Dave Mazza on New Roundhill ETF Focusing on MLCCs, AI's 'Unsung Heroes'

SPEAKER_1 (0:00)

And joining me right now, Dave Mazza, CEO Roundhill Investments. I'm so glad you are with us. I know we've talked so much about the Roundhill ETFs, and we've had multiple times that we've talked about the Memory ETF, or the Photonics and Optics ETF, the NeoCloud ETF. We're going further down and being more pure play in the circuit board. What does that mean? Tell me about this AI exposure now and what it means to have CCML.

Dave Mazza (0:28)

Yeah, you're absolutely right. Really, Roundhill has prided itself on providing precise exposures for investors to access various components of the broader AI trade.

And the launch this week, ticker CCML, is actually focused on what I would call the unsung hero of AI, of the semiconductor circuit board, which are MLCCs and electronic components. And these are actually smaller than a grain of rice. Your phone, if you have a smartphone, has thousands of them. A circuit board has tens of thousands of them.

The Vera Rubin is going to have even more than that. And really what these components do is help control the power, right? So the GPU is getting taxed at different times. And really these are incredibly important to be able to control that power flow. And until recently, I think they were very, very underappreciated component of what's needed for our continued leadership in AI when it comes to the US.

SPEAKER_1 (1:29)

What kind of names are we talking about? And as you noted, these are obviously shipped in size, the MLCC ship in the trillions of units annually for these circuit boards components. Tell me more about this.

Dave Mazza (1:45)

Yeah, this is really important when we came to constructing CCML to provide that precision that I think investors come to think of Roundhill for.

There really is only almost like Memory or like Photonics, there's really only a few pure play companies. But what's interesting is most of them have been really difficult to access for US investors because the companies that make them like Murata in Japan, Samsung Electro in Korea are all based outside of the US. In fact, most of the production of this by a vast majority happens in Asia. So Korea, Japan, China, Taiwan, and really CCML hones in just on those names to provide exposure to this really unique component that's needed nearly in every device, and particularly when it comes to building AI servers.

SPEAKER_1 (2:34)

So I was looking at some of the details, what Goldman Sachs wrote, the Japan Ministry of Finance data, what that told you. You talked even about Murata, which is one of the most held in your fund there, the global share that it has. This is a global story.

Dave Mazza (2:51)

Yeah, it's a completely global story. Murata itself actually controls 35 percent of this market, and as you noted a moment ago, there's actually trillions of these shipped globally.

So it really is a global access play, but there's not US companies that investors really have known about because the companies that make them have almost been untouchable for most, and so they've been packaged together in a precise way, like we built DRAM and other ETFs to give the ability for investors to express this view on what is becoming another bottleneck in the AI trade, and one that if you look at the bill of goods for server racks or for a Nvidia chip, one of the components that has had the highest price increases, particularly for the high-end components that's needed in AI, is actually these MLCCs.

SPEAKER_1 (3:43)

So in these MLCCs, each one actually releases small amounts of electrical charges. So as the power system grows and AI needs more power, we need more of these circuits and such. Do you expect AI will need more power? She said sort of tongue-in-cheek.

Dave Mazza (4:02)

No, absolutely.

So again, what's interesting is that these are on one hand ubiquitous. They're in nearly every device that you can think of. They actually first came to be in 1960 as part of the NASA space program. The requirement to your point to be able to deliver this power with precision so that it's not overloaded, that would have been happening before. And as AI systems get more complicated, as the chip design itself gets more complicated and more powerful, you need more and more of these components to be able to make it work effectively in a way that people come to expect, especially when again, we think about the usage of a large language model. You put in a query, you want your answer as robustly as possible, as quickly as possible, and the MLCCs working inside the circuit board are helping make that happen.