Same Land Grab, Different Structures: Scoring the Small-Cap Neocloud Blitz

Below CoreWeave and Nebius, a second tier of AI-infrastructure blitzers is fighting the same land grab — power, campuses, GPUs — with different capital structures and very different scores. Seven companies, one scoreboard, and an honest explanation of why none of them clears the 65-point threshold.

5 min read · 7-company scoreboard · financial structure · Not investment advice

The scoreboard nobody expected

Score all seven AI-infrastructure blitzers on the same frozen v1.0 model, and a pattern emerges that no one asked for but everyone should see:

All seven sit below the 65-point success threshold. That fact is itself a finding. The field shares the same land grab — power secured, campuses planned, GPUs financed — but the scoring sorts them by a single axis: how much of the blitz has actually arrived in the profit and loss statement. Cloud operators that recognize real AI revenue sit on top; landlords that recognize some sit in the middle; contract-first companies whose revenue is still a promise sit at the bottom.

Why nobody clears 65

The Blitz Index v1.0 is deliberately conservative about one thing: it scores recognized revenue, not contracted revenue. Growth (A1) and gross margin (B4) are measured from filed numbers. That discipline bites hard in this field, because the defining feature of the small-cap neocloud is the gap between what's sold and what's delivered.

Cipher Digital is the extreme case: trailing revenue of roughly $191M — all of it legacy bitcoin mining, and declining — against ~$11.4B of contracted lease revenue. A sixty-fold gap. The contracts are real (Barber Lake's $9B+, 20-year lease was filed with the SEC in September 2026; Google's backstop of the Fluidstack lease is in the 10-Q), but the P&L shows none of it yet. v1.0's rule is a conscious choice: in a field where everyone is selling the future, only the realized present counts. That choice is what puts CIFR at 32/D — the model refuses to score the sold future as if it were the present.

The same logic runs through the rest of the deductions. Debt (D1) is measured against recognized revenue, and every one of these companies borrowed against the contracted future: CIFR's debt is ~29x its trailing revenue, WULF's ~24x. The leverage looks apocalyptic precisely because the denominator hasn't caught up with the numerator. Whether that turns out to be brilliance or folly is a question for the P&L of 2028 — which is why none of these companies can clear 65 today, no matter how impressive the backlog.

IREN vs APLD: the cloud operator and the landlord

The most instructive comparison in the field is the top pair. IREN (56) and APLD (51) look similar at a glance — both are blitzing AI infrastructure, both are leveraged, both burn cash. But they are different businesses wearing the same costume. IREN is a cloud operator: it runs GPU capacity itself and sells compute, with contracted ARR above $4B against operating ARR of roughly $1B and FY2027 capex guidance of $25–30B. APLD is the landlord: it builds AI data centers and leases them — 1.41GW contracted ($36B base term, ~$86B with renewals), but only 175MW operating, about 12% of what's contracted. The "+322% quarter" that headlines its latest results was 53.7% tenant fit-out — one-off construction pass-throughs, not recurring rent. The recurring engine (base rent $65.8M/quarter) is still small.

Three structural differences matter. First, interest coverage: APLD's quarterly interest ($77.4M) exceeds its adjusted EBITDA ($64.4M) — the sharpest leverage in the field, coverage of 0.83x. IREN's operating cash flow is positive. Second, whose money builds the future: IREN funds part of its build with customer prepayments; APLD's deferred revenue is only $39M against $2.87B of FY2026 capex — it borrows first and collects later. Third, the debt itself: IREN's is mostly low-coupon convertibles plus non-recourse GPU financing tied to Microsoft contract cash flows (the parent company's exposure is limited); the $6.75B of convertibles defers dilution risk into the future — new notes convert at $51–$86, well above the current price, with capped calls cushioning dilution up to roughly $82–$110 (company filings), so the dilution is real but long-dated and contingent on the stock rising before the 2029–2033 maturities; APLD's is high-coupon secured notes with a hard maturity wall around 2031. Both reach their refinancing test around the same year — by which time the AI revenue must be real.

The five-point score gap (56 vs 51) tracks exactly this: IREN recognizes a faster AI revenue ramp and runs with gentler leverage; APLD is at an earlier delivery stage with harsher financing. This is the same "same industry, different leverage" lens we applied to CoreWeave vs IREN — now within the second tier.

Core Scientific: the second blitz after bankruptcy

Core Scientific (52/C) is the field's most honest test case, because it has already failed once. The company filed for Chapter 11 in December 2022 and emerged in January 2024 — roughly $400M of debt wiped out, existing shareholders diluted about 40%. Blitz Index v1.0 has no memory of past failures except through present numbers, so we made a deliberate judgment: the bankruptcy history flows into the D1 (debt) deduction as demonstrated leverage failure. A capital structure that has already collapsed once deserves a heavier discount than one that hasn't been tested. That's how the bankruptcy history is reflected in CORZ's D1 score of −9.

But honesty cuts both ways. In October 2025, CoreWeave offered roughly $9B to acquire CORZ, and shareholders rejected it — a 10-to-1 vote against a board that had recommended acceptance. Shareholder oversight functioned exactly when it mattered. So the governance deduction (D3) scores zero: no penalty for a governance system that provably worked. The present picture is mixed but real — colocation is now 83% of revenue, though 83% of that comes from a single customer (CoreWeave), and total debt of $7.68B still exceeds assets of $5.26B. The score of 52 says: execution is genuine, leverage is inherited, and the customer concentration is the swing variable.

WULF vs CIFR: where recognized revenue decides

External observers often rank Cipher Digital above TeraWulf — CIFR's contracts are bigger ($11.4B vs ~$19B for WULF's Anthropic deal, comparable), and Google's backstop of its Fluidstack lease looks like validation. Our model reverses them: WULF 41 vs CIFR 32. The reason is the recognized-revenue rule described above, and it's worth stating plainly because the disagreement is methodological, not factual.

WULF already recognizes HPC lease revenue — $31.9M in the quarter, up 52% quarter-over-quarter — and its Anthropic contract has the anchor tenant's name directly on the lease (per the 8-K, not through an intermediary). CIFR's first delivery (Black Pearl, two months early in August 2026) is genuine execution, but no HPC revenue has appeared in any filed statement yet; the P&L still shows only declining mining. The model scores what has arrived. When CIFR's deliveries start hitting the P&L, the score will move — that's the mechanism working as designed, not a verdict on the company's future.

Conclusion: watch the delivery, not the deal

The small-cap neocloud field is one business in two costumes. The landlords (APLD, CORZ, CIFR) lease power and buildings; the operators (IREN, NBIS, CRWV) sell compute. The scoring gap between them tracks how much of each blitz has arrived in the P&L — and every one of these companies is still below the threshold, because every one of them is still building ahead of revenue. That is not a condemnation; it is the definition of the blitz phase. The honest question for each is the same: does the delivered future arrive fast enough to service the borrowed present?

Scores from the Blitz Index v1.0 (frozen October 2026). Company figures from SEC filings and company disclosures, press-reported where labeled. Companion piece in this series: "Same Industry, Different Leverage: CoreWeave vs IREN." This is an assessment against a scholar's standard — not investment advice. No price targets, no buy, sell, or hold recommendations.