Nvidia capex

SpaceX's 5-yr CDS rose to 194bps on Oct 7, 2026, from ~110bps at June inception (FT/Refinitiv) — the fastest-moving credit price in the AI financing web, on the borrower of the largest debt-for-chips raise ($40B, Apollo-led, close expected 2027). Fires if the spread crosses 250bps: that would signal genuine credit distress in the marginal AI borrower and raise the financing cost of the outer edge of the demand web. Distinct from the deal-pricing watch (01a115a0-12ba), which fires on the transaction itself.

FT (Oct 6, 2026): SpaceX seeking ~$10B bank loans + $30B investment-grade debt, led by Apollo with Pimco among lenders in talks, to purchase Nvidia chips (xAI Colossus 2 expanding; Nvidia-exclusive data centers). Transaction expected to close in 2027. Fires when the deal prices/closes or is reported dead. A closing at IG-adjacent spreads institutionalizes the largest debt-for-chips raise yet and extends [[nvidia-infrastructure-leasing-circular-financing]]; a collapse or repricing would signal credit markets balking at the outer edge of the demand web — the sharpest financing-leg bear datapoint since the circular-financing debate began.

Bloomberg (Oct 5, 2026): JPMorgan pitching ~11% yield on $5B leveraged loan for Volta Infrastructure — $3.7B chip tranche buying ~36,000 Nvidia GPUs as collateral + $1.3B LC-collateralized lease payments to Bitdeer (Norway); Anthropic signed a 6-year offtake. Structured as a "strip," excluding CLOs. Lender commitments were due within days of the report (~Oct 14). Fires when the loan prices or is pulled: a pricing at ~11% institutionalizes double-digit marginal AI-debt costs and confirms balance-sheet appetite for GPU-collateralized paper; a pullback would signal the leveraged market is refusing the outer edge of the financing web — the sharpest financing-leg bear datapoint since the circular-financing debate began.

Microsoft reports fiscal Q1 2027 on Oct 27, 2026 — the first quarter carrying the new lease treatment (data-center useful lives 15→25 yrs; future DC leases reclassified from finance to operating leases, which took the calendar-2026 capex guide from ~$190B to ~$175B with spending "unchanged"). Key reads: (1) FCF vs Q4's $19.6B — above suggests the capex/FCF squeeze is stabilizing; (2) Q1 capex vs the guided >$50B; (3) Azure growth vs the ~45% guide and whether "demand exceeds available capacity" still holds; (4) any GPU useful-life change (would fire the Burry depreciation watch).

One month after the Oct 5 NYMEX listing, check whether the H100/B200 rental-index futures attracted real institutional volume/OI — the test of whether "institutional money treats compute as a real commodity" (DataStorage). Also check whether the B200 term curve's below-spot easing is deepening (Rubin volume arriving) or flattening, and whether ICE/Ornn or Nodal lists competing contracts.

FT (Oct 2, 2026): Amazon is seeking to offload about $8B of advanced Nvidia chips — thousands of Grace Blackwell GPUs across more than a dozen US data centers — into an SPV funded by outside investors, keeping use of the hardware under an asset-light model; the vehicle could issue debt and sell up to a 10% equity stake. Talks with investors have been held in recent weeks. Fires when the vehicle launches/prices (or when talks are reported dead). A pricing extends the fifth layer of the circular-financing web ([[nvidia-infrastructure-leasing-circular-financing]]) and institutionalizes hyperscaler GPU asset recycling; a collapse would signal outside investors are refusing GPU residual-value exposure — the same risk the Howden Re insurance talks are meant to lay off.

FT (Sept 29): Nvidia is in early talks with insurers — via broker Howden Re, led by financial-solutions head Ingemar Lanevi — on structures to insure lenders against neocloud defaults and GPU residual-value shortfalls, with possible syndication to hedge funds because deal scale could overwhelm insurer balance sheets. Talks "may not lead to any deals." Fires when the first insurance-backed GPU financing structure or deal is announced (or when the talks are reported dead). This is the next concrete step of the "chips as an investable asset class" push and extends [[nvidia-infrastructure-leasing-circular-financing]].

Man Group (Kristina Hooper, Sept 28) projects the 10-yr reaching 5.5% before year-end and warns surging long-end yields "threaten to topple the two pillars propping up US economic growth: AI capital expenditures and consumer spending." Baseline: 5.17% (FRED, Sept 25), >5.25% intraday Sept 28. Fires if the 10-yr crosses 5.5% — the level at which the marginal cost of the ~$1.1T 2027 AI buildout starts forcing capex re-phasing decisions. Distinct from the Warsh FOMC watch: this is the market-rate tripwire, not the policy path.

