CoreWeave borrowing costs fall for data-center financing

An Applied Digital subsidiary raised $1.59 billion in high-yield bonds priced to yield 7% for a fourth building at the Polaris Forge 1 campus that will supply 150 megawatts of IT load under a 15-year contract, reporting by The Next Web and TipRanks shows. The 7% yield is down from the 10% investors demanded for part of the same project in November, and the offering reportedly drew roughly five times its size in demand, according to Bloomberg coverage cited by The Next Web and TipRanks. Credit-default-swap spreads tied to CoreWeave fell to about 4.52 percentage points earlier this month from a 8.81 peak in December 2025, a 49% decline, per The Next Web. Editorial analysis: This package of reported moves fits into a broader repricing of AI infrastructure credit risk, lowering the apparent cost of capital for projects tied to specialized AI cloud providers and narrowing gaps with hyperscaler-linked deals.
What happened
An Applied Digital subsidiary sold $1.59 billion of high-yield bonds priced to yield 7% to finance a fourth building at its Polaris Forge 1 campus in North Dakota, which will provide 150 megawatts of computing capacity under a 15-year contract, reporting by The Next Web and TipRanks shows. The 7% pricing represents a sharp move from the roughly 10% yield investors demanded for an earlier tranche of the same project in November, according to The Next Web and TipRanks. The deal reportedly attracted demand equal to about five times the issuance size, a detail attributed to Bloomberg in both outlets. The Next Web also reports that CoreWeave-linked five-year credit-default-swap spreads fell to about 4.52 percentage points from a 8.81 peak in December 2025.
Technical details
Editorial analysis - technical context: Lower bond yields and tightened CDS spreads both reflect how fixed-income markets are assessing counterparty and project risk, not changes in machine learning models or software. For data-center financing, the spread between a project-linked high-yield bond and a hyperscaler-backed deal is a shorthand for perceived default and revenue stability. Reporting shows the Applied Digital sale now sits within roughly 1 percentage point of Cipher Digital's recent Amazon-linked bond sale, whereas the gap was nearly 2.9 percentage points versus a Cipher deal tied to Alphabet in November, per The Next Web and TipRanks.
Context and significance
Industry context: Public reporting links the repricing to a string of commercial developments and capital moves that market participants cite. The Next Web documents CoreWeave revenue growth of 168% to $5.13 billion in 2025, management guidance for more than $12 billion in 2026, a contracted backlog above $66 billion, and an additional $2 billion investment from Nvidia in January 2026 at $87.20 per share. Those items are reported facts; they serve as inputs market participants use when assessing credit risk. Reporting by The Next Web and TipRanks places the Applied Digital transaction within a larger wave: data-center developers have raised billions of dollars of high-yield financing for projects leased to CoreWeave, according to Bloomberg data cited in coverage.
What to watch
For practitioners: Watch three observable indicators that will clarify whether this is a sustained repricing versus a single strong issuance:
- •secondary-market trading and spreads on existing CoreWeave-linked bonds
- •new issuance coupons for other AI infrastructure projects, including Cipher/ hyperscaler-linked sales
- •confirmed uptake and duration of long-term capacity contracts documented in filings or press releases
Monitoring those metrics will show whether financing costs for specialized AI infrastructure are converging with hyperscaler economics or if the change is issuance-specific.
Scoring Rationale
Lower borrowing costs for CoreWeave-linked projects materially affect build economics for AI data centers and reflect a broader market repricing; the change is notable for infrastructure and financing teams but not a paradigm shift.
Practice interview problems based on real data
1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems

