tkalAI Capital Flows
Earnings deep dive · Synthesis edition · AI Cloud & Infrastructure

NBIS Q2 2026 — $2.25B of Cash Flow, and Almost None of It Was Nebius's

Revenue grew 454% to $582.3M and adjusted EBITDA turned positive for the first time at $236.2M on a 41% margin — a genuine milestone, and not the number that should move you. The number that should is the one the market cheered: $2.246B of operating cash flow. Decode it and the applause fades. Roughly $1.2B was a jump in deferred revenue and another ~$1.19B was collecting old receivables, so essentially none of that cash came from the business — it came from customers pre-funding the build and from banking prior bills. Nebius is not yet a company that funds itself; it is a company whose customers and creditors fund it, and the cash-flow statement dressed that up beautifully. Then the market paid roughly 22x forward sales for it, versus about 5x for CoreWeave. My read: the EBITDA inflection is real, and it is not the risk. The risk is that prepaid growth is a customer loan in a cash-flow costume, and you were just asked to pay a premium multiple for the costume.
Published · Post-print synthesis (reported before open ) · NBIS · Committed analyst read — not advice · For analysts covering AI cloud, neoclouds, hyperscaler capex, and the data-center credit stack

The only question that matters

Nebius came into this print as the market's favorite small neocloud — the pure-play the buy-side had already re-rated into a levered call on GPU demand. Shares closed at $193.23 on into the before-open release, having round-tripped from the low-$180s to above $215 and back over the prior week, with 30-day IV rank around the 42nd percentile and a beta near 3.6 (unverified — options and market data, not filing figures). Read the setup honestly and the demand question was already closed: revenue compounding past 400%, capacity contracted years forward, a multi-billion exit run-rate guided. So the real questions were further down the statements, and only one of them mattered. Not "can it inflect to positive EBITDA" — that was likely and it happened. The one that mattered: with a $20–25B capex year in front of it, where does the money come from, and how much of the reported cash generation is the business versus customers and financiers pre-funding the build? This is the gap between the narrative and what the data supports, and on Nebius it sits in the cash-flow statement. That is where I'm spending this piece.

Read the EBITDA and the cash flow side by side

Take the good news at face value first, because it is real. Group revenue landed at $582.3M, up 454% year-over-year from $105.1M and ~46% sequentially (first-half revenue of $981.3M implies a ~$399M Q1). The Nebius AI cloud core did about $575M — ~98% of the group, with Avride and TripleTen small consolidated remainders and ClickHouse and Toloka carried off the income statement. ARR reached roughly $3.0B at end-June, up from ~$1.9B in March (unverified — a company operating metric from the shareholder letter, not a GAAP line). And the margin inflected: adjusted EBITDA of $236.2M at a 41% margin versus a $21.0M loss a year ago, as cost of revenue fell to 23% of sales (from 29%) and depreciation to 45% (from 72%) on rising utilization. For a neocloud, that is a structurally important data point — the unit economics of renting GPUs at scale can clear their own operating cost. Genuinely good. Also not the argument.

Now decode the number everyone quoted. Nebius reported $2.246B of operating cash flow — arresting for a company that simultaneously booked a $190.4M net loss. Sit with that gap for a second, because it is the whole tell. Where did $2.25B of operating cash come from if the P&L lost money? The statement answers plainly: deferred revenue contributed +$1.197B and receivables collection contributed +$1.187B (receivables fell from $720.3M to $288.6M). Add those and you have essentially the entire operating inflow — meaning it is working capital, not margin. Translate what that actually is: a concentrated set of very large customers wrote Nebius large checks in advance, and Nebius banked bills it had already sent. That is real cash and it is a position of strength — customers prepay because they need the compute. But it is not the business generating cash, and it does not repeat by default; a quarter without a fresh mega-prepayment produces a very different statement. The cash-flow line the headline celebrated is, mechanically, a measure of how much the customers chose to pre-fund this quarter.

And it went straight into the ground. Capex was $5.657B in Q2 ($8.13B first-half) — it dwarfs even that flattering operating inflow. The gap was bridged by financing: for the half, $4.34B of convertible notes, $2.0B of pre-funded warrants, and a $2.85B treasury-share sale. So follow the full chain: customers prepay, creditors and equity holders fund the rest, and the build gets paid for by everyone except the income statement.

Below adjusted EBITDA: where the $236.2M profit goes Nebius Q2 2026 · adjusted EBITDA vs. the costs beneath it · US$ millions Adjusted EBITDA +$236.2M 41% margin BELOW THE LINE — THE DRAGS Depreciation & amort. $259.7M Interest expense $119.1M from $4.8M a year ago Stock-based comp $102.5M up ~600% WHERE IT LANDS GAAP net loss $(190.4)M $(0.68)/sh Adjusted net loss $(33.2)M narrowed 64%
The three drags below adjusted EBITDA total $481.3M — more than double the $236.2M of adjusted EBITDA — which is why a 41%-margin quarter still prints a GAAP loss. Adjusted EBITDA excludes D&A, interest and stock-based comp; the GAAP net loss also reflects other income and tax, so this shows the major drags, not a full line-by-line reconciliation.

The chart makes the GAAP bridge plain: the costs stacked below adjusted EBITDA outweigh it more than two to one. That interest line is the one to watch: the carrying cost is senior to the shareholder, and it went from a rounding error to $119M in a single year while the company is still early in its debt curve.

Paid subscribers

The rest of this dive is for paid subscribers.

The headline numbers are above. The 8 sections below carry the mechanism, the peer read, the valuation work, and the dated tests that decide it.

  • The deferred-revenue balance is the engine — and the dependency
  • The guide tells you the payoff is a 2027 event
  • How the rest of the group should read this
  • Valuation: you paid a premium for prepaid growth
  • The strongest case against me
  • What decides it from here
  • The trade (analysis, not advice)
  • Bottom line
Unlock the full dive →

Already a subscriber? Sign in.

Subscribe · free weekly

This is one print. The pipeline runs every week.

Every Monday, the single sharpest call from the pipeline — a dated, falsifiable read on credit, demand, and capex, then graded in public on the scorecard. Free.

One email to confirm. Unsubscribe any time.

Sources