Nebius Group: contracted demand meets a $25B capex bill — the 20% dip is a watch, not a buy

Nebius Group: contracted demand meets a $25B capex bill — the 20% dip is a watch, not a buy

Last week Nebius Group priced $5 billion of convertible notes, upsized the deal from its original $4.5 billion, and closed it Monday at roughly $5.75 billion in gross proceeds. The stock fell through the offering: down just over a fifth in five sessions to near $211, about 30% below the $300 peak it touched in early July. The obvious read is dilution fear hitting a crowded, heavily shorted AI growth name — and the obvious read is only half the story. The quarter Nebius reported a week before the raise, a report that had sent the stock up in response, did not look like the work of a company in trouble.

June-quarter revenue was $582.3 million, up 454% from a year earlier and ahead of the roughly $570 million consensus, with the AI-cloud business that drives the company up 514%. Adjusted EBITDA was positive again at $236.2 million, roughly a 40% margin, full-year guidance held at $3.0–3.4 billion, and management reaffirmed an exit annualized run-rate of $7–9 billion by year-end 2026. This is not an unfunded AI narrative: Microsoft agreed in September 2025 to a multi-year GPU-infrastructure deal reported near $17.4 billion over five years, and Meta signed on in March for up to $27 billion over five years. On the balance sheet the commitments surface as $5.98 billion of deferred revenue — customers prepaying for capacity they will use later.

The raise exists because of the number above revenue. Nebius plans to spend $20–25 billion of capital in 2026 against revenue of $3.0–3.4 billion. It is building and filling GPU-packed data centers before the contracted capacity becomes revenue, and the quarterly comparison shows it: $5.66 billion of capex in Q2 against $2.25 billion of operating cash flow, the latter flattered by about $1.2 billion of new deferred revenue and a similar swing in receivables. Strip out customer prepayments and the build-out runs on external capital. That is the neocloud model — the tenants prepay and sign five-year deals, while the landlord fronts the construction and borrows to cover it. March brought a $3.75 billion convertible offering; August brought the $5 billion deal, split between $3 billion of 0.50% notes due 2030 and $2 billion of 4.50% notes due 2034.

The terms tell you what management believes. The notes convert at $313–325, a 40–45% premium above the roughly $224 reference price — meaning lenders convert only if the stock still compounds to about where it peaked this summer. Nebius chose near-zero-cost debt over selling stock into a weak tape, and it paid the lenders with cheap call options on its own shares. The cost arrives later: conversion overhang for existing shareholders if the stock rallies, and, in the 4.5% tranche, real interest on top of a GAAP result that swung to a $190.4 million loss last quarter as depreciation (up 245%) and stock-based compensation (up 597%) ramped behind the expanding fleet.

At $211, the market cap is roughly $57 billion — about 42x trailing revenue and roughly 17x even the 2026 guidance number, with free cash flow sharply negative by design. The closest public comparable, CoreWeave, trades near 6x trailing sales despite having recognized roughly $7.6 billion of revenue over the last twelve months. Part of the spread is deserved: Nebius is compounding revenue at a 450%+ rate and carries a roughly 74% reported gross margin. The rest is the price of a hockey stick. Wall Street, for what it is worth, still stands above the market: Goldman Sachs carries a $286 Buy target set in early July, Citi sits at $287, and this week DA Davidson maintained its Hold at $250.

Judged by the standard test of a beaten-down stock — did the business deteriorate faster than the valuation reset? — Nebius passes. Revenue beat, adjusted EBITDA stayed positive, guidance held, and the selloff is financing mechanics rather than weakening demand. The reset from $300 to $211 has outrun any change in the operating story. That is the setup in which a selloff can be overdone.

But it is not yet a bargain, and the reason is structural. A business spending $20–25 billion of capital to generate $3–4 billion of revenue has one unforgiving failure mode: if Meta, Microsoft, or the next tenant reprice, cut, or build capacity in-house, the depreciation and interest bills do not shrink with the revenue. Nor is the financing treadmill finished — another raise on worse terms would land on top of the conversion overhang the noteholders already hold. The honest stance is watch, not buy, with the proof sequenced across the next two quarters: whether the Q3 revenue ramp is tracking toward the $7–9 billion exit run-rate, what the formal 2027 guidance expected later this year implies for cash conversion, whether each capex dollar starts producing more revenue, and whether a third mega-raise appears. If the ramp holds at a capex pace that is no longer accelerating, the ~17x forward multiple starts to look like a landlord with a contracted tenant list; if the build-out slips or the tenants blink, the current price has already paid for success that has not happened yet.