How to Judge the Order Quality of an AI Cloud Provider: Four Metrics You Can Actually Verify

The four metrics at a glance ​

Why "big orders" does not mean "good business" ​

AI cloud providers share one trait: they all quote very large numbers. CoreWeave cites $132.5 billion of revenue backlog, Nebius cites $9.409 billion of deferred revenue, Oracle cites hundreds of billions in remaining performance obligations. All of these dwarf current revenue and are easily read as "future revenue is locked in."

But order quality depends on four things: whether it has become cash, whether it has become deliverable capacity, whether it is concentrated in a few customers, and whether the revenue it generates covers the capital cost incurred to win it. All four can be checked in public filings without any forecasting.

Metric one: how much of the backlog is estimated ​

Revenue backlog (or RPO) is not defined uniformly, which is the first trap.

CoreWeave's footnote is worth reading word by word: remaining performance obligations, plus other amounts the company estimates will be recognized as revenue in future periods under committed customer contracts, in each case subject to delivery and service availability requirements. That sentence carries three implications:

• It includes amounts the company itself estimates, not pure contracted value;
• It is conditioned on delivery and service availability — meaning if the capacity is not built, that portion does not exist;
• It is cumulative over multiple years, not any single year's revenue.

By contrast, a company reporting only "remaining performance obligations" uses a harder basis, because that maps to contracted obligations under accounting rules. So when you see "revenue backlog," the first step is to read the footnote for whether estimates are included; the second is to check whether the committed-but-unsigned portion is disclosed.

A usable check: divide backlog by trailing four-quarter revenue to get a multiple. If that multiple exceeds 8x and the company also does not disclose customer concentration, discount the credibility of the backlog — a long-duration cumulative figure cannot be assessed for risk concentration without customer structure information.

Metric two: deferred revenue is cash in, prepayment may not be ​

These two terms are routinely conflated, but they sit at different points in the cash chain.

Deferred revenue is a balance sheet item: cash has been received, revenue not yet recognized. It corresponds to cash already in the door. Nebius reported $9.409 billion of deferred revenue at the end of September 2026 ($1.446 billion current plus $7.964 billion non-current) against $1.578 billion at the end of 2025 — an increase of about $7.8 billion in nine months. That figure is hard: it is already on the books.

Prepayment is often called "advance receipts" in casual usage, but when a company says payments are "expected," it is a forward-looking statement. CoreWeave's CEO referenced roughly $9 billion of customer prepayments and deposits expected within the next three months — expected, not booked.

So the order of judgment is: first the sequential change in deferred revenue (hard data), then the forward statement on prepayments (soft data). If a company's growth narrative rests mainly on the latter, its order quality has not yet been validated by the balance sheet.

A supplementary check: deferred revenue as a share of next-year revenue guidance. A high share means customers have already voted with money; a low share means the orders are still on paper.

Metric three: conversion from contracted to active power ​

For AI cloud providers, power data is the physical evidence of order quality. Compute must be energized before it can be delivered, so the conversion rate directly reflects whether orders can become revenue.

Two concepts must be separated:

• Active power: capacity already energized and deliverable;
• Contracted power: capacity signed but not yet energized.

At the end of Q3, CoreWeave had more than 1.7 GW of active power (double the start of the year) against more than 4.6 GW of contracted power, with a year-end target above 1.85 GW of active power. That leaves roughly 2.9 GW to energize over coming quarters.

The gap cuts both ways: positively, it is capacity already locked for the future; negatively, it is an execution commitment that must be met on schedule, and any slip pushes revenue recognition back while interest does not move.

A usable check is the conversion rate: net addition to active power this quarter divided by contracted power at the start of the quarter. If that ratio stays below 15% for several quarters, delivery is lagging signing — orders pile up on the books without becoming revenue.

