AI ETFs Shift From Semiconductors to Memory, Photonics, Compute

Investors are reallocating from broad semiconductor ETFs into funds focused on specific AI infrastructure layers such as memory, photonics, neocloud platforms and compute capacity. The Roundhill Memory ETF (DRAM) has attracted nearly $11 billion in net inflows since its early April launch. Over the past year the VanEck Semiconductor ETF (SMH) recorded more than $9 billion in net inflows.

High-bandwidth memory is a component used in AI servers. Roundhill’s DRAM holds roughly 15 names and gives exposure to memory makers including SK Hynix, Samsung Electronics and Micron Technology. The fund concentrates positions in South Korean memory firms that many U.S. investors find harder to buy directly.

Photonics-focused products have appeared in new launches and filings. Corgi Funds introduced the Corgi Lithography & Semiconductor Photonics ETF (EUV) in early May; the fund has about $200 million in assets and lists Taiwan Semiconductor Manufacturing Co., ASML Holding and Lam Research among its largest positions. Tuttle Capital launched the Tuttle Capital Pure Play Photonics ETF (FOTO) on May 29; FOTO holds roughly 16 companies with top weightings in Lumentum, IPG Photonics and Fabrinet. Several sponsors, including REX, Roundhill, KraneShares, Tema and Defiance, have filed or discussed photonics-themed products. A proposed Aura AI Photonics ETF would track the VettaFi AI Photonics Index, which breaks the photonics value chain into layers from raw materials through testing and quality control.

Roundhill filed for a Neocloud ETF with proposed ticker NCLD to cover companies building GPU-as-a-Service platforms, high-density colocation centers adapted for AI workloads, AI software platforms, data center development and operations, cooling and thermal management, power infrastructure and high-speed networking. The filing describes “Neocloud Companies” as firms tied to those elements. GPU-as-a-Service lets customers rent GPU clusters in the cloud rather than buy and operate the hardware themselves.

Exchanges and issuers are also preparing tradable compute instruments. ICE and CME Group plan to launch compute futures later this year that reference the price of renting AI compute capacity. ETF filings tied to that concept include a ProShares AI Computing Power ETF and proposed leveraged and inverse variants; a Roundhill compute fund filing that mentions futures and swaps; and filings from Leverage Shares and Volatility Shares for unlevered and leveraged computing-power ETFs.

Focused ETFs can concentrate holdings in fewer names and sectors, which may produce heavier single-stock or regional weightings and overlap across specialized funds. Investors considering memory, photonics, neocloud and compute-power products should review fund holdings, geographic exposure and the instruments used to deliver returns, including futures, swaps and leverage.