Analysis
Three separate financing deals for AI infrastructure priced within the past week, and the direction is the same every time: junior-tranche debt gets more expensive. Project Big Sky priced its junior slice at 8.5%. Anthropic's Project North Star, where Blackstone funds anchor the junior tranche, priced that tranche above 10% (see the byTheNumbers box below for the exact deal sizes). CoreWeave's bonds are now trading near 15% in the secondary market, according to Value Add Pulse's own reporting this week.
That is not a rounding error. It is roughly a 75% increase in the cost of junior-tranche capital across three deals signed within months of each other, all backing the same thesis: that AI compute demand justifies locking up gigawatts of power and chips for years at a time.
What the barbell actually looks like
Pulse's own funding tracker shows why "AI financing is getting more expensive" undersells the shift. The tracker has logged $196.2 billion across 521 disclosed deals in 2026, but the median disclosed round is just $110 million. A handful of debt-backed infrastructure mega-deals -- North Star, Big Sky, CoreWeave's balance sheet -- sit at one end. A long tail of ordinary seed and Series A rounds sits at the other. Very little capital is landing in the middle, where a company would be big enough to need serious infrastructure but too early to access investment-grade-adjacent debt pricing.
Firmus Grid is the clearest casualty of that gap: it withdrew its roughly $30 billion data-center IPO after investors balked, despite having delivered only a fraction of its planned capacity. A company that size needed either cheap debt or a receptive IPO market to keep building, and got neither this month.
However, rising yields are not proof the AI capex story is broken -- they are proof lenders are finally pricing the risk that building for an assumed level of demand is not the same as having contracted revenue to match it. Blackstone is still underwriting billions into Project North Star at those higher yields, which means credit investors are still willing to fund the trade, just no longer at last year's pricing.
The next wave of chip-financing deals will clear somewhere between CoreWeave's roughly 15% print and Big Sky's 8.5% -- that spread, not any single yield in isolation, is the number that tells you whether credit markets still believe the AI capex story. LPs who think they've diversified away from AI-equity risk by also holding infrastructure-debt funds should check whether those funds are underwriting the same demand assumptions through a different legal wrapper.