Research report
The theme is going up. Which name, at what risk?
The Neocloud Report · Edition 2, data as of 2 November 2026 (latest filed quarter of each company)
Five US-listed GPU-cloud names, one business model, and a valuation spread from about 3.6x trailing sales to about 240x. Edition 2 (data as of 2 November 2026, each company's latest filed quarter) ranks CoreWeave, IREN, Applied Digital, Nebius and SharonAI by risk-adjusted value: what you pay against what is actually de-risked, across valuation, leverage and customer concentration.
Look inside before you buy
3,190 words · about 14 min to read · 13 sections · data as of 2 November 2026
Table of contents
- What changed since Edition 1
- What this is
- What this is not
- 1. The one-page thesis
- 2. The method: risk-adjusted value on three axes
- 3. The five names, ranked
- 1. CoreWeave (CRWV): the cheapest proven pure-play, and the most levered
- 2. IREN: the balanced middle, now financed like the others
- 3. Applied Digital (APLD): the landlord with long leases (up from fourth)
- 4. Nebius (NBIS): best execution, demanding price (down from third)
- 5. SharonAI (SHAZ): the convex tail
- 4. The comparison, one table
- 5. The shared fault line, measured
- 6. How to access
- 7. The shape of the risk: a curve, not a block
- 8. What to watch
- 9. What this report cannot do
- 10. Sources
Excerpt
1. The one-page thesis
The market trades "neocloud" as one ticker with five names on it. That is half-right: the business model is uniform, the risk is not even close.
A neocloud is a GPU data-center operator. It buys NVIDIA chips, or leases the power and campus to house them, wires them into clusters, and rents that compute to AI labs and hyperscalers on multi-year contracts. The catch is structural: the cluster is bought before its revenue arrives, so underneath every neocloud is a financing story rather than a technology story. How cheaply it funds the build, and how locked-in the demand that pays for it, is the business.
Because the technology is a commodity (everyone buys the same NVIDIA silicon), the names do not differentiate on the product. They differentiate on three balance-sheet axes, and on those they spread across the entire risk curve instead of clustering. The October sell-off made the point: every name fell, by amounts that ranged from a third to more than half.
One table from the document
4. The comparison, one table
| Name | Market cap | P/S now | P/S Ed. 1 | 1 month | Leverage | Concentration | Where the risk lives |
|---|---|---|---|---|---|---|---|
| CoreWeave (CRWV) | ~$27.7B | ~3.6x | ~7x | -43% | High (~$35B debt) | Top two 62% of Q2 revenue | The balance sheet |
| IREN | ~$8.9B | ~12.6x | medium | -46% | Medium (GPU-backed, 6% to 9%) | Microsoft ~half of contracted run-rate | One anchor, revenue still ahead |
| Applied Digital (APLD) | ~$4.9B | ~5.6x | theme | -34% | Medium to high (secured notes) | CoreWeave and one hyperscaler | A tenant that is CoreWeave |
| Nebius (NBIS) | ~$38.4B | ~28x | ~62x | -39% | Low (cash ~ debt) | ~80% (Ed. 1 figure) | The price and the book |
| SharonAI (SHAZ) | ~$0.74B | ~240x | ~860x | -57% | Dilution and GPU debt | NVIDIA > half of contract value | The whole thing |
The preview stops here. The full document is delivered by email after purchase.
Same chips, opposite bets
Everyone buys the same NVIDIA silicon, so the names do not differ on the product. CoreWeave trades around 3.6x trailing sales with about $35 billion of debt; SharonAI around 240x on revenue that has barely started. One owns its sites outright; one is a landlord with 15-year leases. Ranked by what you pay versus what is de-risked, they spread across the whole curve, and the ranking moved since Edition 1: Applied Digital up to third, Nebius down to fourth, with the reasons.
The risk they all share
Customer concentration is the bear case in one sentence: a handful of AI labs and hyperscalers fund the entire buildout. CoreWeave's top two customers were 62% of its second-quarter revenue; IREN's Microsoft agreement is roughly half of its contracted run-rate; NVIDIA is more than half of SharonAI's contract value. That means concentration is not the differentiator, it is the correlation: a capex wobble at the top re-rates all five at once.
The part a chart cannot give you
The measured shape: CoreWeave and Nebius move together at about 0.80, which makes two of them about 1.1 independent bets, not two. IREN and Applied Digital trade with the converted bitcoin miners, not with the pure neoclouds. So a five-name basket is about two bets plus a tail. None of it is a recommendation, it is the arithmetic of how a convex, correlated cluster behaves.
What is inside
- What changed since Edition 1, in one list: the October fall, CoreWeave's debt, IREN's end of mining, the re-ranking.
- The three-axis method (valuation, leverage, customer concentration), scored high / medium / low per name so you can see which axis carries the risk.
- A scorecard for each of the five names with its latest filed quarter and a one-line bull and bear: CRWV, IREN, APLD, NBIS, SHAZ.
- The comparison table: market cap, price-to-sales now and in Edition 1, one-month move, leverage, concentration, and where each name's risk lives.
- The shared fault line, measured: the correlations between the names and the effective number of bets.
- The access decision: picking names inside the bloc versus the Roundhill Neocloud ETF (NCLD).
- Forward triggers, from hyperscaler capex to CoreWeave's interest bill and the November prints, and a numbered source list with every SEC filing URL.
Buy it if
- You already hold or are weighing a neocloud and want the cross-sectional view (which name, at what risk), not the theme view.
- You want each name's risk located on a specific axis rather than averaged into one number.
- You are deciding between picking a name and holding the NCLD wrapper, and want the trade-off spelled out.
Skip it if
- You want a buy or sell call. There are none here, by design and by law.
- You want a price target or a timing view. The report trades the cross-section, not the direction.
- You want figures that never move. Edition 2 is dated 2 November 2026 and refreshed after the November prints; a report older than 60 days is withdrawn from sale until it is refreshed.
How delivery works
Pay through Stripe, and the document arrives by email within seconds as a permanent link tied to your address. Read it in the browser or print it to PDF. Nothing to install, nothing recurring: a one-time purchase. Lost the email? Sign in to your library with the same address and it is there.
Included in Pro. See Pro
An educational research publication, not investment advice, trade signals, a managed account, or an advisory service. Nothing in it is a recommendation to buy or sell any security. See our disclosures.