Web Watch

Web Watch in One Page

The Nebius thesis comes down to five live questions that the report says will be answered inside the next four quarters. First, the July 29, 2026 Q2 ARR print — the single number management has telegraphed and the bull case is written against (above $2.5B confirms the $7–9B exit ARR guide; below $2.2B breaks it). Second, whether a third $1B+ hyperscaler or sovereign customer signs and drops the Microsoft + Meta concentration below 70% — the bear's only true cover signal. Third, anchor-customer behavior on the other side: any disclosed tenor-shortening, repricing, or renegotiation by Microsoft or Meta, or evidence those two are accelerating owned AI capacity in a way that displaces reserved purchases. Fourth, the GPU supply cycle — H100 / B200 / Vera Rubin NVL72 spot pricing and first-deploy timing, since a 5x supply step in H2 2026 is the cleanest "2001 fiber" template trigger and is not priced into either consensus or bear PTs. Fifth, the financing-and-audit stack — the first asset-backed debt deal, the 2026 AGM ratification of Deloitte, and any interim disclosure of a third ICFR material weakness, all of which decide whether the forensic grade resolves up or down.

Active Monitors

Rank	Watch item	Cadence	Why it matters	What would be detected
1	Q2 2026 earnings, ARR print, and firm-vs-flex backlog disclosure	1d	The July 29, 2026 print is the highest-leverage event of the year; the $7–9B exit-ARR guide stands or falls on this single data point and the bull's $300 path is written down to ARR >$2.5B	Confirmed Q2 release date, pre-announcements, the ARR number itself, segment Adj EBITDA, Q3 capacity-activation commentary, and any new disclosure of the firm vs flex split inside the $46B backlog
2	Third $1B+ hyperscaler, sovereign, or large enterprise contract	1d	Customer concentration (~95% with Microsoft and Meta) is the bear's load-bearing argument; a new $1B+ anchor that drops top-2 concentration below 70% is the single cleanest cover signal	New named reserved-capacity contracts with AWS, Google Cloud, Oracle, Anthropic, xAI, a UK/EU/US sovereign program, or any other $1B+ multi-year compute commitment to Nebius
3	Microsoft and Meta contract status and AI capex-mix signals	1d	A renegotiation, tenor-shortening, or trim by either anchor would impair the entire switching-cost narrative and re-rate the equity by 30–50%; both customers are spending $80B+ and $65B+ on owned AI capacity and are the customer-becoming-competitor risk the 20-F itself flags	Press, transcripts, or filings indicating Microsoft or Meta repricing, walking back, restructuring, or expanding their Nebius reserved-capacity contracts, or accelerating owned AI campuses in a way that displaces external compute
4	GPU rental pricing, Vera Rubin NVL72 deployment, and NVIDIA NCP tier	1w	The cycle template question — H100 spot up 40% in six months is late-cycle behavior and a 5x supply step at the Vera Rubin launch in H2 2026 could reset reserved-contract economics before any demand recession arrives	SemiAnalysis spot-price prints, Vera Rubin first-deployer announcements, NVIDIA Cloud Partner Reference Platform tier changes, on-demand vs reserved-rate spreads, and competitor (CoreWeave, Oracle) Rubin deployment news
5	Capital structure execution and auditor / ICFR status	1w	Funding the $20–25B 2026 capex cycle and resolving the adverse FY2024 ICFR opinion are the two governance / financing levers that gate the multiple; either a clean ABS print inside the 1.25% convertible or a clean Deloitte sign-off would materially de-risk	First asset-backed financing terms (size, coupon, tenor), AGM ratification of Deloitte, any interim disclosure of a third ICFR material weakness or restatement, and any new debt, ATM equity, or rating-agency action

Why These Five

The report's central tension is that the quantitative case (ARR trajectory, $0.9B net cash into a $20–25B capex year, NVIDIA equity backing) leans bullish while the forensic file (customer prepayments propping cash flow, useful-life extension lifting margin, two ICFR material weaknesses, ~95% backlog with two customers) leans bearish — and the verdict explicitly waits for the Q2 ARR print before sizing in. Monitor 1 captures that exact event. Monitor 2 isolates the bear's single cover signal (concentration de-risking). Monitor 3 isolates the bear's single trigger (anchor-customer renegotiation or in-sourcing). Monitor 4 watches the cycle template — supply normalization, not a demand recession, is the variant-perception risk that neither the bull nor median PT prices. Monitor 5 covers the financing channel and the audit-quality reset, which together determine whether the forensic grade moves to Medium or to High before the FY2026 20-F lands in spring 2027. Together they map every open question the report still leaves on the table.