Nvidia's Q3 FY2027 earnings (guided date Nov 18, 2026) are the first checkpoint where the RTX Pro 5500 China channel (MIIT signaled approval for ByteDance/Alibaba; ByteDance plans a 1M-unit order) and H200 flows could show up as restored China revenue or a guidance change. Current guidance: zero China DC revenue. Fires on the earnings report: check whether China revenue/guidance changes and whether the RTX Pro 5500 channel is quantified.

GMO's quarterly letter ("A Catalyst for the AI Bubble Break") argues equity supply, not demand, may break the AI bubble: the final ~$2T tranche of SpaceX shares exits lockup June 12, 2027, and Anthropic/OpenAI IPOs could add ~5% to investable market cap; GMO's flow work finds each 1% supply increase historically cut market value ~4% over the next year (~20% implied return drag). Fires at/around the June 12, 2027 release: check whether the supply event pressures the broad market and NVDA's multiple, and whether GMO's 4x flow multiplier is holding up. Earlier SpaceX lockup tranches (if any) should also be tracked as they arrive.

New Mexico environment department faces a Nov 23 deadline to decide the air-quality permit for the 2.45 GW Bloom fuel-cell plant at Project Jupiter (Oracle/OpenAI Stargate campus). Oracle has already sent a force majeure notice to developer Stack Infrastructure over power delays; the Energy Transfer gas pipeline is delayed to Feb 1, 2027. A denial or further delay would validate force majeure as a template for power-constrained AI buildouts; approval keeps the 2028 online target plausible.

The Sept 24 Trump-Xi summit's only tangible outcome was extending the US-China trade-war pause to Jan 10, 2027 (Guardian). Its expiry is the next dated checkpoint for China chip policy: a lapse or new escalation would hit the sentiment around Nvidia's China option (currently guided to zero DC revenue) and the broader sovereign-AI narrative; a further extension extends the frozen status quo.

FT/Tom's Hardware/Sifted (Sept 23-24) report Nscale's Norway data center gave ByteDance's Singaporean subsidiary Spring access to 2,000+ Nvidia B200s — 73% of Nscale's 2025 revenue — "swerving US rules," and that the IPO filing is "largely silent" about its biggest customer. Open questions: (1) Will Commerce/DOJ respond — enforcement, guidance, or new neocloud due-diligence rules? (2) Does the controversy delay or reprice the Nscale IPO? (3) Does it complicate Nvidia's neocloud financing playbook or the NDAA export-control conference? Check next cycle for any official comment, IPO-prospectus amendments, or enforcement action.

CNBC reported (2026-09-16) that investors approached OpenAI about a funding round at a valuation as high as $1.5T, with no active raise underway; OpenAI reportedly burns ~$12.3B/quarter. Eisman (2026-09-21 podcast) called OpenAI the ecosystem's single point of failure: "one of them I think is in trouble, which is OpenAI," with ~$250B of incremental Azure RPO behind Microsoft's capex. Fires when OpenAI's next round officially prices/announces (validating end-demand at ~$1.5T) or when reports of a raise collapse or are postponed (the loudest possible bear datapoint for the capex chain). Distinct from the Anthropic IPO watches.

Nvidia trades under 17x forward earnings (decade low, half its 2025 multiple) as of 2026-09-22 per Bloomberg — the equity market's live verdict on capex durability while the stock sits near highs on price. Track the forward multiple each cycle. A further slide below 14x would signal the market is pricing an outright earnings break (circular-financing unwind, margin collapse, demand cancellation), not just skepticism; a recovery above 20x would signal the skepticism is clearing. Baseline: ~15.75-17x forward as of 2026-09-22.

Nvidia's 10-Q disclosed $279B of supply and capacity commitments as of July 26, 2026, up from $119B the prior quarter (+134% QoQ) — the hardest number in the circular-financing debate (Seeking Alpha, Sept 21-22). Track the figure at each subsequent 10-Q (next: late Nov 2026 for fiscal Q3). Fires if the number exceeds $350B (continued escalation of the vendor-financing web); a decline below $250B would instead signal pullbacks/cancellations, which cuts the other way — either direction is a material datapoint for the capex-sustainability verdict.

