Goldman now expects hyperscalers to issue $420 billion of debt in 2027, with the broader AI capital bill approaching $5 trillion through 2030. At the same time, an AI system hallucinated nuclear weapons components aboard a Chinese vessel and nearly triggered a Special Forces deployment. Credit markets are pricing the first story. They have not priced the second.
The Dark Side Of The Boom is the only desk inside the 12-hour window. ING and MUFG are both outside the gate. So this edition draws entirely from one source, and says so plainly.
That source ran two connected stories today. The first: Goldman now expects hyperscalers to issue $420 billion of debt in 2027, with Bank of America putting the broader AI capital requirement near $5 trillion through 2030. The second: an AI system hallucinated nuclear weapons components aboard a Chinese vessel bound for Iran, nearly putting US Special Forces on board. One story is about money. The other is about what it is actually funding.
The crack is not desk against desk here. It is what the credit market appears to be pricing versus what the AI hallucination story makes visible.
Credit markets are funding $5 trillion of AI infrastructure on the assumption that the reliability question gets solved. The hallucination incident, as Dark Side Of The Boom frames it, suggests that assumption is still an assumption. A system that can fabricate a weapons manifest with enough confidence to trigger a near-military response is not a solved problem. That gap between investor conviction and operational reality is the thing worth watching. Nobody is pricing it.
The AI capital story and the hallucination story feel like separate beats but they are the same story told from two ends. At one end, the biggest companies on earth are borrowing at a pace that will stress credit markets by 2027. At the other end, the outputs of that investment are still capable of producing confident fabrications in high-stakes contexts.
There is a version of this that resolves fine. Reliability improves, the debt gets serviced, the buildout pays off. There is another version where one of these hallucination events lands differently, trust cracks faster than the technology improves, and the $5 trillion capital stack looks a lot less comfortable. The credit market, right now, appears to be pricing only the first version.
China adds a third angle. Dark Side notes that China's economy is no longer moving as one piece, making single headline numbers increasingly misleading. That has implications for any model, AI or otherwise, that ingests Chinese macro data as if it still represents a coherent signal. The economy splitting internally is a data quality problem as much as a growth problem.
Watch whether credit spreads for hyperscalers show any sensitivity to AI reliability news. If bond markets start distinguishing between AI companies with auditable outputs and those without, that is the signal the hallucination risk is being priced. Right now there is no sign of it.