Zacks Investment Ideas feature highlights: CoreWeave, Meta Platforms and NVIDIA

CoreWeave said it has a multi-year agreement with Anthropic and an expanded long-term deal with Meta Platforms to provide AI cloud capacity through 2032 for about $21 billion. The company’s sales backlog reached a record $66.8 billion at FY25, supporting strong revenue visibility even as near-term EPS remains pressured by heavy AI infrastructure CapEx. The article also underscores NVIDIA’s strategic link to CoreWeave and highlights continued bullish revisions for NVIDIA and CoreWeave amid AI-driven investor enthusiasm.

Analysis

The market is increasingly pricing AI infrastructure as a capacity-constrained utility rather than a software multiple story. CoreWeave’s new contracts matter less for near-term earnings than for de-risking a long-duration financing machine: when backlog visibility extends several years, equity can absorb heavy CapEx today as long as lenders and hyperscaler customers keep validating demand. That said, the same model creates a reflexive trap—if utilization or pricing weakens even modestly, fixed-cost leverage can reverse quickly because the asset base is front-loaded and depreciation lags cash spend.

Meta is the more interesting second-order beneficiary. Extending external cloud capacity implies Meta is preserving optionality without committing fully to internal buildout, which should help it keep training velocity high while avoiding near-term bottlenecks. But it also signals that the largest buyers of AI compute are still willing to outsource a meaningful slice of demand, which is bullish for capacity providers and subtly bearish for any assumption that hyperscalers will fully internalize the stack.

NVIDIA remains the cleanest exposure, but the incremental upside is increasingly tied to supply-chain control and attached ecosystem economics rather than headline model demand. The tight relationship with CoreWeave reinforces the idea that scarce compute allocation is the real moat; however, that also concentrates risk in a small set of counterparties and financing channels. If debt markets tighten or customer concentration becomes a concern, NVIDIA’s demand is likely the last thing to break, but adjacent names with weaker balance sheets could re-rate sharply lower.

Consensus may be underestimating how much of this move is capital-markets driven rather than fundamental EPS driven. The rally can continue for months if investors keep rewarding backlog and deployment speed, but the setup is vulnerable to any signal that AI spend is outpacing monetization. The biggest contrarian risk is that capacity announcements are being read as durable demand when they may simply reflect a temporary scramble to secure compute ahead of a more disciplined procurement cycle later this year.

Key Decisions for Investors

• Long NVDA vs short a basket of smaller AI infrastructure beneficiaries with weaker balance sheets over the next 1-3 months; buy quality/short leverage as the market starts distinguishing funded demand from financing risk.
• Use pullbacks in CRWV to establish a tactical long only if volume-backed and above prior breakout levels; target 15-25% upside over 4-8 weeks, but keep a tight 10-12% stop because the name is highly reflexive.
• Consider a pairs trade long META / short software-enablement peers that depend on enterprise AI spend, as META’s ability to self-fund compute reduces execution risk while many software names still need proof of monetization.
• For conservative exposure, prefer NVDA call spreads over outright long CRWV: 2-4 month tenor, limited downside, and participation if AI capex remains resilient.
• If CRWV gaps higher again without a fresh contract catalyst, fade part of the move via call overwriting or partial profit-taking; the risk/reward deteriorates quickly once backlog news is fully capitalized.