The DRAMa Behind Wall Street’s Fastest Selling ETF

While the fastest-selling ETF this year has had monster returns, investors aren’t necessarily cashing in.

That’s because so much of the flows into the Roundhill Memory ETF (DRAM) happened following the fund’s performance tear from early April to mid June, after which its price fell by about 37%, data from Morningstar show. In fact, the fund returned 80% this year, but the money invested in it has on average lost 20%. It might be a scenario of eager investors chasing returns, one of true believers buying at what they see as a discount, or more likely a combination of those. In any case, money hasn’t flowed out of DRAM, which since it launched this year has pulled in money faster than almost any ETF in… well… memory.

“It has just been a machine, hoovering up cash from investors,” Morningstar managing director Jeff Ptak said. The fund’s overall performance “has not accrued to their benefit, because so much of that money has come in such a short amount of time.”

Long-Term Memory

The difference between the fund’s overall returns since its launch and how the average dollar invested has performed is based on the timing of flows into and out of the ETF. It’s essentially a calculation for internal rate of return, Ptak said. And, though the difference between fund and dollar-weighted returns is particularly big in this case, it’s common for there to be a gap, as Morningstar’s recent report on the topic found. The figures Ptak calculated for DRAM are based on flows, assets and performance from April 1 through Aug. 3. If investors stay put and DRAM rallies the gap will narrow, he noted.

Still, the fact that so much money went into the fund when it did is a testament to the opportunities investors see in AI, Roundhill CEO Dave Mazza told ETF Upside. “The flows tell the best story: From DRAM’s peak on June 22 through its low on July 29, the fund took in $10.4 billion of net inflows even as the price fell, which is the opposite of what performance chasing looks like,” he said. “Investors are treating memory as a long-term secular growth story tied to the AI buildout, and less of a momentum trade. Typically, momentum money exits when price falls.”

While memory is one category that the firm sees as a bottleneck in AI, there are others, and it last week launched two ETFs in that vein:

• The Roundhill Photonics & Optics ETF (LYTE) focuses on companies replacing copper transceivers with optical ones.
• Its Neocloud ETF (NCLD) invests in companies renting out capacity from graphics-processing units (GPUs), something necessary for AI models to train and run.

The Rent is Too DRAM High: “Hyperscalers are building out datacenters, but are going to neocloud and asking for compute,” Roundhill ETF strategist Thomas DiFazio said, describing the neocloud as toll-road providers. Bitcoin miners have moved in on that as well, renting out their GPUs, he noted. And in the LYTE ETF, the company sees potential for photonics, as “light can operate over much longer distances and at much greater speeds, with less heat” than copper, he said. “There is continuous innovation that’s going on within photonics and optics.”