Verdict: KILL. No position at current levels. Active monitoring for factor-driven re-entry.

What This Is

Nebius Group N.V. ($NBIS) is a European-domiciled AI cloud infrastructure company built from Yandex's carcass. $24.8B market cap. Revenue grew 547% YoY to $228M in Q4 2025. ARR $1.2B. Guidance: $3.0-3.4B revenue in 2026 with 40% adjusted EBITDA margins. CapEx plan of $16-20B. Anchor customers are Microsoft (majority H2 delivery) and Meta (fully deployed). Owns >25% of ClickHouse (≈$3.75B). $3B cash, $3.16B in convertible notes at $138.75 strike.

The bull case is real: $18B operating EV (after subtracting $3B cash and $3.75B ClickHouse) on 500%+ growth with sold-out capacity and hyperscaler anchors. At 15x 2028 EBIT the stock is a 3-4x in 2.5 years. 5.6x operating EV/Revenue is not heroic for this growth rate.

The question isn't whether NBIS is cheap. It's whether we have any informational advantage in knowing that.

We don't. And the cost of that conclusion — passing on a potential 53% return at consensus target — is real. At low-edge sizing (2%), that's ≈1% portfolio contribution left on the table. The kill is not costless. It's a bet that our next-best allocation has higher risk-adjusted return per unit of edge. Given the structural dynamics below, we think it does.

The Kill Conditions

1. No delta on any forward variable. Six variables tested. On every one, our estimate falls within sell-side range. Revenue ($2.8-3.2B vs guide $3.0-3.4B), EBITDA margin (35-40% vs guide ≈40%), customer concentration, financing mix, demand sustainability, governance. None produce a view materially different from the 12 analysts covering this stock.

2. No factor edge. Three regression specs produced idio variance of 65.7%, 69.4%, and 70.4% — a 5-point range depending on factor choice. Honest statement: idio is probably 65-75%, roughly at our threshold, not cleanly below it. The point estimate is noisy with 250 observations and 92.6% idio vol. What IS clear: 20-30% of variance is tech momentum (QQQ beta 17%, MTUM 22.5% in best spec). We have zero edge on either factor. The stock partially trades as a leveraged tech momentum bet — available through ETFs at zero fee.

3. No forced buyer. The strongest structural observation. Nobody is FORCED to buy on any positive scenario. Convert arbs sell into rallies (delta hedging increases as price approaches $138.75). On the downside, multiple forced sellers: momentum funds unwind, margin longs liquidate, and the company itself may sell stock (ATM authorized with Goldman/MS/BofA/Citi). This asymmetry — zero forced buyers, multiple forced sellers — is the actual signal, more reliable than any valuation math.

4. No informational advantage. $24.8B market cap, 12 analysts (75% bullish, median target $150), +170% in 12 months. Zero insider P-code purchases. Every filing is public and well-parsed.

What We Actually Found (Primary Sources)

Seven 6-Ks, the 20-F, and the Q4 transcript. Cross-referenced MSFT and META transcripts, CRWV 10-Q.

Microsoft doesn't know Nebius exists. Cross-corpus search of 3,962 earnings transcripts: zero mentions of "Nebius" in any MSFT or META transcript. Zero mentions of "third party cloud" or "GPU as a service" in MSFT. Classic asymmetric dependency — NBIS is one of several commodity GPU suppliers to MSFT, not a strategic partner. MSFT is farming out AI infra buildout to multiple vendors (NBIS, CRWV, IREN, others).

The financing gap is real and unfilled. $6.4-8B needed for 2026 capex beyond internal funding. The September 2025 6-K contains intent language: "expects to finance... debt secured against the contract... at terms that reflect the credit quality of the counterparty." Six months later — no binding terms filed. The CFO repeated the same language on the Q4 call word-for-word. This is the single most important variable and it's unknowable from public sources. Debt-heavy = equity preserved. Equity-heavy = 20-35% dilution. Without this, any per-share price target is fiction.

