What ETFs does Roundhill offer for the AI theme?

Roundhill runs a shelf of thematic AI ETFs built around one idea per ticker. The broad ones are MAGS (Magnificent Seven, equal-weight) and CHAT (generative AI) and WTAI (broad AI and big data). The concentrated 'foundational AI layers' sleeves each isolate one bottleneck of the buildout: DRAM (memory), LYTE (photonics and optics), NCLD (neocloud / GPU-as-a-Service), and HUMN (humanoid robotics). Roundhill's pattern is to file a single-theme sleeve early in a narrative and let it track the obvious pure-plays. The complete live lineup, with expense ratios, AUM and flows, is tracked on QuantAbundancia's /etfs page.

What is the Roundhill DRAM ETF?

DRAM is the Roundhill Thematic Memory ETF, a concentrated fund built around the memory layer of the AI buildout: the DRAM and high-bandwidth-memory (HBM) makers whose supply is the tightest bottleneck in the AI cycle. It isolates the memory names (the US-listed pure-plays plus the Korea-listed leaders where accessible) rather than burying them inside a broad semiconductor fund. See /etfs for its live holdings and expense ratio.

Roundhill LYTE vs NCLD vs DRAM: what is the difference?

They isolate three different bottleneck layers of the same AI buildout. DRAM is the memory layer (the HBM and DRAM makers). LYTE is the optical-interconnect layer (silicon photonics, transceivers, co-packaged optics, the names that move data between accelerators). NCLD is the neocloud layer (the companies that buy GPUs at scale and rent the compute back out). Same concentrated-single-theme format, three different links in the chain from chip to model.

Are Roundhill thematic ETFs concentrated?

The single-theme sleeves are, deliberately. Where a broad semiconductor or AI fund holds 40 to 80 names, a concentrated Roundhill sleeve holds roughly 25 to 40, weighted toward the pure-plays rather than the mega-cap diversifiers. That concentration is the product: it gives clean exposure to one layer without diluting it with NVDA, AVGO or the mega-cap ballast. It also cuts both ways, the fund moves harder in both directions than a diversified basket.

How do I find a Roundhill ETF's holdings and expense ratio?

For a newly filed fund (like LYTE or NCLD), the holdings, expense ratio and NAV are not final until the SEC declares the registration effective and the fund lists, which for Roundhill has typically run 60 to 120 days from filing. QuantAbundancia tracks the whole Roundhill lineup on its /etfs page with issuer, expense ratio, holdings and 30-day net flow where coverage exists, updated when each listing prints.

Is a concentrated thematic ETF riskier than a broad AI ETF?

It carries more single-theme and single-name risk, yes. A concentrated sleeve rises and falls with one layer of the buildout, and a small number of pure-plays drive most of the move, so a stumble in one bottleneck (a memory price cycle, a capex pause hitting neoclouds) hits the whole fund at once. A broad AI ETF dilutes that with mega-caps and more lines. The concentrated fund is a way to express a specific view on one layer; the broad fund is a way to own the theme generally. Neither is 'safe'. This is educational research, not investment advice.

September 1, 2026·9 min read·QuantAbundancia Research

Roundhill AI ETFs explained: the foundational-layers shelf (MAGS, CHAT, DRAM, LYTE, NCLD)

Roundhill files a concentrated single-theme ETF for each layer of the AI buildout. The list, what each one isolates, filed vs live, and how a concentrated thematic ETF actually behaves.

ETFRoundhillthematic ETFAI ETFMAGSCHATDRAMLYTENCLDAI infrastructure

Roundhill has quietly built the most complete shelf of single-theme AI ETFs on the US tape, and the logic behind it is worth understanding before you buy any one of them. The firm's playbook is consistent: file a concentrated, single-theme sleeve early in a narrative, give it a ticker that names the layer, and let the basket track the obvious pure-plays. One bottleneck of the AI buildout per fund, filed first, before the theme is crowded.

This is the map of that shelf: which fund isolates which layer, what is live versus still filed, and how a concentrated thematic ETF actually behaves once you own it. It is the hub for QuantAbundancia's per-fund explainers (the $LYTE photonics ETF and the $NCLD neocloud ETF each get their own deep dive) and for the live data on /etfs.

