Key Takeaways

  • CoreWeave reportedly increased the yield and offering discount on a $2.6 billion loan after investors pushed for more compensation.
  • The financing would support capacity tied to Anthropic, adding to scrutiny of CoreWeave’s roughly $21 billion debt stack.
  • CoreWeave’s $99.4 billion backlog supports its growth case, but customer concentration and undisclosed contract economics remain key risks.

CoreWeave has reportedly improved the terms of a $2.6 billion leveraged loan intended to finance computing capacity associated with Anthropic, a sign that debt investors are becoming more selective about funding the artificial intelligence infrastructure buildout.

According to Bloomberg, pricing on the loan widened to as much as 5.5% over the benchmark rate. The offering discount was cut to 97 cents on the dollar, meaning investors would pay less than the loan’s face value. Both changes effectively increase the prospective return for lenders while raising CoreWeave’s financing cost.

The adjustment indicates that investors are asking harder questions about leverage, contract quality, and how quickly expensive computing assets can generate cash.

CoreWeave is pursuing a capital-intensive growth model at unusual speed. Its debt stack has been reported at around $21 billion, while its backlog reached roughly $99.4 billion following Q1 2026 bookings. That backlog gives CoreWeave a substantial pool of contracted demand to present to lenders, although backlog and immediately available cash flow are not the same thing.

The Anthropic agreement illustrates the resulting uncertainty. CoreWeave has described it as a multibillion-dollar arrangement, but detailed financial terms were not disclosed. Investors therefore have limited public information about pricing, margins, payment schedules, and the protections available if demand changes. The loan repricing suggests creditors want a wider cushion to offset concerns regarding whether expected contract revenue is sufficient to compensate for construction, power, chip, and borrowing costs.

Long-duration infrastructure agreements sometimes use take-or-pay provisions, under which customers commit to pay for reserved capacity even when actual usage is lower. Such provisions can make revenue more predictable and support debt underwriting. Their value still depends on the customer’s credit profile, the duration of the commitment, and the provider’s ability to deliver the promised capacity.

Customer concentration introduces additional risk. Meta added a $21 billion commitment in 2026 alongside earlier commitments, demonstrating how a relatively small group of hyperscale companies and model developers account for a large share of CoreWeave’s commercial pipeline. Reuters has covered CoreWeave’s expansion and the role of major customer commitments in its growth trajectory.

While large contracts provide revenue visibility, they also magnify counterparty and renegotiation risk, especially when infrastructure is built around specialized workloads or a limited number of customers. If deployment schedules shift, the financing remains in place even when the associated revenue arrives later than expected.

The broader AI market is absorbing enormous spending on accelerators, data centers, networking equipment, and electricity. Loan buyers are increasingly separating strong demand signals from durable economics. Morningstar has examined CoreWeave’s financial profile as markets weigh rapid revenue growth against leverage and ongoing capital requirements.

These financing terms offer a broader market signal, as higher borrowing costs can eventually influence capacity pricing, contract duration, and minimum-spend commitments offered by infrastructure providers. Customers may also place greater emphasis on supplier resilience, delivery milestones, and remedies for delayed deployments.

CoreWeave still enters those discussions with a large backlog and commitments from Anthropic and Meta. Yet the revised loan terms show that contracted demand alone may no longer secure the most favorable financing. Credit investors appear willing to fund the next stage of AI capacity, but they want to be paid more for the execution, concentration, and leverage risks attached to it.