Neoclouds ranked by risk-adjusted value: CoreWeave, Nebius, IREN, Applied Digital, SharonAI

The market trades "neocloud" as one ticker with five names on it. Buy the GPU-cloud theme, the thinking goes, and it barely matters which one - they all rent NVIDIA silicon to AI labs and they all ride the same capex wave. That is half-right. The business model is uniform. The risk is not.

The five names are running the same playbook on completely different balance sheets, customer books, and valuations. $CRWV trades around 7x sales; $SHAZ trades around 860x. One owns its data centers outright; one has not recognized revenue yet. Ranked by risk-adjusted value - what you are paying versus what is actually de-risked - they do not cluster. They spread across the entire curve, from a cheap-but-levered base to a pure lottery ticket.

What a "neocloud" actually is

A neocloud is a GPU data-center operator that rents AI compute as a service - the same shape as AWS or Azure, but pure-play on accelerated compute and without the legacy enterprise stack. They buy NVIDIA GPUs (or lease the power and campus to house them), wire them into clusters, and sell capacity to AI labs and hyperscalers on multi-year contracts. They sit one layer below the model labs and one layer above the chip and power suppliers.

The model is capital-devouring by construction. A cluster is bought before the revenue arrives, so every neocloud is, underneath, a financing story: how cheaply can it fund the build, how locked-in is the demand, and how concentrated is the customer that backstops the debt. That is why these names belong in QA's hyperscalers bubble but trade with far more dispersion than the megacap clouds - the megacaps self-fund; the neoclouds borrow, dilute, or both.

Here is the stack, ranked from the most de-risked value to the most speculative.

CoreWeave (CRWV): the cheapest proven pure-play

$CRWV is the value end of the curve. It trades around 7x sales against Nebius near 62x, on the largest real revenue base in the group and a $99.4B backlog. On the multiple alone it is the cheapest proven name in neoclouds.

The entire bear case is the balance sheet. Debt jumped from roughly $8.2B in 2024 to nearly $30B, and OpenAI plus Meta make up about 55% of the backlog. So the discount is not free - you are paying a low multiple precisely because the leverage and the customer concentration are real. The drawdown, somewhere between 26% and 43% off the highs depending on the window, is either the discount or the warning, and which one it is depends on whether the backlog converts to cash faster than the debt compounds. Momentum is the weak point here: CRWV has been the laggard of the group, not the leader.

Nebius (NBIS): best execution, demanding price

$NBIS is the cleanest operating story and the hardest entry. Q1 printed 684% year-over-year growth, it signed a $27B Meta contract, took a $2B NVIDIA equity investment, and roughly doubled EBITDA margin to 45%. The balance sheet is cleaner than CoreWeave's and the consensus calls it the stronger story. The full company read is in the Nebius explainer.

The problem is entirely the price. As of late June 2026 NBIS had just made an all-time high, up about 210% year-to-date and only 6% off the highs - no pullback, priced for perfection. And the customer book is the tightest in the group: Meta and Microsoft are roughly 80% of it. Best execution does not mean best entry; the risk here is paying the perfection multiple right before the first quarter that is merely good.

IREN: the balanced middle

$IREN is the one that splits the difference. It owns its assets outright, which is a genuine cost moat the leasing-based names do not have, and its legacy Bitcoin cash flow cushions the build instead of forcing dilution to fund it. It has more than 5GW of secured power and, on the sell-side read current in June 2026, Bernstein reaffirmed a Buy with a $100 price target, noting IREN's $10.4M of revenue per megawatt puts it closer to CoreWeave and Nebius than the miner-origin reputation implies.

The framing writes itself: cheaper than NBIS, less levered than CRWV, with real operating assets underneath. The offsets are the miner-origin discount the market still applies and Microsoft at roughly 55% of 2026 revenue - the same single-customer dependence that runs through the whole group.

