The Financial Times (McMorrow, Chan and Taffe — known here only via an echo; the original article has not been inspected) reports that Big Tech is expanding guarantees, notably residual-value guarantees behind SPV financing of AI data centres and chips, to keep roughly $300B of AI exposure off its balance sheets. The supplied snippets do not reproduce the FT's $300B figure, but the mechanism is independently corroborated at larger scale: the BIS's March 2026 Quarterly Review coins 'shadow borrowing' for >$120B of AI compute debt moved off hyperscaler balance sheets in ~18 months via SPVs, leases and guarantees; a Nikkei Asia study of SEC filings tallies ~$1.65T of off-balance-sheet AI obligations at Alphabet, Microsoft, Amazon, Meta and Oracle versus $1.35T of reported debt; and Morgan Stanley counts >$3.1T in off-balance-sheet commitments, >$2.7T of it hyperscaler guarantees alone. Estimates diverge by more than an order of magnitude depending on what is counted (guarantees only versus leases, GPU procurement and construction commitments — all footnote-disclosed rather than hidden), S&P has reportedly downgraded Oracle to BBB- citing AI-commitment leverage, and an independent Wharton/MIT Technology Review analysis (~$1.1T of hyperscaler capex through 2027 requiring a ~2.7x productivity gain by 2030) frames the demand side the guarantees ultimately rest on.
Independently converges with the AI-capex-bubble / funding-cycle downside his corpus already governs: BIS 'shadow borrowing', the Nikkei SEC-filings tally, Morgan Stanley's commitment count and S&P's Oracle downgrade are dated receipts that the leverage premise of that scenario is firming, extending the bust path with a credit-contagion channel (guarantees → SPVs → private credit) that his Re-Roll capital analysis and Fog option governance should now carry. Residual-value guarantees also mean someone is underwriting GPU depreciation curves — the same fast-deflating GPU economics his radar tracks via cost futures and price-decline claims — giving citable board-facing material, though no supplier, pricing or operating consequence for Scott is established and velocity is near zero, so this is medium scenario-feed, not an alert.
ip:framework.the-re-rollip:framework.fog-is-a-race-between-two-clocksdev:project.strategyradar:concept.ai-financingradar:concept.compute-financingradar:nvidia-500b-ai-infrastructure-financingradar:cme-gpu-cost-futuresradar:epoch-price-of-thought
queries asked of Scott's wikis
- AI capex bubble compute overbuild oversupply position
- local inference economics GPU cost decline cheap compute
- frontier API dependency vendor risk agent harness token costs
- AI winter downturn funding-cycle scenario planning
- self-hosting versus cloud frontier AI sovereignty enterprise advice
- open-weights releases under financial pressure distressed labs
now 0 pts/hpeak 8 pts/hcomments 0/hpeers p50momentum: steady3 platformsage 557h
points/hour across evidence · reading as of 2026-10-13 05:35:48.225472+11:00 · deterministic, not a model opinion
2026-10-02T13:30:47Z
The velocity wake traces to the already-attached Reuters credit-market story accruing modest engagement (~3 pts/h against a floor baseline, now 25/9) plus the stale magnitude-valve flag from the adjudicated Nvidia essay — no new evidence and no change of meaning from the 2026-10-01 reprice ('financing market beginning to price the risk'). Case stays priced as is, armed on force-majeure spread, measurable financing repricing/failed issuance, or Anthropic-item traction.
2026-10-01T07:26:50Z
The Reuters credit-market story adds the case's first market-pricing datapoint — lenders resisting AI borrowers, with a linked KKR warning on AI-borrowing credit risk — shifting the meaning from 'contingent exposure is disclosed and modeled' toward 'the financing market is beginning to price the risk' along the guarantees→SPVs→private-credit channel. The item itself is cold and the magnitude-valve flag is stale residue of the adjudicated Nvidia essay, so heat holds low; material_change keeps the case armed for the next financing repricing or force-majeure event.
2026-10-01T07:24:07Z
evidence attached: hn.story.49918147 — Independent Reuters reporting that credit markets are balking at AI borrowers materially contextualises the AI-infrastructure financing-risk case.
2026-09-29T23:21:26Z
Reuters' quantification of Anthropic's $518B largely non-cancellable buildout is the substantive-periphery expansion this case was waiting on: the commitment web now demonstrably spans the lab demand side, so the revenue anchor that guarantee-backed SPV financing rests on is itself contractual and non-cancellable — two-sided contingent exposure, not just hyperscaler guarantees. The item is nonetheless cold (HN 3/0), every other on-hypothesis object has been static ~5 days, and the magnitude-valve flag is stale residue from the adjudicated 1993 Nvidia-essay spike, so heat holds low; material_change is set to keep the case off the quiet clock and armed for the next force-majeure or financing-disclosure event.
