Nebius (NBIS): what it does, the $50B Microsoft–Meta backlog, and the AI-utility read

The standard $NBIS story is that the ex-Yandex spinout reinvented itself as an AI cloud, signed roughly $50B of contracts with Microsoft and Meta, and is the next hyperscaler in waiting. That story is half-right.

The half it misses: Nebius is not a software-margin cloud. It is a leveraged data-center utility that rents NVIDIA GPUs by the hour. The backlog is real, but it is compute Nebius has mostly not built yet — and building 4 gigawatts of AI factories is a capital problem before it is a demand problem. This piece walks through what Nebius does, how it makes money, where it sits in the AI-compute stack, and the financing question the momentum tape is ignoring.

What Nebius actually does

Nebius runs an AI cloud: on-demand access to clusters of NVIDIA GPUs, high-performance storage, and the managed software layer that makes a training or inference run actually work. The customer is an AI lab, an enterprise model team, or — increasingly — a hyperscaler that needs more accelerated capacity than it can build itself. They rent Nebius compute instead of buying chips and standing up their own data centers.

The company is the AI-infrastructure remnant of Yandex N.V. After divesting its Russian businesses in 2024, what stayed listed kept the AI-cloud engine plus a cluster of adjacent bets: Toloka (data labeling for model training), Avride (autonomous driving), and TripleTen (technical reskilling). The core Nebius AI Cloud is the thesis; the rest are option value. The platform runs the latest NVIDIA Blackwell and Hopper systems on non-blocking InfiniBand fabric, across a mix of owned facilities and partner data centers.

The category has a name now: the "neocloud." $CRWV (CoreWeave) is the best-known of them. These are not $AMZN, $MSFT, and $GOOGL reselling spare capacity — they are purpose-built GPU landlords whose entire business is buying NVIDIA silicon, wrapping it in power and networking, and renting it back to the AI economy.

How they make money

Nebius makes money two ways: on-demand consumption (you spin up GPUs and pay by the hour) and, increasingly, multi-year reserved capacity contracts that lock customers into dedicated compute. The reserved deals are what changed the company's profile.

The contract spine, as disclosed through Q1 2026:

• Microsoft ($MSFT) — a five-year agreement worth up to $19.4B, with the first tranche of capacity delivered on schedule in November 2025.
• Meta ($META) — a $27B, five-year capacity contract: roughly $12B of dedicated compute plus a $15B option Nebius can exercise at its discretion.
• NVIDIA ($NVDA) — a $2B equity investment in March 2026, plus NVIDIA Exemplar Cloud status on the GB300 platform.
• Total contracted backlog — approaching $50B for the 2027–2031 window.

That backlog drove the headline ARR. Core AI-cloud annualized run-rate revenue reached $1.92B in Q1 2026, up 54% quarter-on-quarter, against Q1 revenue of $399M. Management reaffirmed FY2026 guidance of $3.0–3.4B in revenue and $7–9B in ARR.

Two things to hold in mind on the income statement. First, customer concentration is the headline risk: Microsoft and Meta dominate the backlog, so the revenue visibility everyone is pricing rests on two counterparties. Second, Q1 net income of $621M exceeded revenue of $399M — a tell that it is driven by non-operating gains (equity-stake revaluations), not by the cloud business, which is still in heavy build-out. Pin these to data_as_of 2026-05; the contracts and capacity guidance move quarter to quarter.

Where it sits in AI compute

Nebius lands in QuantAbundancia's AI-Capex Hyperscalers bubble — though the classifier flags it as an extended-tier member (correlation ~0.34 to the hyperscaler centroid), not a core constituent. The cleaner editorial framing is the AI Utility theme: a name whose economics look more like a power-and-real-estate utility than a software platform.

The peer set tells the same story. NBIS correlates most tightly with $CRWV (CoreWeave, ~0.64) — the pure-play neocloud comp — followed by $IREN, $CORZ (Core Scientific), and $CIFR (Cipher Mining), the crypto miners pivoting their power and shells into AI compute, and $ALAB (Astera Labs) on the connectivity side. That cluster is the tell: Nebius trades with the names that turn megawatts into GPU-hours, not with the software hyperscalers it supplies.

Where it sits on the stack: Nebius is downstream of the chip (NVIDIA) and the memory bottleneck, and upstream of the AI lab. Its constraint is not silicon allocation alone — it is power and the capital to build. The contracted-capacity guide was raised to 4 GW, anchored by a 1.2 GW proprietary facility in Philadelphia, a 310 MW AI factory in Lappeenranta, Finland, and a gigawatt-scale site approved in Independence, Missouri. For the broader version of that constraint, see the data-center power bubble — power, not GPUs, is becoming the binding limit for everyone in this bloc.