Applied Digital (APLD): theme beta, not a leader

$APLD sits on the power-and-campus layer rather than the cloud layer - it is closer to a landlord building and powering the buildings than an operator selling the compute inside them. That means it captures less value per megawatt than the names that run the GPUs themselves. It works as broad exposure to the buildout, but it has the weakest differentiation of the established names: you own the theme, not an edge.

SharonAI (SHAZ): the lottery ticket

$SHAZ (SharonAI Holdings) is the speculative tail of the bloc - highest convexity, highest blow-up risk. It trades around 860x sales on roughly $1.6M of actual revenue, revenue recognition does not start until late 2026, and in June 2026 it diluted hard with a $1.6B raise (about $900M equity plus $700M of convertible notes) anchored by Situational Awareness L.P. and Oaktree to fund a six-year NVIDIA collaboration and one of Australia's largest AI factories.

That Situational Awareness anchor is the same Leopold Aschenbrenner fund whose concentrated AI-compute book QA has tracked before (see the Aschenbrenner portfolio). On any fundamental, risk-adjusted basis SHAZ ranks last - it is behind on every axis except upside-if-it-works and short-term momentum. It belongs on this list only as the convex tail of a Compute-bloc position, sized accordingly.

How they stack up

The axes that actually separate them

Three observations fall out of the table:

1. Valuation spread is enormous for one business model. From ~7x sales (CRWV) to ~860x (SHAZ) is not a rounding difference - it is the difference between paying for delivered revenue and paying for a six-year promise. The multiple tells you where on the de-risking curve each name sits.
2. Customer concentration is the shared risk, not the differentiator. CRWV ~55%, NBIS ~80%, IREN ~55% - every name in the group leans on one or two hyperscalers. The neocloud bull case and the neocloud bear case are the same sentence: a handful of AI labs are funding the entire buildout. If one of them trims capex, it hits all five.
3. Asset ownership is the quiet moat. IREN owning its power and campus, and SHAZ/APLD living on the power-and-campus layer, matters more than it looks - in a financing-driven business, the operator that does not have to re-lease or re-borrow every cycle has the most durable margin.

This is also why the bloc is more beta than alpha at the index level. QA's residualization work flags the hyperscalers cluster as largely market beta wearing a thematic costume - so the cross-sectional read (which name, at what risk) matters more here than the directional one (is the theme up).

How to access

All five are US-listed, which is the easy part - unlike the memory makers, where the best names sit on the Korea Exchange (see HBM is the tightest bottleneck in the AI cycle). CRWV, NBIS, IREN, and APLD trade on US exchanges, and SHAZ lists on NASDAQ. A US-retail account can build the entire bloc directly; to trade them from one, see /stack/ibkr. For a single-ticker wrapper of the whole cluster rather than picking names, there is now the Roundhill Neocloud ETF ($NCLD), the first US pure-play on the theme.

Live price, bubble correlation, and bot positioning on each name sit on their /stocks pages. Bubble shifts and rule-based alerts across the Compute bloc are part of /pro.

What to watch

• CRWV backlog-to-cash conversion. The $99.4B backlog versus the ~$30B debt is the whole story - watch whether contracted revenue lands faster than interest compounds.
• A first NBIS pullback. The name has had none. The risk is structural until the multiple resets or the growth rate justifies it; the first merely-good quarter is the test.
• IREN's Microsoft renewal. With Microsoft near 55% of 2026 revenue, the contract cadence is the single most important data point for the "balanced middle" read.
• SHAZ revenue recognition (late 2026). Until revenue actually starts, the ~860x multiple is a promise, not a metric. First recognized revenue is the binary.
• Hyperscaler capex. The shared dependency. If Meta, Microsoft, or OpenAI signals a capex slowdown, it re-rates all five at once, regardless of which one you picked.

Live data on these names: /stocks/CRWV, /stocks/NBIS, /stocks/IREN, /stocks/APLD, /stocks/SHAZ - price, ETF holdings, bubble correlation, bot positions.

Bubble context: /bubbles/hyperscalers - the cluster these names belong to and how it's moving.

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