Politico (2026-09-21): the Senate NDAA draft would require tracking of exported chips, allow Congress to block advanced-AI-chip exports to China, and impose new chipmaking-equipment controls; House-Senate negotiations expected after the November midterms, enactment deadline end of 2026. Nvidia is lobbying to strip the provisions. Direct China revenue at stake is small (<1% of DC revenue per Nvidia's earnings call), but the outcome sets the regulatory template for the export leg of the sovereign-AI buildout narrative. Trump's alignment with Huang is a backstop but "it's not a bill he can veto."

Anthropic has confidentially filed and chosen Nasdaq; what remains is the public S-1 release — the first time public investors see audited financials (claimed >$100B annualized run-rate, ~$42B 2025 losses, second straight profitable quarter per FT). Hyperliquid pre-IPO perps (~$2.15T implied) trade 24/7 and react first. Fires when the S-1 is publicly released; key reads: revenue trajectory vs the $100B claim, capex/commitment schedule, and how the risk factors handle Amodei's own 'We Must Pace the Frontier' thesis. Distinct from the IPO-pricing watch (01a0b3c7-b060) — this fires earlier.

Apollo chief economist Torsten Slok reports 5-yr CDS on an AMZN/GOOGL/MSFT/ORCL basket climbed above 100bps as of mid-Sept 2026 — the highest in his 8-year series — vs bank CDS flat near 40bps, and attributes the widening to a debt-financed spending cycle with rising leverage, negative FCF, and uncertain payback (not dealer hedging). This is the cleanest market-priced signal of stress in the buyers' credit, the key vulnerability behind Nvidia's otherwise-intact capex story. Baseline: >100bps (8-yr high) as of 2026-09-16.

Anthropic targets an October 2026 Nasdaq IPO raising up to $100B at ~$2T valuation (Reuters/Business Insider), expected to complete before the November midterms. The largest IPO in history and the hardest test of public-market appetite for AI capex economics. A strong pricing supports the demand-financing loop behind Nvidia's capex story; a cut price, pulled deal, or undersubscribed book would be the single loudest bear datapoint. Distinct from the existing watch on Nvidia's $10B anchor-participation talks (01a0a97b-245b), which tracks whether Nvidia confirms/denies. Anthropic revenue run-rate crossed $65B by end-July 2026, up from ~$9B at end-2025.

Burry estimates ~$176B of understated depreciation and overstated profit across hyperscalers 2026-2028 (Oracle ~27%, Meta ~21% overstated by 2028) because GPUs are depreciated over 5-6 years vs a claimed 2-3 year economic life. Fires if any major hyperscaler (MSFT, AMZN, GOOGL, META, ORCL, CRWV) shortens GPU/server useful-life assumptions or takes GPU-related impairments in an earnings report — which would validate faster obsolescence and compress reported AI earnings industry-wide without any order cancellation. Counter-evidence to monitor: H100 rental rates (now $3.28/hr, +22% m/m), CoreWeave's ~95% contract-rebooking rate, A100 rentals contracted through 2029.

The Sept 16 dot plot shows 16 of 18 members expect at least one more 25bp hike by end-2026 (4.00-4.25% by end-2026 and 2027; ~90% CME FedWatch odds). Each subsequent FOMC decision (next expected late Oct/Dec 2026) is a live check on the marginal financing cost of the ~$1.1T 2027 AI buildout and on long-duration semiconductor multiples. A hold or pivot would relieve the pressure; the signaled follow-up hike extends it.

Reuters reported (2026-09-11) that Anthropic is in talks to bring Nvidia in as an anchor investor ahead of a possible IPO, with one source pegging the potential commitment at as much as $10 billion. Talks are non-binding. Fires when either company officially confirms or denies the investment, or when terms are disclosed. Would extend Nvidia's role as financier of the AI ecosystem and enlarge its antitrust surface area alongside the DOJ Groq probe.

The DOJ has sent Nvidia a formal request for information over whether the $17B Groq non-exclusive license plus executive hiring (Jonathan Ross, Sunny Madra) was structured to evade Hart-Scott-Rodino antitrust review. Fires when the probe escalates to formal enforcement (suit, settlement, fine, imposed conditions) or is publicly closed/resolved. Outcome sets the template for DOJ's approach to Nvidia's broader expansion playbook, including the potential Anthropic anchor investment.

Track whether any of the four major hyperscalers (Microsoft, Alphabet, Meta, Amazon) guides down its capital expenditures or delays its cluster deployment schedules in upcoming Q3 2026 earnings reports, confirming whether the 'Pacing the Frontier' debate has translated into actual hardware demand reductions.

