Deck
Nebius Group · NBIS · NASDAQ
Netherlands-based AI infrastructure company renting GPU compute to hyperscalers and AI labs under multi-year reserved-capacity contracts, funding the buildout with customer prepayments, $4.3B of convertibles, and a $2B equity investment from NVIDIA.
$207
Price
$52.6B
Market cap
$399M
Q1'26 revenue (+684% YoY)
$1.92B
ARR, end Q1'26
Listed 2011 as Yandex; halted February 2022 after Russia invaded Ukraine; resumed October 2024 as NBIS at $17 after divesting Russian assets, then compounded to $207 — roughly 12× in nineteen months.
2 · The tension
Every bull number compounds and every bear number explodes off the same fact
- The single variable. Roughly 95% of the $46B contracted backlog sits with Microsoft (~$17B) and Meta (~$27B). The bull reads that as counterparty validation; the bear reads it as a put written on two customers each running $65B+ of annual AI capex on owned capacity.
- The trigger date. July 29, 2026 — end-Q2 ARR. Above $2.5B (around 30% QoQ) is consistent with the $7–9B exit-2026 guide; below $2.2B (under 15% QoQ) impairs the case and forces a multiple reset.
- The asymmetry. Net cash $0.9B going into a $20–25B capex year (raised from $16–20B on May 13, 2026), but ~$15B of headline backlog is flex capacity Nebius must resell rather than take-or-pay, and the Q1 45% segment EBITDA contains roughly 10pp of useful-life accounting relief.
One quarterly print sets the tape for the next twelve months.
3 · Variant perception
Three load-bearing inputs the bulls quote are quietly weaker than printed
- $46B backlog is closer to $29–31B firm. Per the Q1 2026 letter, ~$15B of Meta's $27B is capacity Nebius can resell at market rates — not take-or-pay. That cuts the headline 87× forward-revenue visibility to roughly 55–60×.
- 45% Q1 segment EBITDA contains ~10pp of accounting. The same quarter, Nebius extended server useful life from four years to five, cutting FY2026 depreciation by a disclosed $167.6M. Normalize and the segment margin is closer to 34%. CoreWeave uses six years; H100/H200 obsolescence runs three to four in practice.
- $402M FY2025 operating cash flow is financing in costume. Strip $982M of customer prepayments and a $982M accounts-payable build and continuing-ops CFO is roughly −$1.56B against $4.1B of capex. The flywheel only spins while two anchor customers keep prepaying.
The market is paying for $46B / 45% / +$402M. The footnotes read $31B / 34% / −$1.56B.
4 · Money picture
An operating-leverage inflection on a balance sheet still half-built
$399M
Q1'26 revenue
+684% YoY, +75% QoQ
45%
Segment EBITDA margin
vs −106% one year ago
$2.5B
Q1'26 capex
6.2× revenue
$0.9B
Net cash
after $4.3B converts + $2B NVDA
ARR went from $90M at end-2024 to $1.92B at end-Q1'26, a compound run-rate of roughly 85% per quarter as Microsoft and Meta capacity lit up. FY2026 capex is guided to $20–25B (raised from $16–20B on May 13, 2026) against $3.0–3.4B of revenue. The next twelve months turn on whether Q2 ARR holds the 30%-QoQ trajectory while owned sites in Pennsylvania, Missouri, and Alabama come online on schedule.
5 · Forensic flags
Numbers driving the rerate sit on a control environment the prior auditor failed
- Two material weaknesses, auditor in transition. Reanda — a small Amsterdam firm — issued an adverse internal-control opinion on FY2024 and re-flagged fixed-asset controls and TripleTen revenue recognition on FY2025. Deloitte is up for shareholder ratification at the 2026 AGM; the first clean opinion would land spring 2027.
- Reported earnings are not operating earnings. FY2025 net income of $83M is entirely non-operating: a $598.9M ClickHouse remeasurement, $55M of money-market gains, $27M of FX. Continuing-operations operating loss was $612M.
- Receivables grew 4,795% on revenue growth of 479%. Ending DSO is 587 days. The same period's useful-life extension, $52.6M of capitalized interest, and a $43.6M equipment-loss-in-transit charge excluded from adjusted EBITDA all sit inside the flagged controls gap.
Investors are extending hyperscaler-class credibility to a margin print produced inside a control environment the auditor called inadequate.
6 · Bull & Bear
Lean long, wait for the print — quant case carries; forensic file is real enough to size in only after Q2
- For. 6.6× EV on exit-2026 ARR at the guide midpoint vs CoreWeave at 12–18×, on faster growth and lower leverage.
- For. Net cash $0.9B going into a $20–25B capex year (raised May 13, 2026) while CoreWeave carries ~$11.5B net debt; NVIDIA's $2B equity stake is the cleanest supply-priority signal in the cohort.
- Against. Roughly 95% customer concentration with two counterparties who each run $65B+ of annual AI capex on owned capacity.
- Against. $402M operating cash flow becomes −$1.56B once prepayments and the AP build are stripped; two ICFR material weaknesses; useful-life extension, capitalized interest, and one-time exclusions all point the same direction.
My view — bull owns the next twelve months on the multiple gap and the balance sheet; bear owns the tail. Verdict moves to clean long on a disclosed third $1B+ anchor that drops top-2 concentration under 70%; moves to avoid on Q2 ARR under $2.2B or any shortening of the Microsoft or Meta tenor.
Watchlist to re-rate: Q2 2026 ARR print (July 29); firm-vs-flex backlog disclosure in the Q2 letter; first asset-backed deal coupon and size; Deloitte ICFR opinion on FY2026 (spring 2027).