Orbis exited 74% of its position. 12.8M shares to 3.3M in Q4 2025. ≈$925M sold into strength. This is less bearish than it first appears — Orbis is a value fund (Allan Gray family) that rode the Yandex restructuring. They never had a growth thesis. A value fund selling a stock that now trades as an AI momentum name at 5x their entry is mechanical profit-taking, not a directional signal. The signal would be if a growth fund that bought at $80+ was dumping. What IS significant: zero Form 4 insider purchases anywhere. Nine Form 144 sale filings. Management has information we don't, and none of them are buying.

Material weaknesses in internal controls. 20-F FY2024 (lines 4706-4717): three specific MWs — IT general controls, revenue recognition ("inability to consistently establish accurate revenue statements"), and fixed asset reconciliation. Reanda issued an adverse IC opinion. Deloitte appointed Feb 2026 as replacement. Management says no misstatements resulted. Remediation expected Q4 2025 — status unknown until FY2025 20-F. Revenue recognition weakness in a company doing $834M quarterly operating cash flow on customer prepayments warrants monitoring.

Depreciation change: real but narrower than first assessed. 4-year to 5-year GPU depreciation starting Q1 2026. Does NOT affect EBITDA (which excludes D&A by definition — we got this wrong initially). Does affect EBIT by ≈$275M/yr on a $5-6B average GPU base, or ≈8-9 points of margin. Their "20-30% medium-term EBIT target" would be 12-22% under the old schedule. Aggressive but not unique — CRWV does similar.

GPU pricing compression is not happening. Cross-corpus search found zero evidence of pricing pressure, oversupply, or capacity normalization across 3,962 transcripts. Every source says sold out, strong pricing, allocation situations. Our initial bearish call on pricing was base rate projection with no supporting data.

The CRWV Catalyst (Feb 26 — 3 Days)

CRWV reports Q4 earnings on Feb 26. This is not a monitoring item — it's an imminent factor event that could create the re-entry conditions this memo describes.

Both stocks trade as AI infra momentum names. Identical short interest (NBIS 17.3%, CRWV 17.4%). NBIS has 22.5% momentum factor loading. CRWV is the bigger, more leveraged, more MSFT-concentrated expression of the same trade ($46.5B cap, $14B debt, 67% MSFT).

If CRWV misses and drops 15-20%:

NBIS sympathizes -10-15% on factor contagion. That puts NBIS at $83-88 — inside the zone where momentum funds start unwinding and the reflexive ATM loop activates. March max pain is $90; June max pain is $80. Put OI is concentrated at $80 (6,925 contracts) and $75 (4,120 contracts). The institutional hedging is already positioned for this scenario.

At $80-85, NBIS operating EV/exit ARR approaches 1.5x. At $70, ClickHouse alone covers >25% of market cap and the convert arb floor activates. That's where the risk/reward flips — buying a real company at a factor-crash price, not trying to out-analyze 12 sell-side analysts.

If CRWV beats: Both stocks rally. Convert arbs absorb NBIS upside above $130. Rally is capped by structure, not fundamentals.

If CRWV credit event (tail): $14B debt stack under stress reprices the entire AI infra sector. NBIS — with $3B cash, lower leverage, and no debt covenants on converts — becomes the relative winner. Sector panic creates the entry that idio analysis couldn't find.

The asymmetry favors patience: upside is capped by convert structure, downside is where the opportunity opens. Watching CRWV Feb 26 is not "monitoring a trigger" — it's watching the most likely near-term mechanism that could change this from kill to buy.

Why the Bull Case Doesn't Create Edge

The bull case IS consensus. Median analyst target is $150 (53% upside). 75% of covering analysts are bullish. The stock is up 170% in 12 months. The re-rating from Yandex-shell to AI-infrastructure-play already happened.

Per-share economics diverge from EV economics. The $56-75B EV bull case at 15x 2028 EBIT assumes current share count. If 20% of $40-50B cumulative capex is equity at average $90, that's ≈90M new shares — 35% dilution. Convert dilution adds another 9%. The company has Goldman, Morgan Stanley, BofA, and Citi on retainer for the ATM. They have every incentive to issue into strength. The EV math works. The per-share math might not.

