This Janus Henderson fund manager owns CoreWeave debt — here's why he's sticking with it

CoreWeave’s 9.75% bonds due 2031 yield 15.9%, up 73% since issuance, while the company carries about $51.6 billion in total debt and its five-year CDS spread is 1,092 basis points. Its shares fell 20% over the past week, though a Janus Henderson credit investor called the bond selloff an overreaction and expects the company to manage financing through operating cash flow, GPU financing, or customer prepayments. CoreWeave’s estimated 2025 capex of $36.4 billion exceeds consensus EBITDA of $7.5 billion; analysts estimate 2026 revenue of $26.5 billion, EBITDA of $16.7 billion, and capex of $47.9 billion.

Analysis

The key issue is not whether AI compute demand exists, but whether contracted demand converts into cash quickly enough to fund capacity before debt service and equipment economics catch up. High utilization or customer commitments do not by themselves establish attractive returns after financing costs, power, and GPU replacement. Customer prepayments could bridge the timing gap, but economically they resemble financing: they may secure capacity for Microsoft, Meta, or OpenAI while shifting bargaining power away from CoreWeave and constraining its future pricing flexibility. GPU-backed borrowing adds collateral and obsolescence risk if accelerator values or rental rates fall.

Near term, elevated rates and distressed credit pricing can become self-reinforcing: higher funding costs weaken equity, which can make new equity or unsecured borrowing more dilutive or expensive. The absence of an immediate maturity wall reduces urgency, not the eventual need to fund the buildout. Over 1–3 months, verify customer prepayment terms, committed versus utilization-based revenue, cash interest, and financing availability; a frontier-lab IPO would be a sentiment catalyst, not proof of CoreWeave cash generation. Over 6–18 months, the decisive test is whether operating cash flow covers a meaningful share of capex without repeated expensive financing.

Contrarian point: tight AI capacity can support utilization while still failing to protect provider returns if financing, power, or hardware-refresh costs absorb the economics. The reported yield is not sufficient evidence that the 2031 bonds are cheap without recovery, covenant, and liquidity analysis.

Key Decisions for Investors

• No immediate directional CRWV trade: wait for evidence on cash conversion and funding terms rather than treating high bond yields or an AI-demand narrative as standalone signals.
• Credit watch: consider the 9.75% 2031 bonds only after reviewing price, covenants, collateral priority, debt structure, and realistic recovery assumptions. Reassess if financing terms improve and cash generation begins funding capex; avoid assuming the quoted yield will be realized.
• For the next 1–3 months, track cash interest, contracted backlog conversion, customer prepayments, GPU financing terms, and any equity issuance. A material deterioration in funding access or customer commitments would weaken the recovery case.
• Falsifiers: sustained operating cash flow sufficient to fund a growing share of capex would support the credit thesis; persistent capex far above internally generated cash, rising funding costs, or weaker utilization/pricing would argue against it. Treat any frontier-lab IPO as a sentiment catalyst, not a substitute for those metrics.