Neocloud Brief · neocloud stocks · what is a neocloud · neocloud ETF · neocloud ETF
Neocloud ETF: NCLD, and the ways to own the theme
For most of 2026 there was no way to buy the neocloud theme off the shelf. Then Roundhill launched NCLD on August 6 — a pure-play neocloud ETF where two companies make up about 58% of the fund.
01The 30-second read
Fast read
For most of 2026 the answer to “is there a neocloud ETF?” was no — the theme was traded name by name. That changed on August 6, 2026, when Roundhill launched NCLD: actively managed, 0.65% expense ratio, roughly 21 holdings, with Nebius and CoreWeave alone making up about 58% of the portfolio.
There are now three distinct ways to own the theme, and they are not substitutes:
- NCLD — the actual neoclouds (GPU-as-a-Service, AI data centers). Concentrated, new, small.
- DRAM — the memory supply chain underneath. A different bet (HBM pricing cycles).
- DTCR — data-center REITs and digital infrastructure. Older, larger, income-oriented, far less AI-levered.
The choice is really about which part of the stack you want, and how much concentration you can tolerate.
02What NCLD actually is
The Roundhill Neocloud ETF (NCLD) is an actively managed fund targeting companies whose business is the neocloud buildout. Roundhill’s definition is concrete: a neocloud company is one with at least 50% of revenue, contracted backlog, or committed capex tied to GPU-as-a-Service platforms, high-density AI colocation, AI training and deployment platforms, data-center development, GPU-specific cooling and thermal management, power infrastructure, high-speed networking — or crypto miners transforming into AI/HPC operators.
Active, but not hyperactive
- Turnover generally expected to be limited to quarterly rebalancing
- Options trade on the fund; distributions are annual
Part of it is held in swaps
- The fund uses total return swaps to stay within RIC diversification tests
- That’s why position lists show swap lines alongside stock — a structure detail, not a red flag
03What’s inside
Top holdings as reported in August 2026. Weights shift with prices and quarterly rebalancing.
| Holding | Ticker | Approx. weight |
|---|---|---|
| Nebius Group | NBIS | ~29–30% |
| CoreWeave | CRWV | ~25.6–25.8% |
| IREN | IREN | ~8.6% |
| Hut 8 | HUT | ~6.2% |
| Galaxy Digital | GLXY | ~4.9% |
| TeraWulf | WULF | varies |
| ~15 other positions | — | remainder |
The concentration is the story. Two names at roughly 58% of the fund means NCLD behaves much more like a two-stock bet than a diversified basket — and those two are the most expensive, most leveraged companies in the group. A bad quarter from Nebius or CoreWeave dents the ETF regardless of the other 19 holdings. AUM was around $56.7M shortly after launch, with the fund up roughly 15% in its first week and daily volume above $46M in early sessions. Small and new — check spreads before trading.
04The alternatives, compared
| NCLD | DRAM | DTCR | Direct names | |
|---|---|---|---|---|
| What you get | The neoclouds themselves (GPUaaS + AI data centers) | Memory / HBM supply chain | Data-center REITs + digital infrastructure | One company, your choice |
| Style | Active | Index | Index (Solactive DC REITs & Digital Infra) | N/A |
| Expense ratio | 0.65% | 0.65% | ~0.50% | $0, plus your diligence |
| Size | ~$57M | ~$26B | ~$2.4B | N/A |
| Concentration | ~58% top two | 24 memory names | ~68% top ten | Total |
| AI leverage | Highest | High (cycle-driven) | Moderate | Highest |
| Income | Annual distributions | Minimal | REIT dividends | Company-dependent |
| Main risk | Launch-era crowding + top-two concentration | Memory pricing cycle | Rates/supply, diluted AI exposure | Single-name blow-ups |
NCLD is the purest expression of the thesis and the most concentrated. DRAM plays a different link in the same chain — memory pricing rather than compute rental. DTCR is the conservative, dividend-paying version, with meaningful exposure to tower REITs that aren’t really AI stories. Owning names directly removes the fee and the internal concentration, but concentrates your risk in a single balance sheet.
