Morgan Stanley Destroys Bear Case Against Nebius. Proves The 35% Plunge Was a Huge Mistake

Morgan Stanley projects total hyperscaler compute capacity will rise from 30.5GW in 2025 to 116.6GW by 2028 (+282%), implying $4T–$8T of infrastructure capex through 2028—undercutting the “compute oversupply” bear narrative. Nebius posted 1Q revenue growth of 684% YoY and reiterated a $7B–$9B annualized run-rate by year-end, while launching an asset-light partnership model that shifts new data-center financing to third parties. The article argues Nebius’ roughly 35% sell-off may be an overreaction to a problem that may not exist, though execution risk remains.

Analysis

The investable signal here is less about “AI demand” and more about who controls scarce bottlenecks: power, interconnects, and long-duration capital. If hyperscaler buildouts stay on track, the economic rent likely shifts toward balance-sheet fortressed platforms and away from asset-heavy neoclouds that must keep refinancing growth. That makes AMZN, MSFT, and META the cleaner structural winners; NBIS can still compound, but only if its partner-funded model really lowers cost of capital faster than competitors can copy it.

For CRWV and IREN, the market is likely overfocusing on revenue acceleration and underpricing funding dilution risk. In a strong demand tape, these names can still underperform if every incremental MW requires expensive leases, convertibles, or equity issuance; the stock can lag even while the business grows. The second-order winners are not just hyperscalers but also project finance, grid equipment, and land/power developers; MS may see indirect upside if capital markets activity around AI infrastructure stays hot.

The key time horizon is 1-3 months for confirmation that capex intent is turning into signed orders and power delivery, versus 6-18 months for whether utilization and monetization catch up. The contrarian risk is that the bull case is still assuming perfect execution on a power-constrained buildout; if interconnect delays or weaker AI ROI emerge, neocloud multiples can compress again before revenue inflects. What would falsify the positive view is a slowdown in hyperscaler capex guidance, partner-funded expansion failing to convert into backlog, or another quarter where announced capacity outpaces contracted demand.

Key Decisions for Investors

• Long NBIS on weakness over the next 2-4 weeks; thesis is that asset-light expansion deserves a higher multiple than self-funded peers if partner capacity additions are visible. Risk/reward: favorable if management can prove external financing of new MW; falsify if revenue growth decelerates or dilution accelerates.
• Pair trade: long NBIS / short IREN for 1-3 months. NBIS has a cleaner funding profile, while IREN’s market is more likely to penalize capital intensity if the tape turns risk-off. Cover if IREN shows materially better power-secured growth or NBIS loses pricing power.
• Long AMZN or MSFT into the next hyperscaler capex print, but size as a quality growth trade rather than a pure AI momentum trade. Upside comes from being the cheapest capital in the system; risk is a capex pause or margin pressure if AI spending fails to monetize within 6-18 months.
• Use CRWV as the hedge leg against the AI infrastructure basket rather than shorting the whole theme. The stock is the most exposed to financing optics; if it re-rates higher, it likely needs fresh evidence of contracted demand and lower cost of capital.
• Alert, not a trade: if MS starts showing clear underwriting / project-finance wins tied to AI infrastructure, the stock could benefit from fee growth and lending spreads. Verify via capital markets commentary rather than headline AI enthusiasm.