Circular Financing in AI: Who Pays Whom Between Nvidia, OpenAI and Oracle
Circular financing in AI refers to chipmakers putting money into the companies that buy their chips, through equity stakes, loan guarantees and promises to buy back unused computing capacity. The practice has come under closer scrutiny since Nvidia published expanded disclosures with its November 17 earnings, days after Jensen Huang resigned as chief executive and while Anthropic was preparing an IPO in which its own suppliers bought shares. The loops are large, and they connect Nvidia, OpenAI, Anthropic, Oracle and a group of smaller GPU rental companies known as neoclouds. Nvidia's guarantees are conditional and secured on scarce powered buildings, so the nearer risk for Nvidia is that sales to weakly financed neoclouds fall away in a downturn. The most direct vendor-financing exposure sits at Broadcom and Oracle.
How does circular financing in AI work?
In an ordinary sale, a customer pays for equipment with its own cash or with money it borrows. In the AI build-out, the chip supplier often provides part of that money first. It buys shares in an AI lab or a cloud operator, guarantees the operator's debt or leases, or agrees to buy capacity the operator cannot rent out. The customer then signs contracts that turn into chip orders, and the supplier books the sale.
Source1
Nvidia draws most of the accusations, but the practice runs across the supply chain. When Anthropic listed, seven suppliers bought $26 billion of its shares at the IPO price in a private placement, with Nvidia and Google taking $10 billion each, according to Anthropic's S-1 filing4.
AMD gave OpenAI warrants, which are rights to buy stock, for 160 million AMD shares at one cent each. They vest as OpenAI reaches purchase milestones and as AMD's share price reaches targets as high as $6003. TSMC's board approved buying up to $500 million of OpenAI preferred stock3,4.
How much has Nvidia invested in OpenAI and its other customers?
Nvidia has become one of the largest investors in its own customer base. It put $58 billion into public and private companies in the first nine months of its 2027 fiscal year, compared with $5 billion a year earlier, according to its third-quarter disclosures2. Its OpenAI stake replaced a September 2025 letter of intent for up to $100 billion of staged investment1.
| Deal | Type | Amount |
|---|---|---|
| OpenAI | Equity in February 2026 funding round | $30B1 |
| CoreWeave | Equity plus purchase of unsold capacity through 2032 | $2B + $6.3B1 |
| Nebius | Equity | $2B1 |
| xAI | Equity in a $20B round | Up to $2B1 |
| Meta | Shares in October 2026 offering | $3B5 |
| OpenAI's Ohio campus | Guarantee on land, power and buildings | Up to $105B2,3 |
Guarantees are the larger exposure. Nvidia reported maximum guarantees of $118 billion against about $53.5 billion of cash and marketable securities, and it has made no payments on them2. Bank of America estimated a far higher $230 billion of guarantees as of August 20266, and the research does not reconcile the two figures.
Most of the disclosed amount rests on one project. In July 2026 the Wall Street Journal reported talks on a $250 billion backstop for a roughly 10 gigawatt (GW) OpenAI campus in Ohio, and Nvidia shares fell about 5%3.
The deal signed in August covers only the first 4.25 GW. The guarantee switches on building by building once each is financed, built, powered and accepted by Nvidia, with the first expected in fiscal 2029. It shrinks as OpenAI pays rent, and it pays out only if OpenAI defaults and the developer, SoftBank's SB Energy, cannot re-lease or sell the site2,3. Colette Kress, the finance chief who became interim CEO and who reportedly argued against the larger figure, told analysts a guarantee "doesn't exist" if a project stalls3. Nvidia's neocloud commitments fell to $32 billion from $36 billion, and new filing language says they take effect only if the neoclouds secure financing2.
Are AI compute commitments actually binding?
AI labs sign multi-year contracts to rent computing power, and the headline totals are far larger than their revenue. Anthropic, the only lab with detailed public disclosure, lists about $895 billion to $901 billion of commitments against a revenue run-rate of about $81 billion4,5. Trackers put OpenAI at about $1.6 trillion across some 21 deals, but how much of that OpenAI can cancel is essentially undisclosed5, which leaves the larger half of the picture open.
The firm contracts are Broadcom's equipment leases and the deals with Google, Amazon Web Services and Microsoft. Most are take-or-pay, meaning Anthropic owes the money whether or not it uses the capacity5. They tie Anthropic to Google's TPU chips, Amazon's Trainium chips and Broadcom's custom processors.