Process Errors and What They Mean

We built six differentiated observations. Of the four that could be resolved against primary sources, two were wrong: EBITDA/EBIT confusion and unsupported GPU pricing compression. Both errors were bearish — they favored our kill conclusion.

This is not a "33% error rate." It's a 50% error rate on resolved items (2 right, 2 wrong out of 4 resolved), with systematic directional bias. Our analytical process on this stock was biased bearish. The EBITDA error is particularly telling — confusing EBITDA with EBIT is a basic definition, not a subtle analytical judgment. The pricing compression call was a vibes-based projection dressed up as analysis.

The corrected findings actually strengthen the bull case, not the kill. Depreciation doesn't flatter EBITDA margins. Pricing isn't compressing. Customer concentration was lower than assumed. After correcting our own errors, the company looks better than our initial read suggested.

The kill still holds — but on structural grounds (no forced buyer, no informational advantage, unknowable financing), not on the bearish analytical observations that turned out to be wrong. This distinction matters: the kill is about our edge, not about the company's quality.

Positioning and Structural Dynamics

The convert structure creates the most important microstructure dynamic. $3.16B at $138.75 conversion price means:

• Rally toward $130+: Arbs short MORE stock (increasing delta hedge). Structural selling pressure into strength. Ceiling near $138.75.
• Drop below $80: Arbs cover shorts (decreasing delta). Floor support around $65-70 where delta approaches zero.
• Short interest (17.3%, 3 days to cover, ≈$4.2B notional): Not a squeeze candidate. It's structural — convert arbs NEED the short. Hedging, not directional.

The corridor ($70 floor, $139 ceiling) is defined by capital structure, not fundamentals. Stock is pinned.

Below $80: the reflexive loop activates — company needs capital, ATM selling creates supply, price drops further. This is the SMCI/CRWV playbook where capital-intensive companies with ATM programs create their own selling pressure at the worst moment.

What Would Change This

Five re-entry triggers, ranked:

1. CRWV miss + NBIS sympathizes to $80-85 (potential: Feb 26). Not a "monitoring" item — it's a 3-day catalyst. At $80-85, the math changes: operating EV/exit ARR ≈1.5x, put OI support at $80, convert floor approaching. This is the most actionable trigger because it has a date and a mechanism.

2. 6-K filing with binding asset-backed debt terms. Closes the #1 gap. Investment-grade debt backed by MSFT contract at 5-6% resolves dilution bullishly. 15-20% stock move.

3. Stock below $70 on sustained factor crash. ClickHouse covers >25% of market cap. Convert arb floor provides structural support. Operating EV/exit ARR below 1.0x. At that point you're buying the ClickHouse stake and getting the AI business nearly free.

4. Insider Form 4 with P-code >$500K. First conviction signal from anyone with inside information. Against 9 Form 144 sales, a purchase would be genuinely new data.

5. ClickHouse IPO or sale filing. Crystallizes $3.75B of embedded NAV.

LR Signal: 0.85

Mild bearish. Market prices NBIS for execution on known guidance with manageable dilution. Our analysis broadly confirms this is reasonable — the bull math works on EV — but the dilution path is murkier than priced, forced actor asymmetry skews downside, and we identified systematic bearish bias in our own process that, once corrected, made the bull case stronger.

The signal: our error-corrected analysis is mildly more bullish than our initial read, but still without edge. The slight overpricing is in certainty around per-share economics, not in the business quality. This is a company that probably executes on its guidance and probably reaches $140-150 — but "probably" at consensus odds with no informational advantage and structural headwinds on the path isn't an allocation.

The opportunity isn't here. It's in the factor crash that may arrive in 3 days, or 3 months, or never. Watch CRWV Feb 26. If the sector breaks, the entry this analysis couldn't find from fundamentals might present itself through positioning.

No allocation at $98. Active watch at $80-85. Aggressive interest below $70.