The TL;DR. Roundhill splits into two groups. The broad funds give diversified AI exposure: $MAGS (Magnificent Seven, equal-weight), $CHAT (generative AI), $WTAI (broad AI / big data). The concentrated "foundational AI layers" sleeves each isolate one bottleneck: $DRAM (memory), $LYTE (photonics / optics), $NCLD (neocloud / GPU-as-a-Service), $HUMN (humanoid robotics). The concentrated ones are where the differentiated exposure is, and the concentrated risk.

The two groups on the shelf

Broad exposure. These are the diversified AI vehicles, closer to owning the theme than a single layer.

$MAGS - the Magnificent Seven, equal-weighted. Seven mega-caps, rebalanced. The most diversified-away-from-any-single-name of the group, and the least "thematic" in the bottleneck sense.
$CHAT - generative AI software and infrastructure. The original Roundhill AI sleeve.
$WTAI - broad AI and big data, a wider net across the whole complex.

The foundational AI layers (concentrated). This is the interesting collection, and QuantAbundancia's real coverage area. Each ticker isolates one link in the chain that runs from raw compute to a finished model.

$DRAM - the memory layer. The DRAM and high-bandwidth-memory makers whose supply is the tightest bottleneck in the AI cycle. Maps to the Memory / HBM bubble.
$LYTE - the optical layer. Silicon photonics, transceivers, co-packaged optics: the names that move data between accelerators once copper runs out of headroom. Full teardown in the $LYTE photonics ETF piece; maps to the Networking / Optical bubble.
$NCLD - the neocloud layer. The companies that buy GPUs by the tens of thousands and rent the compute back out. Full teardown in the $NCLD neocloud ETF piece; maps to the Hyperscalers / Cloud bubble and the Neoclouds theme.
$HUMN - humanoid robotics. The one layer that is not pure AI-infrastructure, an adjacent thematic bet on embodied AI.

The shelf, side by side

Ticker	Layer / theme	What it isolates	Concentration	Status
$MAGS	Magnificent Seven	7 mega-caps, equal-weight	Very high (7 names)	Live
$CHAT	Generative AI	AI software + infra	High	Live
$WTAI	Broad AI / big data	Wide AI complex	Medium	Live
$DRAM	Memory	HBM / DRAM makers	High (concentrated)	Live
$HUMN	Humanoid robotics	Embodied-AI pure-plays	High	Live
$LYTE	Photonics / optics	Optical interconnect	High (concentrated)	Filed (check /etfs)
$NCLD	Neocloud	GPU-as-a-Service	High (concentrated)	Filed (check /etfs)

Roundhill runs a broader shelf than this (income and options-overlay funds among others); the seven above are the AI-thematic equity sleeves QuantAbundancia tracks. The complete live lineup, with confirmed expense ratios, AUM and 30-day net flows, sits on /etfs, refreshed as each listing prints.

Why "concentrated" is the whole point

A broad semiconductor or AI ETF (SOXX, SMH, AIQ) can't give you clean exposure to one layer without diluting it. Buy SOXX for photonics exposure and you mostly own NVDA, AVGO and TSM, with the optical pure-plays as a rounding error. The concentrated Roundhill sleeve does the opposite: it strips out the mega-cap ballast and holds the layer's pure-plays at real weight. Where a broad fund carries 40 to 80 lines, a concentrated sleeve typically holds roughly 25 to 40, tilted toward the names that actually are the theme.

That is a feature and a risk in the same sentence. The concentration is why the fund tracks the layer cleanly on the way up. It is also why it falls harder when that one layer stumbles, a memory price cycle turning, a capex pause hitting the neoclouds, an architecture shift redistributing the optical winners. A concentrated thematic ETF is a view on one layer, sized accordingly, not a diversified core holding.

Two things to check before you buy any of them

Filed is not live. LYTE and NCLD are registration filings, not listed funds yet. A filed ETF has no confirmed holdings, expense ratio or NAV until the SEC declares it effective, which for Roundhill has typically run 60 to 120 days. Always check whether a fund has actually listed, and at what fee, on /etfs before treating a filing as buyable.

The investable-universe gap. A concentrated ETF can only hold what is listed. That matters most for the neocloud sleeve: the actual capacity leaders (Crusoe, Lambda, Nscale, Together AI) are still private, so NCLD is structurally over-weighted toward the listed names and under-exposed to the private leaders. The photonics and memory universes are more fully listed, so LYTE and DRAM have less of this gap. The first big neocloud IPO reshapes NCLD's basket; watch for it.

