NBIS research

Position in the Physical Layer of AI thesis

NBIS is the datacenter anchor. The Physical Layer of AI thesis says every hyperscaler needs power, networking, memory, AND somewhere physical to put the GPUs. Microsoft, Meta, and Amazon are not bottlenecked on Nvidia silicon, they are bottlenecked on the steel-and-concrete capacity to plug it in. Nebius runs the neocloud that the lead customers pay to skip the build-it-themselves wait. That is a pure capacity arbitrage with multi-year contractual lock-in, not a "GPU rental" commodity.

Quinn's book opened NBIS on 2026-02-09. The 2026-05-11 canonical snapshot has NBIS at +80.59%. That is the second-largest lifetime gain in the book.

Recent catalysts (last 30 to 60 days)

$27B Meta contract (April). Multi-year capacity reservation. Meta CapEx call language shifted from "evaluating" to "executing" on neocloud. NBIS named, indirectly, in the same-week 10-Q footnote about hyperscaler customer concentration.
$19.4B Microsoft backlog (March). Disclosed as a "long-term capacity commitment" in the Q4 print. Backlog cleared $50B total per the Q1 2026 earnings call.
547% revenue growth (Q4 2025 vs Q4 2024). The growth rate is real, the comp is small. The market is pricing the next-12-months curve, not the trailing one.
Capex matched-funded. NBIS issued bonds backed by the multi-year customer contracts. Spread tightened on issuance, which is the bond market voting that the contracts are real.

The thesis (what has to be true)

1. Hyperscaler capacity demand keeps outrunning the in-house build-rate at Microsoft + Meta + Amazon through at least 2027. If they catch up on their own data center pipeline, the NBIS contract value compresses fast. 2. The $27B Meta + $19.4B Microsoft backlog converts to revenue on the disclosed timeline. No renegotiation, no cancellation clauses fired. 3. Power and grid bottlenecks ALSO keep the in-house pipeline slow. The Physical Layer thesis is self-reinforcing: NBIS wins because VST is constrained. 4. NBIS does not get acquired at the wrong price. The capacity arbitrage is more valuable to a strategic acquirer (a hyperscaler buying its own escape valve) than to a private buyer.

Kill vectors (what would break the thesis)

Backlog cancellation. Either Microsoft or Meta substantially redirects 2027 capacity to in-house build. Predator agent fires on a disclosed cancellation or a 10-K backlog re-statement >20% down.
Pricing pressure. A new neocloud entrant (Coreweave international expansion, Lambda Labs scale-up) compresses the per-GPU-hour rate. The contract structure protects revenue for the locked term, but next-cycle pricing breaks.
Power assignment fail. NBIS announced sites depend on grid interconnects that aren't fully closed. A regulator-driven hold (NY, NJ, VA) delays revenue ramp by quarters.
Mechanical exit at break. Discipline rule fires on close below $130 for two consecutive sessions. Written exit price, locked at entry.

Layer context

In the 8-layer Physical Layer of AI map, NBIS is the Datacenters anchor. The layer is currently at ~13 percent of book (under the 30 percent sector cap). Sister names in the layer: CRWV (CoreWeave, pure-play AI cloud), IREN (BTC-to-AI conversion).

NBIS is the cleanest pure-play because the entire backlog is named-customer hyperscaler. CRWV has more public-customer concentration risk. IREN is a hybrid economics play. If you only want one Datacenter anchor, NBIS is the one.

Position discipline (the rules at entry)

Trim trigger: +200 percent from cost. Locked pre-entry.
Kill vector exit: break of $130 for 2 consecutive days. Full exit, no negotiation.
4-quarter thesis-fail rule applies (Discipline-v2 Rule 2). If Meta or Microsoft restates backlog down for 4 consecutive quarters, exit regardless of price.

Moat 7. Asym 9. The asymmetric pay is high because the backlog is real and the next 12 months reprice if even half of it converts on schedule.

Real money. Real position. Real receipts.