NBIS — knowledge base

Overview

Nebius (NBIS) is a neocloud AI infrastructure provider whose Q2 2026 blowout earnings (revenue +454% YoY, AI cloud ~$575M, adj. EBITDA beat, 5 GW contracted-power guide, >$9B prepayments) and ~34% stock surge are viewed as a sentiment turning point after 6–8 weeks of sector selling. Strategy centers on a three-bucket revenue model (hyperscaler anchor deals for cheap financing, high-potential frontier labs, and opportunistic short-term/burst auctions) that emphasizes market agility, rapid paybacks (~1 year 10 months), and multi-year cash-flow tails post-payback; company states it could sell entire 2027 capacity on current terms. Positioned as more reactive and commercially agile than asset-first peers such as Iron (IREN, framed as 4–6 months behind on revenue with vertical-integration margin upside and far smaller ~$15B vs Nebius ~$65B mkt cap); competes in a large multi-winner TAM where GPU access, prepayments, and financing are decisive edges. Colocation comps and longer GPU useful lives further support the broader neocloud/colo complex; bulls argue neoclouds quietly build durable moats versus louder colo LOIs and that the AI-compute cycle remains early.

Key facts & figures

Nebius Q2 2026 consolidated revenue $582.3M (+454% YoY from $105.1M); AI cloud revenue $575M (+514% YoY), ARR $3.0B [[s:124@00:04:28]][[s:124@00:05:04]].
Adj. EBITDA $236.2M vs ~$173M estimate (prior-year loss $21M) [[s:124@00:04:36]].
Stock closed $259.20 on 2026-08-12, +34.14% [[s:124@00:03:53]].
Deal paybacks compressed to ~1 year 10 months (under 2 years), down from prior 2–3 years; average contract/ACV yield >$20M/MW (specifically $20–25M) for Q2 landmark deals; pricing above $20M/MW [[s:124@00:07:30]][[s:124@00:04:45]][[s:122@00:06:45]].
Short-term/burst or auction deals reached $40–50M per MW (some higher); small portfolio slice [[s:122@00:24:21]].
Raised year-end 2026 contracted power target to 5 GW (from prior >4 GW); expects >$9B customer prepayments in 2026 (covering 50–60% of associated capex on many deals); could sell entire 2027 capacity on current terms today [[s:124@00:05:28]][[s:124@00:05:45]][[s:124@00:05:56]].
Post-earnings mkt cap ~$65B (range ~$62–70B); IREN ~$15B (~$14–15.7B) [[s:124@00:12:54]].
Three revenue buckets: (1) hyperscaler anchors enabling cheap financing, (2) high-potential frontier labs, (3) opportunistic burst/auction capacity.
CoreWeave A100 contracts extend through 2029 (Nvidia A100 launch 2020), implying ~9-year useful life and undercutting 2–3-year obsolescence bears [[s:122@00:05:47]].
Colocation comps: Riot–Anthropic Rockdale add-on ~191 MW, ~$9.1B/20-year lease ≈ $2.4M average annual revenue per MW (prior AMD capacity at site; total ~241 MW) [[s:122@00:12:23]][[s:122@00:12:57]]. Fermi–TensorWave Project Matador 222 MW, ~$6.5B/15-year ≈ $1.95M average annual per MW (speaker claim of ~$2.7M overstated) [[s:122@00:12:48]].
White Fiber (WYFI) post-space 100 MW GPU-capacity deal (Krambu, exclusive ops, 2027) drove stock +>20% [[s:122@00:04:36]].
IREN (Iron) guided +730 MW under construction for 2027 ramp toward 1.21 GW total; Goldman Sachs 13G ~9.4% stake (~33.64M shares); JPMorgan also disclosed stake [[s:122@00:10:04]][[s:124@00:14:53]].
CapEx to build ~1 GW HPC/AI factory cited ~$50–60B (Jensen Huang-linked range $40–80B) [[s:124@00:51:36]].
Bitdeer (BTDR) mkt cap ~$2–2.5B [[s:124@00:52:36]].
External headwinds: NY statewide moratorium on large/hyperscale data centers (Jul 2026, up to 1 year) [[s:122@00:14:55]]; Texas ERCOT Batch Zero large-load pause (early Aug 2026, Gov. Abbott-directed) hitting Cipher pipeline (heavily Batch Zero into 2027) [[s:122@00:16:02]].

Thesis & bull case

Large multi-winner AI-infra TAM still early; Nebius’s commercial agility and three-bucket mix (anchors + labs + $40–50M/MW burst) produce sub-2-year paybacks, >$9B prepayments, 5 GW contracted guide, and multi-year cash-flow tails—company could sell all 2027 capacity today on current terms. Hyperscaler anchors lower cost of capital; GPU contracting (exemplified by CoreWeave A100s to 2029) extends asset lives well beyond bear-case obsolescence. Rising colo comps (~$2–2.4M+/MW average annual) and $20–25M/MW ACV validate pricing power. Financing, prepayments, and GPU access remain decisive moats. Neoclouds framed as quietly building durable strategic moats versus “loud” colo LOIs that may prove shorter-cycle. IREN positioned as 4–6 months behind Nebius on revenue, vertically integrated with potentially higher margins, deeply undervalued (~$15B vs Nebius ~$65B); base-case ~4x (~$160) within a year, private fair value argued ~$120–140, institutional ownership (Goldman 9.4%, JPM) and narrative expected to drive next leg; multi-year 4–10x+ compounding preferred over short-term FOMO trims. White Fiber’s retrofit speed, Tier-3 operating team, low-GPU-CapEx cloud contracts, and NC1 financing flywheel illustrate attractive risk-reward for agile operators; Iron’s power/vertical integration still valuable on longer timelines. Mispriced miners (Bitdeer as power + Bitcoin call option/parts value ~$2B mkt cap; Marathon, Bit Digital) offer asymmetric upside. Skilled-labor bottlenecks are cyclical, not structural—pay attracts mobile crews (oil-field analogy). Constructive multi-name stance after CoreWeave/Nebius prints; demand for compute structural and power scarcity durable; quality AI-compute names can compound far beyond 10x over 10–20 years as usage scales.

