The 2026s traded at 87.5 yday and now will get taken out at par in 4 months.
The 2029s trading 70c today.
This shit isn’t that hard gents
No one talks about what happens if AI eventually not “matrix multiplication” shaped
one release could theoretically wreck GPU, HBM, advanced packaging, and low nm production
Examples: energy based models, event-based and sparse, vector-symbolic;
The irony if LLMs design these
There are 2 types of distressed credit trades
Exogenous binaries: things that reduce to A or B based on an exogenous factor (eg a court ruling, a used car pricing path)
Endogenous complexities: complex corporate structures, LME dynamics, other barriers
The latter rewards effort
In many ways, investing is fundamentally changing from AI
Changes the archetype of the “good investor”
From
processing vast info, access to obscure info, pro modeling, pattern recognition via analogy, etc
To
Asking the right questions, knowing what matters, conviction
1/ GPU clouds buy hardware, borrow against it, and lease it out. So the hard question isn't AI demand — it's what a cluster earns after its first contract ends. A GPU-hour isn't a unit of anything. Work is — tokens trained, tokens served….
6/…There's a second clock, and it runs on watts. Datacenters are power-limited, so an old cluster competes with what new chips would earn in its rack. The more profitable new silicon is, the sooner replacement pays. The same scarcity that lifts capture also pressures life…
No ill will against Leopold but alas they did blow up
cnbc.com/amp/2026/07/30/leop…
gg
1/
AMC 15% 1.5L @ 107 isn’t paying you for the risk. ~7% expected return on our scenario weights, basically the CCC index. But it’s a single-factor bet: everything rides on the ’27-’28 slate. Upside is capped, and 6 of the 15 pts are PIK that just accretes into the claim that gets restructured in the bad case.
3/
Liquidity is a dependency chain, not a balance sheet. No revolver. ~$500M cash against a ~$590M annual fixed nut. Any sub-$9.5B box office year gets funded by retail equity issuance, which dries up exactly when the slate disappoints. And $3.3B still comes due in one window, Jan-Feb ’29.