$nbis will fly 👊 $CRWV just gave us an important read-through for $NBIS. CoreWeave Q3 2026: • $3.66B revenue, +168% YoY • $132.5B revenue backlog, +138% YoY • ~$1B sequential revenue growth in ONE quarter • 2026 exit ARR guidance raised to $20–21B • >1.85 GW active power expected by year-end • $37–39B CapEx planned for 2026 • ~$9B of customer prepayments/deposits expected in the next 3 months • Management says AI demand is roughly 10x available capacity To me, the message is clear: AI compute demand is NOT the problem. POWER, GPUs, deployment speed and financing are the bottlenecks. And this is where the Nebius story gets extremely interesting. Because the upside for $NBIS may not simply come from deploying more MW. It may come from generating dramatically more value — and eventually more profit — from every MW deployed. My current 2030 framework: BEAR • 4.0 GW active capacity • $16M ARR/MW • ~$64B ARR • ~33–35% EBITDA margin BASE • 5.3 GW active capacity • $20M ARR/MW • ~$106B ARR • ~43–47% EBITDA margin BULL • 6.2 GW active capacity • $24M ARR/MW • ~$149B ARR • 50%+ EBITDA margin Why could Nebius economics improve so significantly? Because the stack is moving beyond simply renting GPUs: Vera Rubin ↓ More compute per MW Eigen AI ↓ Kernel, model and serving optimization ↓ More throughput from the same hardware Inferize ↓ Lower GPU idle time ↓ Higher utilization Token Factory ↓ Nebius monetizes inference/tokens instead of only GPU hours Tavily ↓ Agentic search, retrieval and real-time web intelligence ↓ More AI calls per user interaction The economic flywheel becomes: 1 MW of power → more compute → more tokens → higher GPU utilization → higher ARR/MW → higher EBITDA/MW → higher FCF/MW That last point is crucial. Eventually I believe ARR/MW may no longer be the most important KPI for Nebius. EBITDA/MW — and ultimately FCF/MW — could become far more important. Take my 2030 Base Case: ~$106B ARR At a 45% EBITDA margin: $106B × 45% = ~$48B annualized EBITDA. Even at only 40%: $106B × 40% = ~$42B EBITDA. And this is where Vera Rubin + Eigen AI + Token Factory + Tavily potentially becomes so powerful. Nebius isn’t just trying to sell compute. It is trying to own more layers of the AI value chain: POWER → GPUs → CLOUD → INFERENCE → MODEL OPTIMIZATION → AGENTIC SEARCH → AI APPLICATIONS That could make the economics fundamentally different from a pure GPU-rental business. Illustrative 2030 valuation: BEAR: $64B ARR × 2.5x = ~$160B EV ≈ ~$450/share at ~350M diluted shares BASE: $106B ARR × 4x = ~$424B EV ≈ ~$1,210/share BULL: $149B ARR × 5x = ~$745B EV ≈ ~$1,800–2,000/share depending on dilution Obviously, these are scenario calculations — NOT company guidance. But CoreWeave’s latest numbers reinforce the key assumption behind the entire model: Demand for AI compute remains enormous. If Nebius can combine: multi-GW capacity growth Vera Rubin economics higher utilization through Eigen/Inferize Token Factory monetization Tavily-driven agentic workloads …then the upside may not be linear. It could be multiplicative. The big question for $NBIS is therefore no longer only: “How many GW can they deploy?” It is increasingly: “How much FREE CASH FLOW can they generate from every MW?” That may ultimately determine whether Nebius gets valued like a Neocloud… or like an AI infrastructure + software platform. $NBIS $CRWV