2026-09-29T20:53:13Z
evidence attached: hn.story.49898457 — Reuters quantifies Anthropic's $518B largely non-cancellable buildout, directly bearing on the guarantee-backed AI financing-risk case.
2026-09-29T16:53:44Z
Third re-fire of the same adjudicated noise: the 1993 Nvidia founder-equity essay peaked at 1076/451 and has decayed to ~0 pts/h (25th percentile), so the magnitude-valve spread reading traces to one tangential object's past peak, not hypothesis-relevant cross-platform growth — every on-hypothesis thread (FT, Oracle force majeure, Wharton capex) is static and no new financing-disclosure or force-majeure event appeared. Case meaning unchanged; cooled to low because the substantive periphery is not expanding and the sole fast mover is spent.
2026-09-28T06:42:46Z
The velocity spike re-fires from the same already-adjudicated tangential object — the 1993 Nvidia founder-equity essay, now 401 pts/175 comments — while every on-hypothesis thread (FT, Oracle force majeure, Wharton capex) is static, so the case's meaning is unchanged. Heat held at medium despite the 99th-percentile reading because the fast mover is sensor noise; the live follow-up trigger remains expansion of substantive periphery (further force-majeure or financing-disclosure events), not engagement.
2026-09-28T03:35:54Z
The newly attached 'Owed a billion dollars in Nvidia stock' (HN 145/59) is a 1993 founder-era personal equity dispute (NV1-era options, statute of limitations) with no bearing on off-balance-sheet financing — it adds nothing to the hypothesis and is inflating measured velocity (99th-percentile object) while the hypothesis-relevant threads (Oracle force majeure, Wharton capex) run steady. Case meaning is otherwise unchanged: corroborated guarantee/SPV mechanism, one live risk-shifting incident, FT figure still unverified.
2026-09-28T03:25:10Z
evidence attached: hn.story.49872723 — Front-page HN account (118 points) of a $1B Nvidia-stock obligation bears on hidden AI-infrastructure financing exposure if the deal structure checks out.
2026-09-24T18:08:06Z
Oracle invoking force majeure on a contested data center converts the case from analyst estimates of off-balance-sheet exposure into a live instance of contractual risk-shifting at the most-levered hyperscaler (S&P BBB-, cited in-thread), corroborating the exact guarantee/SPV mechanism the FT described. Combined with grounding-pass corroboration (BIS shadow-borrowing, Nikkei SEC-filings tally, Morgan Stanley commitments), the case graduates from watching to corroborated, while the FT's specific $300B figure remains uninspected and the concealment framing stays qualified by rating-agency modeling.
2026-09-24T16:38:17Z
evidence attached: hn.story.49830037 — Oracle invoking force majeure on a contested data center is concrete evidence of contractual risk-shifting in the AI infrastructure buildout the case tracks.
2026-09-24T09:46:40Z
grounded: converges/medium — Independently converges with the AI-capex-bubble / funding-cycle downside his corpus already governs: BIS 'shadow borrowing', the Nikkei SEC-filings tally, Morg
2026-09-24T09:37:27Z
The independent Wharton/MIT Technology Review analysis (~$1.1T hyperscaler capex needing a 2.7x productivity gain by 2030) adds a genuinely independent financial-analysis voice to the AI-infrastructure-financing-risk theme, moving the case past pure single-headline testimony; but it addresses aggregate capex sustainability, not the FT's specific $300B off-balance-sheet guarantee claim, which remains unverified, and the episode itself has cooled to near-zero velocity (0.17 pts/h vs 14.8 peak).
2026-09-24T09:25:19Z
evidence attached: reddit.post.1wowoyc — Independent Wharton/MIT-TR analysis that ~$1.1T of hyperscaler capex needs a 2.7x productivity gain materially corroborates the infrastructure-financing-risk case.
2026-09-20T16:23:54Z
The discussion raises an important distinction between off-balance-sheet exposure and concealed or unpriced risk: rating agencies may already account for these guarantees. That qualification remains unverified and does not refute the contingent-exposure claim; no new evidence establishes its scale or consequences for Scott.
2026-09-20T13:39:05Z
grounded: novel/low — No meaningful Scott-specific intersection is established: the hits address delivery underwriting, software control and inference costs, not hyperscaler continge
2026-09-20T13:36:07Z
origin walked (codex/luna, conf 0.93): anchor hn.story.49774991 -> echo.other.d9e220dea7 by Ryan McMorrow, Michelle Chan, and Michael Taffe (Financial Times)
2026-09-20T13:34:53Z
case created — The report identifies a concrete financing mechanism and material exposure estimate, but the supplied evidence contains only its headline.