One dated table per question: what each neocloud and AI data-center landlord reported last quarter, what it has under contract by its own definition, what it guides to, and how violently the stocks have moved. Criteria, not picks.
Market risk reading, cross‑asset context and an events calendar: see Exent Market intel →
They compare as three GPU clouds of very different size: CoreWeave is the largest by revenue and by debt, Nebius grew fastest and holds the most cash, and IREN is still moving from bitcoin mining to AI cloud, with its first quarter in which AI revenue passed mining. None of that says which stock is better; it says which risks each one carries. The plain-language background is in what is a neocloud.
Scope: US-listed companies that rent AI computing (GPU clouds) or lease powered data-center space to AI customers (landlords), using each company’s latest quarterly filing as of October 26, 2026. Excluded: private neoclouds (Lambda, Crusoe, Fluidstack), hyperscalers such as Oracle and Microsoft, and chip suppliers. Every table separates three things that headlines mix up: what was reported (audited or filed numbers), what is contracted (the company’s own measure of signed commitments) and what is guided (management’s plan). Interpretation is labelled as such. This is education, not a recommendation.
| Model | What it sells | Who carries GPU risk | Companies here |
|---|---|---|---|
| GPU cloud | Computing time on GPUs it owns, under 1–6 year contracts or on demand | The neocloud (it buys and depreciates the chips) | CoreWeave, Nebius, IREN |
| Landlord (colocation) | Powered, cooled buildings under 10–20 year leases | The tenant (often a GPU cloud or a hyperscaler) | Applied Digital, TeraWulf, Hut 8, Cipher, Galaxy, Core Scientific |
All of them grew revenue quickly or are about to, and all of them lost money on a GAAP basis in the latest quarter. Definitions, the same for every row: revenue is GAAP revenue for the quarter; net loss is GAAP net income (loss) attributable to the company (continuing operations where the filing splits them); cash is cash and cash equivalents, excluding restricted cash; debt is the carrying value of all debt, notes and loans on the balance sheet, current plus non-current, excluding lease liabilities.
| GPU clouds | Quarter ended | Revenue | vs a year earlier | Net loss | Cash | Debt |
|---|---|---|---|---|---|---|
| CoreWeave (CRWV) | Jun 30, 2026 | $2,575M | +112% | −$626M | $5,524M | $35,068M |
| Nebius (NBIS) | Jun 30, 2026 | $582.3M | +454% | −$190.4M | $8,042M | $8,546M |
| IREN | Jun 30, 2026 (fiscal Q4) | $137.2M (AI cloud $70.5M, bitcoin $66.7M) | not compared here | −$684.0M (incl. $450.4M impairments) | $5,896M | $7,593M |
| GPU clouds | CoreWeave (CRWV) | Nebius (NBIS) | IREN |
|---|---|---|---|
| Quarter ended | Jun 30, 2026 | Jun 30, 2026 | Jun 30, 2026 (fiscal Q4) |
| Revenue | $2,575M | $582.3M | $137.2M (AI cloud $70.5M, bitcoin $66.7M) |
| vs a year earlier | +112% | +454% | not compared here |
| Net loss | −$626M | −$190.4M | −$684.0M (incl. $450.4M impairments) |
| Cash | $5,524M | $8,042M | $5,896M |
| Debt | $35,068M | $8,546M | $7,593M |
| Landlords | Quarter ended | Revenue | vs a year earlier | Net loss | Cash | Debt |
|---|---|---|---|---|---|---|
| Applied Digital (APLD) | Aug 31, 2026 | $341.9M | +322% | −$221.0M | $2,950M | $6,377M |
| Core Scientific (CORZ) | Jun 30, 2026 | $164.2M | +109% | −$1,155.3M (mainly warrant fair value) | $1,770M | $4,298M |
| Hut 8 (HUT) | Jun 30, 2026 | $74.9M | +81% | −$150.2M | $234M (plus $6,787M restricted) | $7,638M* |
| TeraWulf (WULF) | Jun 30, 2026 | $44.8M | −6% | −$939.9M | $2,619M | $5,213M |
| Cipher Digital (CIFR) | Jun 30, 2026 | $24.8M | −43% | −$267.5M (incl. $150.5M warrant loss) | $832M | $5,541M |
*Hut 8 reports one line, “loans, notes payable and other financial liabilities”, so its figure may include items other than debt. Galaxy is left out of this table because its consolidated revenue is mostly gross digital-asset sales; its data-center leasing revenue was $18.9M in the quarter. Large GAAP losses at several landlords come from non-cash changes in the value of warrants, not from operations; read the line items before comparing losses. Stock pages: CRWV, NBIS.