05The bull case
1 · The theme finally has a wrapper
- Own the basket without picking between CRWV and NBIS
- Launch drew real volume immediately
2 · The definition is coherent
- Unlike broad “AI ETFs” that dilute into megacaps, NCLD’s criteria target GPUaaS, AI colocation, power, and cooling
- It owns the buildout, not the buzzwords
3 · The demand backdrop hasn’t changed
- Morgan Stanley estimates roughly $2.9T of global data-center construction through 2028
- Sector contracted backlogs run into the hundreds of billions
4 · Take-or-pay economics
- Neocloud capacity is sold on multi-year contracts where customers pay whether or not they use it
- The structural attraction of the model, wrapped in one ticker
06The bear case
1 · Top-two concentration undoes the pitch
- ~58% in Nebius and CoreWeave — a concentrated bet wearing an ETF wrapper
- Both are loss-making, heavily leveraged, and priced for execution
2 · It’s tiny and new
- ~$57M in assets and a few weeks of history
- Liquidity, spreads, and survival through a drawdown are unproven — ETFs can and do close
3 · The fee isn’t trivial for a concentrated bet
- 0.65% to hold two names that already carry enormous risk
- Versus 0.50% for a larger, more diversified infrastructure fund
4 · It inherits every sector risk at once
- Capital intensity, debt, dilution, delivery delays, customer concentration, GPU depreciation, circular vendor financing
- And thematic ETFs typically launch after a theme has run — when inflows, and risk, are greatest
07What to watch
Fund-level
- Flows and AUM — does NCLD grow past launch curiosity, or fade?
- Quarterly rebalancing — does active management trim the top-two concentration?
- Liquidity — check the 30-day median bid/ask spread before any size
Underlying
- Earnings from the names inside: NBIS (Nov 10), CRWV (Nov 16), IREN (~Nov)
- DRAM’s memory pricing cycle, if comparing alternatives — a different clock entirely
- New competitors: Roundhill also launched LYTE (photonics & optics) alongside NCLD
08FAQ
Is there a neocloud ETF?
Yes — since August 6, 2026. The Roundhill Neocloud ETF (NCLD) is the first pure-play neocloud ETF. Before that, there was no clean way to buy the theme off the shelf.
What does NCLD hold?
Roughly 21 positions in neocloud companies. The two largest — Nebius (~30%) and CoreWeave (~26%) — made up about 58% of the portfolio as of August 2026.
How much does NCLD cost?
0.65% annually, in line with actively managed thematic ETFs.
Is NCLD the same as a memory ETF like DRAM?
No. NCLD owns the compute providers; DRAM owns the memory supply chain underneath them. Different links in the same chain, different cycles.
What’s the biggest risk in a neocloud ETF?
Concentration plus the sector’s own risks: two names dominate, both are loss-making and leveraged, and the group depends on capex not yet delivered. Fund size is also small, which adds liquidity risk.
Is a neocloud ETF a buy?
That depends on your own risk tolerance and time horizon, and this page is not a recommendation. The relevant questions are whether you want the concentration, and whether you’d rather own the names directly.
Sources & method
- Fund facts: Roundhill Investments fund page for NCLD (launch 8/6/2026, 0.65% expense ratio, active management, quarterly rebalancing, total return swaps for RIC diversification, underlying neocloud definition); ETFGI and ETF.com/etfdb launch coverage
- Holdings, AUM, performance and price: ETF Tracker reporting and Robinhood/Morningstar/KuCoin holdings summaries (Aug 2026) — weights vary by date and are subject to change; AUM ~$56.7M and 21 holdings as reported shortly after launch
- DTCR: Global X fund page and holdings reporting (Solactive Data Center REITs & Digital Infrastructure Index; ~0.50% expense ratio; ~$2.42B net assets; top ten ≈ 68% of assets; largest holdings Digital Realty, American Tower, Equinix, Crown Castle)
- Demand backdrop: Morgan Stanley estimate of ~$2.9T in global data-center construction through 2028, as cited by Roundhill
- All figures are third-party claims, approximate, and point-in-time; not independently verified by us