Blackstone’s Jas Khaira says fears over AI-buildout debt are overblown

Blackstone is financing AI infrastructure and companies through lending, equity investments, and AI adoption initiatives, including a $5 billion commitment to Crux AI and a $1.5 billion joint venture with Anthropic and Hellman & Friedman. Risks are rising: CoreWeave’s debt has reached about $50 billion, its shares fell over 20% in recent days, and Firmus cancelled its planned IPO at a target valuation of roughly $30 billion as interest rates and credit spreads increased. Blackstone’s Jas Khaira said demand for compute remains strong, described AI returns across portfolio companies as uneven, and argued that hyperscaler-backed data-center debt does not resemble a broad credit crisis.

Analysis

The key risk is not simply AI demand; it is whether financed compute earns out before the collateral loses value. GPU-backed lending depends on utilization, contract enforceability and resale value. If next-generation chips shorten the useful economic life of current hardware, collateral coverage can weaken just as refinancing costs rise. That creates a feedback loop: wider spreads pressure neocloud equity and private marks, while tighter lending can delay capacity orders. Nvidia is therefore exposed to a second-order financing channel even if end-user demand remains strong. Conversely, Google’s TPU commercialization could broaden supply and reduce dependence on Nvidia, but only if customers can use the ecosystem at scale.

Blackstone has potential upside from arranging and deploying capital, but its AI exposure is not a clean single-direction bet: credit, equity and portfolio-company adoption may be exposed to the same utilization shortfall. The claimed ROI from adoption is uneven, so do not capitalize services activity as proven productivity gains. Over 1–3 months, watch credit spreads, refinancing access and neocloud utilization; over 6–18 months, enterprise adoption and GPU economics determine whether today’s buildout is productive capacity or stranded collateral. The canceled Firmus IPO is a valuation signal, not proof of a systemic credit event. The broader risk case is weaker if obligations are ultimately supported by investment-grade customers, but customer quality alone does not eliminate duration, concentration or collateral risk.

Key Decisions for Investors

• Relative-value bias: favor MSFT or META over CRWV if AI infrastructure credit spreads continue widening. The thesis is lower direct exposure to leveraged neocloud financing, not that hyperscaler capex is immune. Reassess if CRWV demonstrates durable utilization, refinancing access and contract coverage; avoid treating a short as a directional call on AI demand.
• Keep BX as a watch rather than a direct AI long on this evidence. Verify fund-level exposure, realized credit performance, fee contribution and valuation sensitivity before underwriting upside; correlated credit and equity exposure could offset deployment benefits.
• Track CRWV refinancing terms, debt maturities, customer concentration and GPU resale values alongside NVDA order commentary. Worsening spreads plus weaker utilization or canceled capacity plans would strengthen the downside case for neoclouds and could eventually pressure accelerator demand; stable spreads and sustained customer commitments would falsify it.
• Treat Google’s TPU/Crux AI activity as a medium-term competitive watch, not yet a basis to short NVDA. Evidence of scaled workloads, customer portability and attractive economics would support a substitution thesis; without it, alternative compute may expand total capacity rather than displace Nvidia.