Risks & bear case

Execution and communication gaps: Iron criticized as reactive, asset-first, and out-of-touch versus Nebius agility; management-trust and per-share/levered-return sophistication questioned—though bulls now overweight IREN as catch-up story. Pure power/real-estate plays may lag tech-operator multiples. Labor availability at remote Texas-style sites harder than coastal retrofits; construction vs. maintenance pay and regional culture differences matter. Power and interconnection remain constraints (ERCOT Batch Zero pause, NY moratorium). Financing discipline and GPU supply access will separate winners; short-term stock-price tribalism, retail FOMO/shitcoin-style trading, and under-risking generational upside (Samsung textiles-to-electronics parable) risk missing multi-year paths. Cipher heavily exposed to delayed Texas batch process into 2027. Overstated colo comps (e.g., Fermi) can inflate expectations. CapEx bears still argue large absolute build costs (~$50–60B/GW) despite compressing paybacks. Host not buying Nebius at post-print levels (~$259 / ~$65B) while chasing IREN relative value.

Timeline of developments

2026-07: New York statewide moratorium (executive order) on new hyperscale/large data centers (≈50 MW+ thresholds) for up to one year, impacting names such as Wolf [[s:122@00:14:55]].
Early 2026-08: Texas Gov. Abbott directs audit; ERCOT pauses/delays Batch Zero large-load review and transmission planning, affecting Cipher’s pipeline (heavily Batch Zero into 2027) [[s:122@00:16:02]].
2026-08-12: CoreWeave and Nebius Q2 earnings cited as narrative turn after 6–8 weeks of AI-infra selling; A100 contracts to 2029, Nebius sub-2-year paybacks and $40–50M/MW burst disclosed; Riot Anthropic and Fermi TensorWave colo comps discussed; White Fiber 100 MW Krambu deal noted [[s:122]].
2026-08-12: Nebius Q2 blowout detailed—revenue $582.3M (+454% YoY), AI cloud $575M (+514%), ARR $3.0B, adj. EBITDA $236.2M vs ~$173M est.; stock +34% to $259.20; guide raised to 5 GW YE 2026 contracted power; >$9B 2026 prepayments; could sell all 2027 capacity on current terms; ACV $20–25M/MW; IREN ~$15B vs NBIS ~$65B framed as 4–6 months behind with Goldman 9.4% stake; neocloud-vs-colo moat and multi-year IREN compounding thesis emphasized [[s:124]].

Open questions

Will labor or power prove the tighter bottleneck for neocloud/colo builds, and can elevated pay fully clear skilled-trades shortages at remote sites?
Which operating model wins on risk-adjusted returns: Nebius-style commercial agility and burst pricing, Iron-style power/vertical integration (now argued 4–6 months behind with margin upside), or White Fiber-style retrofit + tech-operator flywheel?
How durable are sub-2-year paybacks, $20–25M/MW ACV, $40–50M/MW burst pricing, and ability to pre-sell 2027 capacity once supply catches up?
Can Iron convert institutional ownership (Goldman 9.4%, JPM), revenue catch-up, and vertical integration into sustained multiple expansion and the hypothesized ~4x/~$160 base case within a year while scaling 730 MW into 2027?
Impact duration of ERCOT Batch Zero pause and NY moratorium on 2026–27 capacity ramps.
Do neocloud strategic moats outlast colo LOI cycles as bulls claim, and how much AI surplus accrues to pure-power/Bitcoin miners (Bitdeer et al.) versus dedicated neoclouds?

Notable predictions to track

Nebius three-bucket model continues to deliver paybacks ≤2 years and material burst revenue contribution at $40–50M+/MW; full 2027 capacity saleable on current terms [[s:122@00:06:45]][[s:122@00:24:21]][[s:124@00:05:56]].
Broader neocloud/colo complex experiences sustained narrative recovery post-CoreWeave/Nebius prints after prior 6–8-week drawdown [[s:122@00:03:35]].
Multi-year useful lives (A100s contracted to 2029 ≈9 years) continue to invalidate short 2–3-year obsolescence theses [[s:122@00:05:47]].
IREN base-case ~4x to ~$160 within a year (euphoria $200+; private FV ~$120–140 today) as 4–6 month revenue lag closes and institutions/narrative catch up; multi-year 4–10x+ compounding [[s:124@00:13:50]].
Neoclouds quietly capture durable herd/moat versus shorter-cycle “loud” colo bulls [[s:124@00:21:41]].
Quality AI-compute names can deliver far more than 10x over 10–20 years as usage scales to billions [[s:124]].