Neocloud stocks are leveraged, concentrated bets on AI computing demand. This is the checklist: eight risks, the number that measures each one, where it sits in the quarterly report, a dated example from the filings, and how much of a portfolio such a stock can sensibly take.
Market risk reading, cross‑asset context and an events calendar: see Exent Market intel →
That depends on criteria no stock page can answer for you: how much you can lose, for how long you can wait, and what else you own. What can be said is that a neocloud is a company borrowing heavily to buy chips for a few very large customers, and that each of those features is a measurable risk. If the eight numbers below look acceptable to you after reading the latest 10-Q, and the position is small enough that a 60% fall would not change your plans, you have a starting point for evaluating the risk; the checklist does not establish fair value or suitability, and the position could lose all its value. If you cannot check those numbers, it is a bet you cannot check.
Scope: listed GPU clouds (CoreWeave, Nebius, IREN) and the data-center landlords grouped with them, as defined in what is a neocloud. All company figures are from the latest filed quarter (mostly to June 30, 2026; unaudited quarterly figures) or later 8-Ks and releases; Q3 2026 reports are expected in November. Excluded: price targets, ratings, valuation and any view on which stock to own. This is education, not a recommendation.
One row per risk: the number to look at, where it sits in a quarterly report (10-Q for US filers; Nebius files 6-K interim financials, IREN a 10-K and 10-Qs), and a dated example. The examples are illustrations from the filings, not verdicts on the companies.
| Risk | The number to check | Where in the report | Dated example (filing) |
|---|---|---|---|
| 1. Customer concentration | Share of revenue from the largest customers | Revenue or concentration-of-risk note | CoreWeave: Microsoft ~67% of 2025 revenue (10-K); three unnamed customers 36%, 26%, 10% in Q2 2026 (10-Q). Nebius: three customers 24%, 21%, 14% (Q2 2026 interim financials) |
| 2. Debt and refinancing | Interest expense vs operating income; principal due in the next 12–18 months vs cash | Income statement; debt note (maturity table); liquidity section of MD&A | CoreWeave Q2 2026: $640M net interest vs −$49M operating loss; $10.6B due by end-2027; $7.5B of current debt vs $5.5B cash (10-Q). Nebius: $3.5B senior secured facility with a 364-day maturity (Oct 22, 2026) |
| 3. Interest rates | Change in interest cost per 1 percentage point; cost of the newest borrowing | Item 3, market-risk disclosures; debt note | CoreWeave: about $30M of quarterly interest per 1 point on floating-rate debt (10-Q); its 9.75% notes due 2031 traded at a 15.78% yield on Oct 15, 2026 (CNBC) |
| 4. GPU depreciation | Useful life of computing equipment; any change in estimate | Property and equipment note; accounting-policy note | CoreWeave: six years since 2023 (10-K). Nebius: four to five years from Jan 1, 2026, cutting expected 2026 depreciation by ~$167.6M (20-F). Amazon shortened part of its servers to five years in 2025 |
| 5. Power and construction delays | Active vs contracted power; delivery dates vs the original schedule | MD&A business update; 8-Ks on lease amendments | CoreWeave: 1.5 GW active vs ~3.7 GW contracted (Jun 30, 2026). Cipher: two Barber Lake data halls’ rent start moved from Q4 2026 to Q4 2026–Q1 2027 (8-K, Oct 13, 2026) |
| 6. Dilution | Change in share count; convertible notes outstanding; at-the-market programs | Equity note; debt note (converts); subsequent events | CoreWeave: $3.0B of new equity in H1 2026 (10-Q) and $4.2B of 2.875% convertible notes due 2033 (8-K, Sep 22, 2026). Cipher: $850M of 0% convertible notes offered (8-K, Oct 13, 2026) |
| 7. Contract terms | Take-or-pay share; term; “up to” amounts; extension options; prepayments | Remaining performance obligations note; risk factors | CoreWeave: take-or-pay committed contracts of 1–6 years were 98% of Q2 2026 revenue; RPO $103.7B can change with credits and delivery delays (10-Q). Nebius–Meta: “up to” ~$27B, of which up to $15B depends on unsold capacity (Mar 16, 2026) |
| 8. Competition and consolidation | New capacity from customers themselves; pricing per GPU hour; consolidation | Risk factors; management commentary | Microsoft, Alphabet, Amazon and Meta plan about $725–745B of 2026 capital spending, much of it their own data centers (company guidance, Oct 22, 2026). Gartner forecast on Oct 14, 2026 that over 80% of neocloud providers will cease independent operations by 2030 |
Where each risk is explained in depth: customers in who are CoreWeave’s customers, debt and rates in CoreWeave’s debt explained, depreciation in how long AI GPUs last, contract measures in backlog, RPO and ARR, dilution in stock dilution, and the companies side by side in neocloud stocks compared.
About fifteen minutes per report, in this order. The filing is on sec.gov (EDGAR) and on each company’s investor-relations page; what a 10-Q is and how it differs from a 10-K is in 10-K vs 10-Q. Quarterly statements are unaudited; annual 10-K statements are audited.
