Wednesday, October 21, 2026

@ACapitalLP Oct 18 51 clicks

A Capital on X

𝙒𝙝𝙖𝙩 𝙝𝙖𝙥𝙥𝙚𝙣𝙚𝙙 𝙩𝙤 𝙩𝙝𝙚 𝙉𝙚𝙤𝙘𝙡𝙤𝙪𝙙 / 𝘼𝙄/𝙃𝙋𝘾 𝙩𝙧𝙖𝙙𝙚 𝙡𝙖𝙨𝙩 𝙬𝙚𝙚𝙠: I like distilling complex topics down to plain English. Here’s my understanding of what happened. Please correct where wrong… $CRWV and its peers rent $NVDA GPUs to AI labs and hyperscalers under multi-year take-or-pay contracts. The capex is so large relative to cash flow that the model only works if they can keep raising tens of billions a year in new financing to build what customers have already signed for. Much of that comes from secured loans backed by specific chips and contracts, while the rest comes from unsecured bonds, converts or stock. Importantly, not all secured debt is non-recourse, and many of $CRWV’s GPU loans are guaranteed by the parent. Still, the secured facilities held up considerably better than the unsecured bonds, which went from ~12% to ~15% yields, while the cost of insuring them via CDS roughly doubled… Nothing fundamental appears to have caused it. No major customer walked, GPU pricing held up, and the quarter looks fine. What happened was a cluster of headlines landing on a nervous market with the 10-year near 5.2% and hundreds of billions of AI debt issuance expected: Chamath calling it a private-credit bubble, an analyst saying the equity could go to zero, Gartner forecasting that over 80% of neoclouds stop operating independently by 2030, a few smaller peers failing to get IPOs and debt deals done. Underneath sat the bigger question of whether neoclouds are durable businesses or really just financing wrappers that hyperscalers eventually insource and cheap open-source models undercut. Once investors doubt the parent can fund growth, the stock falls, which makes the equity leg more dilutive, which worries bondholders more. Credit and equity fed on each other and the whole group fell 10-30% in sympathy. Was it justified? In three parts. (1) The $CRWV unsecured bond move was likely overdone: a company with ~$17bn of liquidity and a $129bn contracted backlog doesn’t suddenly become a distressed credit because of a handful of headlines. But backlog isn’t cash flow, and secured lenders have first claim on specific assets and cash flows, with many loans also carrying parent guarantees. The question for unsecured creditors is how much cash actually makes it back to the parent after servicing the secured debt. Still, a move to ~15% yields looks more like aggressive HF de-risking with headlines as the accelerant than a sudden collapse in the underlying business. (2) The $CRWV equity move was more understandable, just not for the headline reasons: this is a levered financing business whose equity returns depend on the cost of the marginal dollar of capital. When financing gets more expensive, the residual value on every future GW shrinks, and any gap that needs to be filled with equity becomes more dilutive. That said, unsecured bonds trading at 15% doesn’t mean every new GPU deployment costs 15% to finance. A 20% haircut for a stock that may not self-fund until 2030 could be seen as fair, but it ultimately depends on how much parent-level capital is actually required. (3) The group ($CORZ, $CIFR, $WULF) selloff was the least justified: $CRWV finances depreciating GPUs and bears the risk of obsolescence and replacement. The miners converting to HPC are increasingly financing long-lived power and data center infrastructure leased to hyperscalers under 15-20 year contracts, often with the customer owning the GPUs. Their risks are construction, power delivery and financing, not GPU replacement or compute pricing. Different assets, different contracts, different capital structures. Yet the market sold them as though they had the same problem, and that’s where the mispricing most likely sits…