Meta Is Spending $48 Billion With CoreWeave and Nebius. Which Neocloud Stock Is a Better Buy?

Meta is leaning on neocloud providers CoreWeave and Nebius to secure AI computing capacity, with Meta reportedly committing $21 billion to CoreWeave and up to $27 billion to Nebius. Both providers are funding rapid data center expansion with debt—Nebius issued $8.3 billion in March and CoreWeave added $6.25 billion in April—while remaining loss-making, with 2025 losses of $447 million and $1.2 billion, respectively. The piece is primarily a valuation and business-model comparison, arguing CoreWeave may have less uncertainty than Nebius, but long-term returns for both remain highly uncertain.

Analysis

The immediate winner is not the neoclouds themselves but the GPU and networking ecosystem that sits one layer upstream. Every incremental dollar Meta pushes into third-party AI capacity supports accelerated purchasing of Nvidia systems, interconnects, power, and liquid-cooling infrastructure, which extends the capex cycle even if Meta is simultaneously building in-house silicon. That creates a second-order read-through to suppliers with pricing power and long lead-time backlogs, while also tightening the financing market for any AI infrastructure player with contracted revenues.

The hidden risk is that these contracts may be economically attractive only in the current scarcity window. If hyperscalers close the capacity gap over the next 6-18 months, the neoclouds could face margin compression exactly when debt service is peaking and depreciation is ramping. In other words, the asset base is being built for peak demand, but the exit multiple will be determined by mid-cycle utilization, not today’s backlog.

The market’s current focus on growth rates is likely underweighting capital structure fragility. In a lower-rate environment these balance sheets look manageable; in a slowing AI spend environment, refinancing risk becomes the primary equity variable. The more fragile name is the one that needs equity more often to fund expansion, so dilution risk is not just a shareholder issue—it can also pressure customer confidence if counterparties start questioning long-run platform stability.

Contrarian angle: the consensus is treating neocloud exposure as a clean “AI picks-and-shovels” trade, but the better trade may be to own the enabling monopolies and fade the capital-intensive intermediaries. The crowded bullish narrative on AI infrastructure can persist for quarters, but the first sign of any capacity normalization will hit the neocloud equities far harder than the hyperscalers or Nvidia. The asymmetry is that upside is tied to continued scarcity, while downside is triggered by merely adequate supply.

Key Decisions for Investors

• Long NVDA vs short CRWV on a 3-6 month horizon: capture continued upstream demand while fading the more levered, dilution-prone vehicle most exposed to utilization normalization. Use a small net debit if pairing with call options on NVDA.
• Avoid initiating fresh long NBIS here; if already owned, trim 25-50% into strength over the next 2-4 weeks. The combination of earlier-stage losses and probable financing needs makes the equity highly sensitive to any slowdown in contract announcements.
• Buy META on pullbacks as the more durable way to own AI infrastructure spend: 6-12 month horizon, with the thesis that in-house build plus third-party spillover lowers effective execution risk versus owning the providers.
• For tactical upside, consider a call spread on NVDA rather than outright long CRWV/NBIS. Risk/reward is better because the embedded leverage sits in the supplier with balance-sheet resilience, not the customer-funded constructors.
• Watch for any commentary on build-out delays, secondary equity issuance, or refinancing spreads in CRWV/NBIS over the next quarter; those are the first indicators that the scarcity trade is transitioning into a financing trade.