How this fits the QA map

Each concentrated Roundhill sleeve is the ETF wrapper for one of QuantAbundancia's bubbles or themes, which is why they are useful as a single-ticker way to express a bubble-level view: $DRAM for the Memory bubble, $LYTE for the Networking / Optical bubble, $NCLD for the Neoclouds theme at the seam of the AI Compute, Datacenter Power and Hyperscalers bubbles. The 12 AI bubbles ranked by empirical realness places all of these inside the supercycle taxonomy, and the bubble map validates whether each cluster trades as a real bloc or just market beta before you pay up for a concentrated wrapper on it.

To trade any of the listed names or funds from a US-retail account, see /stack/ibkr. Bubble shifts and rule-based alerts across these clusters are part of /pro.

Roundhill's bet is that each layer of the AI buildout deserves its own ticker. Whether any single layer is a durable investment or a bubble is the question the rest of QuantAbundancia exists to help you answer; that each one is now ETF-wrappable, and that Roundhill filed first, is the structural fact.

Live Roundhill lineup: /etfs - every fund, AUM, expense ratio, 30-day net flow, and bubble mapping, updated as each listing prints.

The per-fund deep dives: $LYTE photonics ETF · $NCLD neocloud ETF.

QuantAbundancia is educational research. Nothing here is investment advice. See /disclosures.

Perguntas frequentes

What ETFs does Roundhill offer for the AI theme?: Roundhill runs a shelf of thematic AI ETFs built around one idea per ticker. The broad ones are MAGS (Magnificent Seven, equal-weight) and CHAT (generative AI) and WTAI (broad AI and big data). The concentrated 'foundational AI layers' sleeves each isolate one bottleneck of the buildout: DRAM (memory), LYTE (photonics and optics), NCLD (neocloud / GPU-as-a-Service), and HUMN (humanoid robotics). Roundhill's pattern is to file a single-theme sleeve early in a narrative and let it track the obvious pure-plays. The complete live lineup, with expense ratios, AUM and flows, is tracked on QuantAbundancia's /etfs page.
What is the Roundhill DRAM ETF?: DRAM is the Roundhill Thematic Memory ETF, a concentrated fund built around the memory layer of the AI buildout: the DRAM and high-bandwidth-memory (HBM) makers whose supply is the tightest bottleneck in the AI cycle. It isolates the memory names (the US-listed pure-plays plus the Korea-listed leaders where accessible) rather than burying them inside a broad semiconductor fund. See /etfs for its live holdings and expense ratio.
Roundhill LYTE vs NCLD vs DRAM: what is the difference?: They isolate three different bottleneck layers of the same AI buildout. DRAM is the memory layer (the HBM and DRAM makers). LYTE is the optical-interconnect layer (silicon photonics, transceivers, co-packaged optics, the names that move data between accelerators). NCLD is the neocloud layer (the companies that buy GPUs at scale and rent the compute back out). Same concentrated-single-theme format, three different links in the chain from chip to model.
Are Roundhill thematic ETFs concentrated?: The single-theme sleeves are, deliberately. Where a broad semiconductor or AI fund holds 40 to 80 names, a concentrated Roundhill sleeve holds roughly 25 to 40, weighted toward the pure-plays rather than the mega-cap diversifiers. That concentration is the product: it gives clean exposure to one layer without diluting it with NVDA, AVGO or the mega-cap ballast. It also cuts both ways, the fund moves harder in both directions than a diversified basket.
How do I find a Roundhill ETF's holdings and expense ratio?: For a newly filed fund (like LYTE or NCLD), the holdings, expense ratio and NAV are not final until the SEC declares the registration effective and the fund lists, which for Roundhill has typically run 60 to 120 days from filing. QuantAbundancia tracks the whole Roundhill lineup on its /etfs page with issuer, expense ratio, holdings and 30-day net flow where coverage exists, updated when each listing prints.
Is a concentrated thematic ETF riskier than a broad AI ETF?: It carries more single-theme and single-name risk, yes. A concentrated sleeve rises and falls with one layer of the buildout, and a small number of pure-plays drive most of the move, so a stumble in one bottleneck (a memory price cycle, a capex pause hitting neoclouds) hits the whole fund at once. A broad AI ETF dilutes that with mega-caps and more lines. The concentrated fund is a way to express a specific view on one layer; the broad fund is a way to own the theme generally. Neither is 'safe'. This is educational research, not investment advice.