If you've been following the AI trade, you know the usual suspects: the chip designers, the cloud giants, the power utilities. But there's a quieter, arguably more critical layer of the AI stack that's starting to get its own spotlight. Defiance ETFs just launched a fund that's all about the stuff that stores and moves the data AI models need to function.
The Defiance Memory & Photonics ETF (BATS: PRAM) tracks the Solactive Memory and Photonics Index, which grabs the 20 largest eligible companies by free-float market cap and weights them equally. The index covers memory semiconductors, storage and networking chips, and optoelectronics. In plain English: it's a bet on the memory and optical connectivity that AI infrastructure can't live without.
This launch comes at a time when AI spending is moving beyond just processors. Sure, GPUs get the headlines, but AI models need massive amounts of data to be stored, retrieved, and shuttled between chips and servers at lightning speed. That's where memory and photonics come in.
Memory: The AI Bottleneck You've Heard About
Memory has quietly become one of the biggest winners of the AI buildout, especially high-bandwidth memory (HBM) that sits next to advanced AI accelerators. The fund's portfolio includes heavy hitters like SK Hynix Inc (NASDAQ: SKHY), Micron Technology, Inc (NASDAQ: MU), Nanya Technology (OTC: NNYAF), Winbond Electronics, SanDisk Corp (NASDAQ: SNDK), and Macronix International. SK Hynix has been vocal about strong demand for HBM, AI server DRAM, and enterprise SSDs, and the broader industry is gearing up for sustained AI-driven memory demand.
That demand isn't just talk. SK Hynix has planned a $4 billion advanced packaging facility in Indiana, with HBM4E production targeted for 2029. And the opportunity extends beyond HBM. AI servers are also boosting demand for conventional DRAM and high-capacity storage, creating a broader memory-cycle tailwind.
Photonics: The Other Side of the Equation
But memory is only half the story. The other half is moving that data. As AI clusters grow, electrical interconnects hit limits on bandwidth, distance, and power. That's where photonics comes in, using light to connect GPUs, switches, and data-center infrastructure.
The numbers are staggering. LightCounting expects 800G optical-transceiver shipments to more than double in 2026, while 1.6T shipments are projected to reach tens of millions of ports from a small 2025 base. The fund's photonics exposure includes Lumentum Holdings Inc (NASDAQ: LITE), and its broader semiconductor holdings include Marvell Technology Inc (NASDAQ: MRVL), which plays a role in high-speed connectivity.
Put it together, and you get a different kind of AI infrastructure trade: memory stores and feeds the data; photonics moves it.
A Truly Global Supply Chain
One thing PRAM makes clear is how international the AI hardware supply chain really is. Eleven of its 20 holdings are listed outside the U.S., with significant exposure to Taiwan. Its top 10 constituents include Nanya Technology (5.79%), Marvell (5.68%), Sandisk (5.47%), Winbond (5.35%), Lumentum (5.35%), Macronix (5.25%), Jeju Semiconductor (5.22%), Micron (5.21%), Innodisk (5.21%), and SK Hynix (5.16%).
With AI spending increasingly constrained by memory capacity, bandwidth, and data movement, PRAM gives investors a targeted way to access two infrastructure bottlenecks that sit behind the headline GPU trade. It's a reminder that the AI revolution isn't just about compute—it's about the plumbing.