Track whether Nvidia's gross margin bottoms out in the 71-72% range in Q4 FY2027 as guided, or drops further due to rising memory and packaging costs.

Track when the first individual project-level deal is officially closed and funded under the $500 billion compute financing platform announced by Nvidia and its Wall Street partners.

Track whether NVIDIA's official commercial launch of the Rubin Ultra features a downgraded memory configuration (e.g. 192 GB or standard HBM4) due to the HBM shortage.

Track whether Figure and Nscale commence the initial deployment of up to 100,000 NVIDIA Vera Rubin GPUs in Barstow, Texas, starting in the second half of 2027.

Track when Nvidia officially closes its $12.9 billion acquisition of Hugging Face, including any antitrust or regulatory reviews.

Track whether Nvidia achieves or exceeds its guided ~70% year-over-year revenue growth for the full fiscal year 2028.

Track whether the Vera Rubin platform achieves the guided ~20% of total data center revenue during Q3 FY2027.

Track the first production-scale deployment or commercial workload execution of OpenAI's in-house developed "Jalapeno" ASIC chip in its active training or inference clusters.

Track the resolution of Oracle's six-month gas pipeline delay and the official power tie-up of its planned AI data center facility to unlock its $638 billion backlog.

Track when Lancium achieves its key grid milestones to unlock the remaining $1 billion of Nvidia's $3 billion investment, or when the first gigawatt of capacity is officially deployed under their strategic partnership.

Track when Nvidia and its Wall Street partners (Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR) execute the definitive agreements for the $500 billion compute financing platforms, following the MOUs signed on August 10, 2026.

Track whether the Ratepayer Protection Act passes the U.S. House of Representatives or Senate, imposing federal regulations on AI data center power allocation and ratepayer protections.

Track when SpaceX launches its first prototype Starmind AI1 satellite into orbit, carrying Nvidia's Vera Rubin payload, to begin space-based AI compute testing.

Track whether Nvidia exercises its pre-funded warrants in Nebius Group (NBIS) on or after September 11, 2026, which would convert its passive beneficial ownership of 21 million warrant-backed shares into active common shares.

Track whether other U.S. states (such as Virginia, Georgia, Ohio, or Texas) or federal regulatory agencies impose similar moratoriums or restrictive grid/power regulations on large AI data center construction, following New York's July 14, 2026 moratorium.

Track the official commercial launch of AMD's Instinct MI500 GPU with HBM4E in 2027 and its first head-to-head performance benchmarks against Nvidia's revised dual-die Rubin Ultra. This will confirm whether AMD has successfully narrowed the performance gap.

Track the official launch of the first GPU compute futures contracts on the Intercontinental Exchange (ICE) utilizing the Ornn Compute Price Index. This is a critical milestone for the financialization of compute capacity.

Track whether Micron's quarterly capital expenditures in FY2027 exceed $10 billion as guided, confirming their aggressive capacity expansion program for HBM4/HBM4E.

Track the first commercial deployment of the Nvidia-LG co-developed smart factory standard, integrating Nvidia Isaac/Cosmos/Omniverse platforms into LG's production facilities. This is a key milestone for the physical AI/robotics capex narrative.

Track the commencement of Phase 1 construction for Kazakhstan's $10 billion "Data Center Valley" project in Ekibastuz, which is scheduled to begin in 2027 and deliver 100 MW to 125 MW of capacity.

Track Helix Digital Infrastructure's project announcements, particularly its first data center construction starts and power tie-ups with Vistra. This joint venture is a key indicator of how private capital and alternative power providers are bypassing utility grid bottlenecks to keep the AI capex boom moving.

Track TSMC's monthly revenue year-over-year growth rate. A drop below 20% YoY would signal that the physical semiconductor supply chain is finally catching up with demand or that hyperscaler chip orders are starting to soften.

Track the hourly rental price of Nvidia's Blackwell GPUs (B200/B300) on the Ornn Compute Price Index. A drop below $3.00/hour would indicate that the current supply crunch is easing or that demand for agentic AI compute is slowing down.

Track regulatory action, SEC inquiries, or accounting standard changes targeting the use of off-balance-sheet Special Purpose Vehicles (SPVs) to finance AI hardware leases (e.g. Anthropic's $35B deal). If regulators clamp down on these structures, it could restrict alternative private-credit funding for the AI capex boom.

Track Nvidia's standalone Vera CPU revenue ramp against its initial $20 billion guidance for the fiscal year. This marks Nvidia's direct entry into the non-x86 standalone server CPU market.