Meta commits to an additional $21 billion spend with CoreWeave for AI compute, on top of a prior $14 billion agreement, totaling over $35 billion through 2032.
This is a long-term capital commitment to secure compute years in advance, despite Meta's own CAPEX of up to $135 billion this year for building data centers.
The AI compute bottleneck is severe, with insufficient supply to meet growing demand.
CoreWeave is positioned with NVIDIA GPUs, essential for training and running large AI models.
CoreWeave's customer concentration is improving; no single customer is expected to exceed 35% of revenue, reducing prior reliance on Microsoft.
CoreWeave is raising $3 billion in convertible senior notes due 2032 and $1.25 billion in senior notes due 2031, on top of existing debt of nearly $30 billion.
CoreWeave's share price initially rose 9% premarket but reversed due to concerns over debt proceeds being used for general purposes, including repaying existing debt.
Revenue commitments are long-term, but borrowing is also long-term, highlighting ongoing financial pressures.
Future compute costs may not decrease as expected due to factors like rising shipping costs, building material costs from geopolitical issues (e.g., Iran war), and doubling memory costs.
This challenges the assumption that economies of scale and efficiency will lower compute costs in the near term.