because our team is in Singapore for Das Asia. So anyways, we are there sharing a mic. So, guys, I think maybe Brett, I could ask you to start by, so you guys are very deep in the compute markets. We've got compute references and indices and markets and building perps on compute. And there's, I think for people who haven't been following, there's a lot of compute markets are going to go live at some point in the relatively near future, it seems like.

What is the problem in the market that compute markets are solved? Why are we building markets around compute?

Brett Harrison (2:01)

So the principal cost unit in the AI economy is companies renting GPUs or buying GPUs to be able to run various kinds of either training or inference jobs. And as I'm sure people know, the amount of money that we're talking about here is extraordinarily large. People are spending anywhere from 100 million to a billion to 10 billion at a time on large scale compute contracts. And for those companies that are building those out, sometimes they're called Neo Clouds or Hyperscalers, a lot of different companies in the space who are in the business of selling compute to other companies. There's a question here of what happens if the prices of compute go down over time, but if they go up over time. And that difference in the potential risk and what they could be offering this compute at is an unhedgeable commodity today.

And so that's why we are in the business of trying to collectively track the price of compute and to be able to build derivatives based on compute so people can actually hedge or speculate and have full transparency into what the forward cost will be of compute.

Jason Yanowitz (3:11)

Okay.

Can we maybe define some people? So we have the labs, there's hyperscalers, there's Neo Clouds, there's cloud companies, and many of these are synonyms for each other. Can you maybe, Andrawes, tee up the... Who are the players here?

Andrawes Bahou (3:29)

Yeah. Once upon a time, there used to be hyperscalers. So that's Google, AWS, whatever, and they'd buy GPUs, they'd drag up the hardware, and then they'd make it available to you on AWS, GCP, or Azure.

Then people started needing a lot more GPUs. And so the buyers naturally of hyperscaler capacity used to be small research labs, and then bigger research labs, AI companies, developers needing to train stuff, train new models, run inference, do all matter of ML workloads.

And then you're well aware probably of the bitter pill lesson, which is the idea that you should really scale up the amount of compute you have and the amount of data, but scale up the amount of compute you have, and your models will keep on gaining in capability. And so that along with OpenAI set off a race into doing something new until then where the consensus was, let's just build better algorithms. Well, then at that point in time, they're like, let's just build bigger computers. And so they started building them way, way larger. In the process of doing so, just because of how capital intensive this is, new financing structures started to emerge. So a new entrant into the market came in, and these are the lenders. Lenders, large private credit firms, not necessarily just private credit firms, credit firms of all sorts.