I was circling Atlanta in a holding pattern, grounded by weather, when the WSJ story dropped this morning. OpenAI missed key internal revenue and user targets ahead of its highly anticipated Q4 IPO. Oracle fell roughly 7-8%, SoftBank had its worst day in six months, CoreWeave fell, and NVIDIA, AMD, and Broadcom were all off. A private company's internal miss repriced billions in public market value in real time.
That's not a coincidence. That's a signal.
I entered the workforce during the financial crisis, which gave me a front-row seat to how quickly interconnectedness becomes contagion. Lehman's failure exposed how dense and opaque the connections already were. The warning was visible in the structure long before anything broke.
I'm not saying AI is 2008. I am saying the structure deserves scrutiny.
The AI capital stack is more layered than most people realize. At one end, you have private AI labs with massive compute commitments, revenue assumptions, and equity structures. At the other end, you have retail investors holding target-date funds in their 401(k)s. The distance between those two things is shrinking.
Here's what the data shows. AI-linked companies now account for roughly $1.2T of US investment-grade corporate debt, about 14% of the market, up from 11.5% in 2020. Tech debt issuance hit record highs in 2025 and represented about 16.7% of global non-financial corporate bond issuance, with AI infrastructure and refinancing as primary drivers. In private credit, BIS data from March 2026 shows outstanding loans to SaaS firms grew from $8B in 2015 to over $500B by end-2025. One-third of private credit funds now lend to SaaS companies, a segment increasingly entangled with AI adoption and disruption risk.
Meanwhile, a $16B AI data center campus in Michigan tied to OpenAI, with Oracle as anchor tenant, is being financed with roughly $14B of debt structured as 144A investment-grade securities. That means it ends up in bond benchmarks like the Bloomberg Global Aggregate. Which means it ends up in PIMCO funds. Which means it ends up in retirement accounts. The private AI build-out and the retail fixed-income stack are now touching.
The OpenAI-Oracle-CoreWeave-Nvidia loop makes it more concentrated. Oracle commits data center capacity for OpenAI. OpenAI holds equity in CoreWeave. Chipmakers benefit through equipment demand. Each relationship makes strategic and commercial sense in isolation. Together, they create a network where a demand shock at one node propagates through equity, debt, and contract exposure simultaneously.
This is what systemic risk looks like before it becomes a crisis. Not one catastrophic event. A local shock that propagates through common exposures and layered leverage.
To be precise about what I'm claiming: the conditions for systemic contagion exist in the AI capital stack today. Size, concentration, cross-asset embedding, and rapid transmission from private companies to public markets are all present. That's different from saying a crisis is inevitable or imminent.
But this morning was instructive. A private company's miss moved public equities instantly. The same underlying AI assumptions are now embedded across private credit funds, investment-grade debt, and retail fixed-income products. The structure is denser than the market is pricing.
In 2008, the warning signs were hiding in the plumbing of the global financial system. In retrospect, the system didn't have a good map of its own interconnections.
The question isn't whether the connections exist. It's whether anyone is paying attention.