“Backstop deals sink stocks” is wrong: the guarantor falls, the guaranteed side rises — resolves 2026-12-31

The debate: a big chipmaker puts money into its own customers, who then use that money to buy its chips (the industry calls this circular financing), and some say the market now reads every such announcement as bad news.

Why it matters: if that were true, any company announcing such a deal should see its stock fall, a signal investors and suppliers could act on directly.

Who is on each side: analyst Marcus and those who agree say the market's reaction to these announcements has flipped completely; we split the reaction by side and found that only one side flipped.

Marcus's claim that circular-financing announcements flipped the market's pricing function does not hold as stated. It confuses positions: on the same day the guarantor fell 4.99% while the guaranteed party rose pre-market, and on July 21 the recipient still rose sharply. It has two counterexamples: deals collapsing and deals stalling both read as bad news, so the market punishes a backstop that fails to happen rather than the backstop itself. And its anchor number is wrong: Oracle closed at $328.33, not $307. What does hold is a layered claim. The guarantor's announcement premium decayed to zero and turned negative from September 2025. The recipient's break point sits at the earnings layer, where Oracle beat across the board and still fell 10%. Structural risk is priced mainly in credit, with Oracle's bond market leading by about nine months on a single-source retrospective, and Nvidia's 5-year CDS at a record 82bp pricing the whole $750B package.

Substack / Marcus on AI
“Oracle stock briefly went up by 43% (and…”

BG2 Pod (Brad Gerstner + Bill Gurley)
“I think it increases the chance that we go over…”

SemiAnalysis (Dylan Patel)
“Nvidia provides a take-or-pay commitment to…”

Four things to watch: (1) if the $250B backstop is signed, the closing reactions on both sides on signing day, Nvidia and the benefiting neocloud, are the clean test of position-dependent gains versus a full flip; (2) Nvidia's 5-year CDS falling back to around 40bp before the end of 2026 makes the credit-layer alarm a false one and downgrades this piece, while holding at 80bp or above, or setting new highs, counts in its favor; (3) the next billion-dollar-scale recipient announcement falling rather than rising, with no sector-wide decline available to explain it, establishes Marcus's full-flip version and retires the position-dependent reading; (4) if a formal event study that strips out market and sector moves overturns the direction of the guarantor premium's decay, this piece's position-alignment table, which uses raw returns, is subordinate to it.

What would overturn it

If a formal event study stripping out market and sector moves shows the guarantor premium has not decayed, or recipient announcements keep earning positive returns with no sector-wide drop to explain them, the position-split reading flips to the full-flip version.

Read the full deep dive →

Dec 31, 2026

Near-term checks: whether the $250B backstop gets signed (in negotiation, no timeline); weekly readings of NVIDIA's CDS; Van Nieuwerburgh's paper “Financing the AI Buildout” (the most likely academic second source); a primary-source check of S&P's cut of Oracle to BBB-; a recheck of the guaranteed party's July 27 closing data