00:00So talk to us first and foremost what the relationship with CME is going to be bringing us. How important
00:04is it to have a futures market and compute. It's incredible important. I'm glad that I came in this week.
00:11It's exciting week for the semi industry as a whole. As you mentioned energy span. We're probably going to spend
00:17over 10 trillion dollars this year. Think about the spot market from the energy to colocation to the server itself.
00:24Everything is at a commodity level. You need financial infrastructure to hedge the
00:29manager risk exposures. Think about the banks. They're underwriting trillions of dollars loans. How they manage risk exposure. How do
00:37they hatch their future volatility as compute going online. So from a commodity point of view think about how all
00:44your futures the wonderful job for people hedging oil fluctuation especially during the Iran war time. So it's incredibly important
00:52to have futures. You mentioned the banks but who else would be the players involved in such a market that
00:57need to be hedging right now. So think about
00:59people who naturally long GPUs. So the data center themselves and your cloud design houses fabs everyone long GPU exposure.
01:07They want a way to hedge it through short futures or productions. Whoever needs compute just like you and me
01:13like AI startups. They spend billions of dollars on GPUs. They need to hatch their cost as a big component
01:20to a line item. Traditionally SaaS companies they do not really have a cost line item aside from people. Now
01:26they do either from GPU level or token level. And
01:29it's a big line item coming up. So the whole push is about transparency and an ability to be able
01:36to put these sorts of assets on your overall balance sheet. But then when it comes to transparency just about
01:43the cost of compute. You've already been doing that with the compute exchange. What is it that you built there
01:47and who's using that.
01:48It's a great question. So compute change is spot market for GPU resources. So we do four contracts as well
01:53as reserve contracts for GPUs resources.
01:56So a lot of participants I would say all participants so far on the underlying consumption layer. So AI startups
02:02all the way to the new clouds. So we're two really big sides of companies.
02:06We work with European new clouds. We work with European new clouds. American new clouds. Southeast new new clouds. On
02:11the flip side we have startups and enterprises globally.
02:14This is really fascinating. You see how people try to do longer term contracts because they won't lock in the
02:19rate. They don't want the 40 percent daily volatility comes on demand GPU pricing.
02:23There's other way to hatch that through future options. So we want to make sure we have the right venue
02:28for the right use space.
02:29If you want to reserve GPUs. You want to physical deliveries. Go to compute change. You want to hatch to
02:34financial exposures. Go to CME in the future.
02:38Tell us about what sparked this whole idea. I mean for full transparency for our audience used to be part
02:42of the Bloomberg team.
02:44You used to work in this very building. What was it that you felt was missing in the market. Was
02:48there literally no transparency around access to compute from one AWS versus a Microsoft offering or a Neo cloud offering
02:55from Coreweave for example.
02:56I won't say it's no transparency but very little. So unless you're in the ecosystem you're actually in the Neo
03:03cloud transactions.
03:04You probably do not know what's the reserve price of people paying for longer term contracts. Think about term structure
03:09risk today.
03:11And then think about this lack of visibility with on demand pricing all the way to reserve which is three
03:16years five years on deal.
03:17And also GPU is not homogeneous products. You need to normalize it. You can't just say hey I'm paying the
03:23H100 with certain HVM. You're paying different HVM, different geolocation.
03:28We can't just do a simple average. You need indices to normalize the underlying assets to produce as a data
03:34that's trackable and representative to the price we are paying today.
03:37Look we're just talking to Manip Singh of Bloomberg Intelligence who is saying that we need to get to the
03:41bottom of NVIDIA's numbers that come after the bell on Wednesday about how much HVM or memory costs are really
03:47starting to impact their margin.
03:49What is it that you're seeing in your data about just the cost of compute at the moment.
03:53With the price going up. The price going up since December last year. It was pretty interesting. Last year everyone's
03:58talking about hey shouldn't compute price always come down and come down to zero by year A.
04:04It's not really happening. Supply demand curve shift every single day. The term structure the forward curve pattern shift every
04:10single day because people have different expectations.
04:13There are two different born nights come out all the time. Yesterday last week last year is fab and this
04:17year is memory and going forward maybe geolocation space.
04:22So every time is different. So pricing pressure is there. For you the pressure is on to get this deal
04:27solidified with the CME. You've announced it but we did say there's regulatory oversight.
04:31What sort of hurdles do you need to go through. What sort of questions will be asked. So we are
04:35actively working with CME and the review with CFTC and other regulatory bodies.
04:40It is a very traditional data future products. We're not talking anything that's exotic. Right. We have done that for
04:47oil for gas for any other asset classes.
04:49So I really don't see a major concern from anyone. And going global. How global is the offering that you
04:56currently have with CME or indeed the compute exchange.
04:59Where else are you likely to push in to more broad offerings worldwide.
05:02So we launched GPU indices at Bloomberg terminal last year. The world's first GPU indices. We cover all prices ex
05:10-China. So think about the European from the U.S.
05:13We normalize the geolocation to win data points. So anything ex-China is kept to our indices. So ideally around
05:19the world you can hatch that your exposure with our indices.
05:22CHRIS asymptotees. Viola
Comments