Analysis
Two numbers from the same week tell two different stories about AI-adjacent investing. CoreWeave's 9.625% notes due 2032 touched a roughly 15% yield this week -- a level Pulse has covered as the market pricing real doubt into the debt behind its $52 billion in combined obligations, not just equity jitters. In the same window, cybersecurity and AI-defense startups raised $6.8 billion in the third quarter, up roughly 50% from Q2 and tied for the highest quarterly total in years, according to Crunchbase News.
Both trades are AI-adjacent. Neither is irrational on its own terms. But they're moving in opposite directions, and that's the story: AI infrastructure financed with debt against long-duration, concentrated contracts is getting repriced downward -- CoreWeave, Firmus Grid's withdrawn IPO, and Nscale's wavering NYSE plans are three data points in four months. Meanwhile, software that sells into AI-driven demand without taking on that leverage -- Cyera's $12 billion valuation and Island's $6.4 billion Series F among the quarter's biggest checks -- keeps clearing bigger rounds at a faster pace.
“One trade's returns depend on a macro assumption holding; the other's depend on a product already being used.”
The mechanism isn't complicated. Compute infrastructure is a bet on customers honoring take-or-pay contracts through a demand downturn that hasn't happened yet; bondholders are starting to ask what happens if it does. Cybersecurity software is a bet that enterprises keep paying subscription fees regardless of whether the AI boom cools, because the attack surface AI agents create already exists, not a projection. One trade's returns depend on a macro assumption holding; the other's depend on a product already being used.
Room for disagreement: it's entirely possible this is a single-company story dressed up as a sector split. CoreWeave's concentration risk -- three customers generating 72% of its revenue -- is specific to CoreWeave's contract structure, not a verdict on debt-financed AI infrastructure broadly, and Nebius and Lambda haven't seen the same bond selloff yet. Cybersecurity funding has also run hot for years independent of AI, so crediting this quarter's number entirely to an AI-adjacent thesis overstates how new the trend actually is.
Nebius and Crusoe's bond spreads are the tell: if they move in sympathy with CoreWeave's over the next month, this was the first real stress test of debt-funded AI infrastructure as a category, not a CoreWeave-specific problem -- and cybersecurity's $6.8 billion quarter will look like the safer bet in hindsight.