Each company reports a different measure of signed business, so the honest comparison is one row per company with the measure named exactly as filed. Adding them up, or ranking them, would compare unlike things. How these measures differ is explained in backlog, RPO and ARR.
| Company | Measure, as the company names it | Value | As of | What it includes or leaves out |
|---|---|---|---|---|
| CoreWeave | Revenue backlog | ~$104B | Jun 30, 2026 | RPO plus other amounts expected under committed contracts, subject to delivery; excludes >$25B signed in early Q3 |
| Nebius | Customer commitments; ARR | >$40B; $3.0B ARR | Aug 12, 2026; Jun 30, 2026 | ARR = last month’s AI cloud revenue × 12 |
| IREN | Contracted ARR; operating ARR | $4B; $1B | Aug 27, 2026 | Annualized run-rate for 2026 capacity vs capacity running today |
| Applied Digital | Contracted revenue over initial base terms | ~$36B | Aug 31, 2026 | ~1.41 GW of critical IT load leased across 5 campuses |
| Hut 8 | Expected aggregate base-term contract value | ~$26.6B | Aug 4, 2026 | 949 MW of contracted IT capacity; excludes renewal options |
| Core Scientific | Potential contracted revenue | >$24B | Jul 28, 2026 | ~1.1 GW of leased customer power capacity; “potential” is the company’s word |
| Cipher Digital | Contracted revenue | $16.6B | Sep 30, 2026 | Capacity stated in gross MW, not critical IT |
| TeraWulf | Contracted revenue, Anthropic lease | ~$19B (up to ~$33B with extensions) | Aug 5, 2026 | One lease, ~401 MW critical IT, 20 years; no portfolio total given |
| Galaxy | Anticipated average annual revenue, CoreWeave at Helios | $1.2B+ a year | Jun 30, 2026 | 526 MW critical IT (800 MW gross), 15-year base term |
Running capacity is a small fraction of contracted capacity everywhere, which is why execution risk matters as much as demand. The first column is reported; the second is the company’s target, which can slip. CoreWeave gives its financial guidance on the earnings call rather than in the release, so it is not repeated here.
| Company | Running today (reported) | Contracted (reported) or target (guidance) |
|---|---|---|
| CoreWeave | 1.5 GW active power (Jun 30, 2026) | ~3.7 GW total contracted power (reported, Jun 30, 2026) |
| Nebius | Not stated as a single figure | Target: 5 GW contracted power by year-end 2026; more than 1 GW deployed a year from 2027 |
| IREN | $1B operating ARR | Target: ~0.3 GW (IT) delivered cumulatively in 2026, ~0.8 GW in 2027 |
| TeraWulf | 102 MW revenue-generating (critical IT, early July 2026) | 336 MW under construction (reported) |
| Core Scientific | Billing 437 MW (mid-July 2026) | ~1.1 GW leased (reported) |
Start with two ratios you can compute from the tables above. Ratio 1, interest burden: interest expense ÷ revenue. CoreWeave’s Q2 2026 interest expense, net, was $640M on revenue of $2,575M: 640 ÷ 2,575 = 24.9%. Its operating result was a $49M loss, so in that quarter operating income did not cover interest; the gap was funded from the balance sheet and new financing. Ratio 2, debt ÷ annualized revenue (latest quarter × 4):
| GPU cloud | Debt | Quarter revenue × 4 | Debt ÷ annualized revenue |
|---|---|---|---|
| CoreWeave | $35,068M | $10,300M | 3.4× |
| Nebius | $8,546M | $2,329M | 3.7× |
| IREN | $7,593M | $549M | 13.8× |
Interpretation (ours): a high ratio is not a verdict. These companies borrow before the capacity earns revenue, so the ratio falls if contracted capacity is delivered on time and rises if it slips. IREN’s ratio is high because its Microsoft capacity was only starting to be delivered in August 2026. Repeat the arithmetic after each quarter. Bond prices and credit-default swaps add a market view of the same question, but they are not a bankruptcy probability; how CoreWeave’s debt is built is in CoreWeave’s debt explained.