The numbers

| Metric | Value | As of | | --- | --- | --- | | Market cap | ~$58B | 2026-05-30 | | Q1 2026 revenue | $399M | 2026-05-13 | | Core AI-cloud ARR | $1.92B (+54% QoQ) | 2026-05-13 | | FY2026 revenue guidance | $3.0–3.4B | 2026-05-13 | | FY2026 ARR guidance | $7–9B | 2026-05-13 | | Contracted backlog | ~$50B (2027–2031) | 2026-05 | | Contracted capacity guide | 4 GW | 2026-05 | | NVIDIA equity investment | $2B | 2026-03 |

The shape of the business: revenue is small relative to the backlog because the capacity is still being built. That gap between $399M of quarterly revenue and a ~$50B backlog is the entire investment debate. If Nebius builds and finances 4 GW on schedule, the run-rate catches the backlog and the multiple looks cheap in hindsight. If capital markets tighten or capacity slips, the backlog is a promise it cannot fund. At a ~$58B market cap and a P/E near 67, the tape is pricing the first outcome.

The bull case

• Backlog visibility. Roughly $50B of multi-year contracts gives revenue a floor that consumption-only neoclouds lack, and ARR growth of 54% QoQ shows the reserved capacity is converting.
• NVIDIA alignment. A $2B equity check, GB300 Exemplar status, and early Blackwell allocation mean Nebius sits near the front of the GPU queue — the scarcest input in the cycle.
• Vertical integration. Owning facilities and the software stack (not just reselling capacity) is a margin and reliability edge over asset-light resellers.
• Anchor customers. Microsoft and Meta signing multi-year deals is third-party validation that the capacity is wanted at scale.
• Optionality. Toloka, Avride, and TripleTen are non-core but real, and the retained equity stakes have already shown up as balance-sheet gains.

The bear case

• Customer concentration. Two counterparties — Microsoft and Meta — dominate the backlog. If either reprices, delays, or insources, the revenue floor cracks.
• Capital intensity. Building 4 GW of AI factories is one of the most capital-hungry projects in tech. The financing is the business, and it competes for the same capital and power as every hyperscaler.
• Dilution risk. A neocloud that must out-build its own backlog typically funds the gap with equity and debt. The share count is a moving target, and momentum holders rarely price dilution until it arrives.
• Commoditization. Renting GPU-hours is a thin-margin utility. As Blackwell capacity floods online across the neocloud cohort, pricing power erodes — the same dynamic that compresses any commodity buildout.
• The earnings optics. Net income above revenue flatters the picture. Strip the non-operating gains and the operating business is still pre-profit and capex-heavy.

How to access

Nebius lists directly on NASDAQ as $NBIS, so US retail can hold the stock outright — no ADR workaround, unlike some foreign-domiciled AI names. To trade it from a US-retail or international account, see /stack/ibkr.

ETF exposure is thin. Because Nebius only relisted in mid-2025 and remains a relatively young index constituent, it does not yet show up in the broad-ETF holdings we track — there is no clean "buy SOXX, get Nebius" route the way there is for the megacap names. For now, exposure is mostly direct, or through actively managed AI-infrastructure and thematic funds that have added the neocloud cohort. Track which ETFs pick it up over time on /stocks/nbis.

Bubble shifts and rule-based alerts on $NBIS — when its correlation to the neocloud cluster breaks, or when it crosses an editorial level — are part of /pro.

What to watch

• Capacity delivery. The 1.2 GW Philadelphia facility, the Finland AI factory, and the Missouri gigawatt-scale site are the milestones that turn backlog into ARR. Slippage is the first crack.
• Financing announcements. Watch for the equity and debt raises that fund the buildout — size and dilution terms matter more than any single contract.
• Customer count. A third or fourth anchor beyond Microsoft and Meta would de-risk the concentration; none would confirm it.
• The fib_key at $113. With the stock near $230, $113 is the next observable structural level on the QA chart — data, not a target.
• Bubble correlation. If $NBIS breaks correlation with $CRWV and the neocloud cohort, the "AI utility" read changes — the names that rise and fall together are the cleanest signal of what the market thinks Nebius actually is.

---

Live data on this ticker: /stocks/nbis — price, ETF holdings, bubble correlation, bot positions.

Bubble context: /bubbles/hyperscalers — the cluster this name belongs to and how it's moving.

QuantAbundancia is educational research. Nothing here is investment advice. See /disclosures.