In CoreWeave's Q3, net active power grew by more than 200 MW to above 1.7 GW at quarter end. The company does not disclose the exact starting active-power figure or the starting contracted balance, so the exact quarterly conversion rate cannot be computed — that must be stated honestly rather than replaced with an estimate. What is certain: against the year-end target above 1.85 GW and 1.7 GW at quarter end, Q4 requires another net gain of roughly 150 MW or more, while contracted power stands above 4.6 GW, a gap of about 2.9 GW.

Metric four: how many times adjusted EBITDA covers interest ​

This is the most overlooked of the four, and the one that most determines survival.

AI cloud is a heavy-asset, heavy-debt model. Judging order quality ultimately comes down to whether the cash gross profit from those orders covers the cost of financing them.

The method: take adjusted EBITDA (non-GAAP, but the company's own cash gross profit measure) and divide by net interest expense for the same period.

• CoreWeave: $2.256 billion adjusted EBITDA / $846 million net interest ≈ 2.7x;
• Nebius: $465.8 million adjusted EBITDA / $112.5 million net interest ≈ 4.1x.

A 2.7x level means that if EBITDA fell 60%, interest coverage would drop below 1x. EBITDA is driven mainly by revenue and operating expense, and revenue depends heavily on delivery progress. That is why "delivery delay" is fatal in this kind of company: it hits the numerator (revenue) while leaving the denominator (interest) untouched.

One further check: compare net interest with full-year adjusted operating income guidance. CoreWeave guides Q4 net interest at $1.05-1.15 billion against full-year adjusted operating income guidance of $1.275-1.375 billion — a single quarter's interest already approaches a full year's operating income. Net income is therefore highly sensitive to rates, and any increase in refinancing cost changes its direction directly.

How to combine the four metrics ​

Any one alone distorts. Combined, they carry judgment. Suggested order:

1. Use metric one to screen out "soft-basis" companies: where backlog contains heavy estimates and customer concentration is undisclosed, lower the credibility weighting immediately;
2. Use metric two to verify whether orders have become cash: rising deferred revenue that represents a high share of next-year guidance means harder orders;
3. Use metric three to check whether capacity keeps up: a conversion rate persistently below 15% means orders are piling up while delivery is constrained;
4. Use metric four to judge financial sustainability: EBITDA covering interest below 2x places the company in a high-leverage risk zone.

Only a combination that passes all four qualifies as genuinely high-quality: deferred revenue growing, capacity being energized on plan, backlog basis free of heavy estimates, and EBITDA covering interest with headroom.

Three red flags ​

Red flag one: order figures growing faster than deferred revenue. New contracts are not yet producing cash inflow; the growth narrative rests on the future rather than the present.

Red flag two: active-power conversion below 15% for two consecutive quarters. Delivery capability has become the bottleneck, and no amount of signing converts into revenue.

Red flag three: EBITDA covering interest below 2x while interest guidance keeps rising. Both signals together mean the company is sustaining the same capacity expansion at a higher cost of capital, and rollover risk is accumulating.

Boundaries and inferences not to draw ​

• All four metrics are based on items verifiable in public filings and involve no undisclosed information or estimation;
• Backlog, ARR and adjusted EBITDA are non-GAAP or company-defined and cannot be compared directly with same-named metrics at peers;
• The "EBITDA interest coverage multiple" is a ratio constructed here for comparability, not an official disclosure by any company, and is not equivalent to the interest coverage ratio (which is typically calculated on EBIT);
• Power definitions are not fully consistent across companies (whether under-construction or leased capacity is included), so definitions must be checked before cross-company comparison;
• This article is not investment advice on any company and forecasts no company's next-period results.

Sources ​

• CoreWeave press release: CoreWeave Reports Strong Third Quarter 2026 Results (November 10, 2026)
• CoreWeave Q3 2026 earnings presentation (PDF)
• Nebius press release: Nebius reports third quarter 2026 financial results (November 12, 2026)
• Related on this site: CoreWeave Q3 2026 key data, CoreWeave vs Nebius, Q3 2026, The AI compute funding loop