00:00There seems to be a sigh of relief in the credit markets when it comes to NVIDIA and how perhaps
00:05they're not as exposed as they initially thought.
00:07Does this mean that the risk for NVIDIA falls?
00:10What are some of the concerns around circularity as well when it comes to some of the investments that NVIDIA
00:15is making right now?
00:19Absolutely. I think this is a very clever way of getting the debt of the books and creating a special
00:27purpose vehicle which can take most of the risk and be a conduit between the actual infrastructure players which are
00:35renting infrastructure and the compute to build the AI and train the AI and even have large scale inferencing done.
00:43And everything is boiled down to the token economics of the model.
00:48So if they can have a good margin at the demand side, I think they can repay the leasing fee
00:54for the compute.
00:55And NVIDIA gets revenue as soon as possible as it delivers the compute to these SPVs.
01:05Does it have the risk that you're actually transferring some of the leverage that we've seen from hyperscalers to just
01:11a sector of the financial market that might not be as transparent?
01:19I think it's sort of translucent, I would say, if it's not transparent.
01:24I believe most of these players are, if you look at the overall capex of the industry, right, so it's
01:31almost reaching $700 to $800 billion per year for the top hyperscalers.
01:36And when you look at this $500 billion fund with the different players, it actually is an extra cushion for
01:45a lot of the other players.
01:46It's still because it will be spread out over the few years.
01:50So eventually it's just 20 to 30 percent of what we are seeing annual spend from hyperscalers.
01:56So there will be long tail players which will be requiring some financing.
02:01Maybe it's a tier two data centers or tier three data centers.
02:04So I would say this will drive more of a flywheel effect for the entire ecosystem where we will have
02:10more players.
02:12The finance could be enjoyed by some of the big hyperscalers who are maybe cash strapped or some tier two,
02:19tier three, as I said.
02:21Yeah, it's been interesting to see this development, right?
02:24I mean, you have NVIDIA bringing in private capital.
02:27You have, of course, a big reliance on debt like we've seen from CoreWeave, for example.
02:32I mean, they did very well on their latest results.
02:34But at the same time, Intel's latest news on trying to raise a very large equity raise.
02:40How much of a problem right now is when it comes to capital availability and AI financing?
02:47And will we see these different models also arise from all of these different projects?
02:54Absolutely. You hit nail on the head.
02:57So everything is connected, right?
02:59If you want to build, if you want to invest a trillion dollars in the infrastructure,
03:05you are investing like 50% goes to the servers and 80% of those goes to the compute and
03:10memory, right?
03:11And for building that, for that insatiable demand, you need to have foundries which can churn out those leading-edge
03:19chips, right, from processors to memory.
03:23And if you look at TSMC and all these big players are already maxed out with capacity,
03:28and that's why you need newer players like Intel to expand and take up the boundary services and churn out
03:37those chips on time.
03:38And that's why to do that, they also need enough capital to buy this equipment from ASML of the world
03:46to produce these chips at scale, right?
03:49And that's why I think since the Intel stock has been doing well, it had some pushing to leverage that
03:57and use that money to influx into the overall capital expenditure for their foundry services.
04:05At least when it comes to this earnings season, are you seeing more signs that perhaps the demand for artificial
04:11intelligence compute is out there?
04:16Absolutely. There is a lot of demand we can see, and we are just scratching the surface right now.
04:22If you look at the amount of productivity gains right from content creators, developers to enterprises, we are seeing a
04:31lot of productivity gains.
04:32We are seeing how much they can do more with AI.
04:37Once you start burning tokens and create those software and everything really fast, it's very difficult to go back, right?
04:46And this will be more of a snowball effect where you keep on building and everything becomes a token economy,
04:53basically, right?
04:54And we are just scratching the surface.
04:57Maybe 5% to 10% of the entire world population or the businesses are still using AI.
05:03Imagine when everyone starts using AI at scale, and then you have multiple agents as well, which are per company,
05:11per developer.
05:11There are hundreds of agents doing automated tasks.
05:14That will burn a lot of tokens.
05:15So it will require a lot of AI token generation infrastructure.
05:20And that is where you see all these investments going in.
05:25Neil, going back to the issue of circularity and the fears around which metrics we should be watching when it
05:33comes to some of those red flags,
05:35are we talking mostly about free cash flow, which some of the earnings results from the hyperscalers, for example, led
05:42to that negative reaction, for example, Alphabet Google?
05:45Are we talking about credit spreads, like we saw NVIDIA, for example, reacting to the half trillion dollar of AI
05:52financing and Wall Street partnership?
05:54Which ones are you watching?
05:58So we are watching multiple levels.
06:00We are looking first at the demand side because that is the key driver.
06:03How much of the revenue is being able to be generated by the hyperscalers with respect to AI?
06:12That is a key metric to track in terms of demand.
06:16Coming back from demand side to supply side, you have to also then look at the bottlenecks, which are there
06:22for building this infrastructure to cater to that demand, right?
06:25Whether it's regulations or whether it's power or whether it's the availability of foundry services to build chips, right?
06:35Manufacture chips.
06:36So all of these are different, I would say, pointers we are watching at every stage because everything is connected,
06:47right?
06:47You have to look at end-to-end and obviously the capital expenditure of the books is very important to
06:56track as well.
06:58And that is why you will need a lot of these different creative deals to finance and keep the train
07:04going.
07:08Neil, recently we also saw some concerns around the geopolitics side of things with the U.S. now proposing this
07:15ban on Chinese optical modules, for example,
07:18at a time when we've continuously talked about shortages in different sectors of the AI trade, whether it's GPUs, memory,
07:25networking, power.
07:26What are some of the biggest red flags and biggest concerns in the next few years that we need to
07:31watch out for?
07:35That's a very good point.
07:37So if you look at, there is too much dependence on, I would say, two economies, either it's U.S.
07:43or China.
07:44It goes for compute and memory.
07:47It has been U.S. for manufacturing and for optics, for even materials like in a phosphide, right, which goes
07:56into networking.
07:57And as you move from copper to optics, as you build massive rack-scale systems, rack-scale infrastructure, you need
08:05all those cutting-edge modules,
08:08which are really manufactured precisely by a lot of Chinese ecosystems.
08:14And they have exported that.
08:16So you will see some bottleneck in these different ways.
08:20The first bottleneck we saw was with compute.
08:22The second one was with memory, and you see the memory stock price is going up.
08:27The third we'll see in terms of the CPU glut, which you are seeing in compute.
08:32The fourth is the network optics, which we are seeing with these Chinese players from a regulatory point of view,
08:40as well as from the raw materials point of view.
08:43There is a wave, there is a possible glut coming in that scenario where the capacities are fully hooked.
08:51And the fifth is going to be the power semi, semiconductors within power ecosystem and moving to newer materials.
08:57So these are the five waves you will see, which builds the AI infrastructure.
09:02And you will see these different waves, and many companies will make money or become trillion-dollar companies.
09:06Some will be just acquired.
Comments