Far more volatile than the market: every name with a year of data swung at roughly eight to nine times the S&P 500’s volatility (7.8x for Galaxy, 9.2x for Cipher). Exent market data, year to September 30, 2026, before the October selloff:
| Stock | 1-year total return | Largest fall in daily closes, year to Sep 30 | Annualized volatility |
|---|---|---|---|
| CoreWeave (CRWV) | −36.3% | −57.5% | 95.7% |
| Nebius (NBIS) | +110.1% | −48.3% | 99.2% |
| Applied Digital (APLD) | +6.2% | −53.2% | 101.1% |
| Hut 8 (HUT) | +147.3% | −41.7% | 101.2% |
| TeraWulf (WULF) | +29.6% | −50.0% | 100.2% |
| Cipher Digital (CIFR) | +25.7% | −51.4% | 109.5% |
| Galaxy (GLXY) | −32.7% | −60.7% | 92.3% |
| S&P 500 (SPY) | +15.7% | −9.1% | 11.9% |
Past returns are history, not a forecast. With swings this size, position size decides the outcome more than stock choice; see how big a speculative sleeve should be and high-beta stocks.
There is no single winner; each company leads on a different measure, and each lead comes with its own risk. Latest quarter unless stated:
| Criterion | Leader in these filings | The number | The catch |
|---|---|---|---|
| Largest revenue | CoreWeave | $2,575M in Q2 2026 | Also the largest debt ($35.1B) and a 24.9% interest burden |
| Fastest revenue growth (GPU clouds) | Nebius | +454% year on year | Growth from a small base; capex of $5.66B in the quarter |
| Most cash | Nebius | $8.04B | Debt of $8.55B is larger |
| Investment-grade, contract-backed GPU loan | IREN | $3.6B at 6.0% for the Microsoft contract | Tied to one customer’s contract and to delivery on time |
| Longest lease terms | Landlords | 15–20 year base terms | Tenants are often other neoclouds; renewal options are not revenue |
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.
CoreWeave is bigger by revenue: $2,575M in Q2 2026 against $582.3M for Nebius. Nebius grew faster (+454% vs +112% year on year) and held more cash ($8.04B vs $5.52B) at Jun 30, 2026.
Nebius. At Jun 30, 2026 its balance sheet carried $8.55B of debt against $35.07B for CoreWeave (carrying value, leases excluded). Relative to annualized Q2 revenue the two are closer: 3.7x for Nebius and 3.4x for CoreWeave.
IREN is smaller and earlier in the shift: Q4 FY26 revenue (to Jun 30, 2026) was $137.2M, of which $70.5M was AI cloud, the first quarter AI cloud passed bitcoin mining. It reported $4B of contracted ARR for 2026 capacity and $1B operating.
Applied Digital mainly leases powered buildings to tenants such as CoreWeave (about $36B contracted over initial base terms), while IREN owns GPUs and sells computing time. As landlord, Applied Digital does not carry the GPU obsolescence risk; IREN does.
Not in the strict sense. Both mainly lease data-center capacity: TeraWulf’s largest lease is about 401 MW to Anthropic for 20 years (about $19B initial term), and Applied Digital’s tenants include CoreWeave and unnamed investment-grade hyperscalers.
Yes. Roundhill launched the Neocloud ETF (NCLD) on Aug 6, 2026, with Nebius and CoreWeave at about 58% of the fund at launch, so it is a concentrated thematic fund rather than a diversified one.
They combine fast growth, heavy borrowing and concentrated customers. In Exent market data for the year to Sep 30, 2026, every listed name had annualized volatility above 90%, against 11.9% for SPY.
There is no best stock without criteria. On the latest filings CoreWeave leads on revenue, Nebius on growth and cash, and IREN stands out for a $3.6B investment-grade GPU loan at 6.0% tied to its Microsoft contract; each lead comes with a matching risk.
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.