There is no correct number, but there is a way to choose one: decide the loss you could accept on the whole portfolio, and work backwards using a fall these stocks have already had. Hypothetical illustration, not advice: a $100,000 portfolio, a neocloud slot of 3%, 10% or 25%, and the average fall of CoreWeave (−61.2%), Nebius (−44.6%) and IREN (−67.4%) from their 12-month closing highs to October 23, 2026, which is −57.7% ((61.2 + 44.6 + 67.4) ÷ 3 = 57.73%; Exent market data).
| Neocloud slot | Amount | Loss at −57.7% | Effect on the whole portfolio |
|---|---|---|---|
| 3% | $3,000 | $3,000 × 0.5773 = $1,732 | −1.7% |
| 10% | $10,000 | $10,000 × 0.5773 = $5,773 | −5.8% |
| 25% | $25,000 | $25,000 × 0.5773 = $14,433 | −14.4% |
To recover a 57.7% fall a stock must rise 136.6% (1 ÷ 0.4227 − 1); see loss recovery math. The usual framing for stocks like these is a speculative slot that is small enough to lose without changing the plan, with a written rule for adding or trimming; how to set it is in how big a speculative sleeve should be, position sizing and single-stock concentration limits. A neocloud ETF does not solve this by itself: Roundhill’s NCLD held about 58% in Nebius and CoreWeave at its August 6, 2026 launch (see thematic ETFs).
For the overlap question, see your portfolio if the AI bubble bursts and when high-beta stocks belong in a portfolio.
Counterexample: a risk list counts what can go wrong, not how likely it is. Some of the same filings show the other side. Much of the debt is secured by contracts with investment-grade customers and priced well below the bonds: IREN financed its Microsoft contract with $3.6 billion of investment-grade GPU loans at 6.0%, and CoreWeave’s March 2026 non-recourse GPU loan carried an effective rate near 7% (10-Q). Customers prepay part of the build: 50–60% of the related capex at Nebius and 45–55% of GPU capex at IREN (company figures). Executives at Nebius and IREN told CNBC in October 2026 that customer demand had not changed. If capacity is delivered on time and contracts renew near today’s prices, the model can work with heavy debt.
The checklist is also not a timing tool: a stock can fall without any new number in its filings, as the group did in October 2026 (see why neocloud stocks fell in October 2026), or rise while they worsen. The miners now leasing to AI tenants carry a different mix of these risks; see bitcoin miners’ AI pivot.
Market intel provides market context, cross-asset analysis, an events calendar and regular written notes. These can inform your review of an existing plan; they do not reliably predict crashes or guarantee better outcomes.
It depends on your criteria, not on the stock. They are heavily indebted, depend on a few customers and fell 44.6%–67.4% from their 12-month highs to Oct 23, 2026 (CoreWeave, Nebius, IREN). Check the eight risks on this page and size any position so a 60% fall would not change your plans.
From its filings: customer concentration (top three customers 72% of Q2 2026 revenue), debt (net interest of $640M against a $49M operating loss in Q2 2026; $10.6B of principal due by end-2027) and dilution ($4.2B of new convertible notes in September 2026).
Interest expense vs operating income, the debt maturity table, top-customer percentages, remaining performance obligations, the useful life of GPUs, market-risk disclosures on rates, and new shares or convertibles. Quarterly statements are unaudited.
Operating earnings do not cover interest, so check operating cash flow, capital spending and new financing to see how the gap is paid. In Q2 2026 CoreWeave reported $640M of net interest expense against a $49M operating loss (unaudited 10-Q).
There is no single answer. One way to decide: pick the loss you could accept on the whole portfolio and divide by a fall these stocks have had. At −57.7%, a 3% position costs 1.7% of the portfolio and a 25% position 14.4%.
Less single-company risk, not much less theme risk: Roundhill’s NCLD held about 58% in Nebius and CoreWeave at its August 6, 2026 launch, so two names drove most of the result.
They are also their customers and competitors: Microsoft, Alphabet, Amazon and Meta plan about $725–745B of 2026 capital spending. Gartner forecast on Oct 14, 2026 that over 80% of neoclouds will cease independent operations by 2030, which includes being acquired.
Often. CoreWeave raised $3.0B of new equity in the first half of 2026 and issued $4.2B of convertible notes in September 2026; Cipher offered $850M of convertible notes in October 2026. Convertibles can turn into new shares.
They raise the cost of the debt these companies keep issuing. CoreWeave estimates each 1 percentage point adds about $30M of quarterly interest on floating-rate debt, and new unsecured debt is priced near market yields, which were about 15% for its bonds in October 2026.
Market intel provides a Risk ON / Neutral / Risk OFF reading from four pillars, cross-asset analysis, an events calendar and regular written notes. Context for reviewing a plan, not a crash forecast.
Educational content, not investment advice. Examples are illustrative; past performance does not guarantee future results.