The Nvidia-powered capacity sits in the conditional group. The SpaceXAI contract comes from xAI, which Nvidia backed and which now sits inside SpaceX, and either side can cancel it on 90 days' notice after early 2027. Most of the Lambda contract is terminable, and Nscale is paid only as capacity is delivered5. Analysts estimate Nvidia's share of contracted frontier-lab compute has fallen to about 39%, with Broadcom at about 24%5.
Nscale shows how two conditional promises can meet on one balance sheet. Nvidia's support depends on Nscale raising money for the build, and Anthropic can leave without liability if that money does not arrive. As of mid-November Nscale had about $5.5 billion of debt and no binding financing for the project4.
Is Nvidia the next Lucent or GE Capital?
Critics point to the telecom bubble, when equipment makers such as Lucent and Nortel lent money to customers who then bought their gear. Lucent's customer credit reached about 24% of its revenue by 2000, and its revenue later fell from about $38 billion to $8 billion6. Nortel went bankrupt. By comparison, Nvidia's $16 billion of equity investments in its third quarter equalled about 14% of that quarter's $115.8 billion revenue2.
George Gilder compares Nvidia's financing platform with Apollo, BlackRock, Brookfield, KKR and others, which aims to mobilize $500 billion, to the way GE Capital grew from customer support into a risk that nearly sank GE6. Nvidia may guarantee up to 25% of deals on that platform, though none had closed by November3,6.
On solvency, the evidence favors Nvidia. Its guarantees start years from now, depend on completion and cover powered land and buildings, which hold value longer than chips. Power is scarce: US large data-center connection queues total about 175 GW, only about 44 GW is expected to be built, and the median wait to connect is about 61 months6.
On revenue quality, the critics have a narrower point. Nvidia's AI Clouds, Industrial and Enterprise segment, which contains the neoclouds, brought in $47.5 billion, up about 195% from a year earlier2. If neoclouds cannot raise money, the damage would appear as lost sales in that segment. Lenders are already charging more: CoreWeave's bond yields reached about 15% and Nebius's passed 12%6.
The clearest vendor-financing exposure may sit at Broadcom. Bank of America, Citi, Morgan Stanley and Blackstone fund a $63 billion special-purpose vehicle that buys Broadcom chips and leases them to Anthropic, and Broadcom backs the resale value of those racks with peak exposure estimated at about $43 billion. Broadcom may also lend Anthropic up to $42 billion through convertible notes4.
Oracle is the other concentrated case. Analysts estimate about 45% of its $664 billion of remaining performance obligations, its contracted backlog, is tied to OpenAI, a lab without a credit rating5.
The case for and against
The bull case
Nvidia produced about $43.8 billion of free cash flow in its third quarter6, so it can absorb losses that would have broken Lucent. Hyperscalers, the strongest customers, took about 61% of the quarter's sequential data-center revenue growth2. Gene Munster estimates Nvidia's financing deals will return about $420 billion for roughly $250 billion put in6. The Ohio guarantee ends if OpenAI earns an investment-grade rating, and the scarcity of grid connections supports the value of the buildings behind it.
The bear case
The Bank for International Settlements found about 30% of AI investment deals by value had AI firms on both sides6, so demand is partly funded by the sellers. Anthropic owes about $70.7 billion in the next 12 months4, and its suppliers' share purchase covers only about a third of that. Lenders assume GPUs last three to four years while hyperscalers depreciate them over longer periods6. The cancellability of Nvidia's supply commitments, mostly memory and larger than its guarantees, is not disclosed.
What to watch
- December 4, 2026Deadline for Nvidia's 10-Q quarterly filing, which should show customer concentration, receivables and how its guarantees reconcile with outside estimates2.
- Early 2027Anthropic's right to cancel the SpaceXAI contract on short notice opens, the first test of the conditional, Nvidia-powered tier of its commitments5.
- Q1 2027Meta begins up to $60 billion of at-the-market share sales, a measure of whether public investors keep funding AI build-outs that its chip suppliers already back5.
- 2027First deals expected under Nvidia's Wall Street financing platform, which would show whether third-party capital replaces Nvidia's own balance sheet3.
- Fiscal 2029The first Ohio building is expected to be ready for service, when Nvidia's guarantee on the OpenAI campus first takes effect2,3.
Get Custom Research Like This
Start Your ResearchSource Research Reports
The full underlying research reports cited throughout this analysis. Tap a report to expand.
Report 1 Map every publicly reported Nvidia equity investment, commitment, or financing arrangement with OpenAI, xAI, Perplexity, Anthropic, and neoclouds (CoreWeave, Nscale, Lambda, Crusoe and others). For each, record the amount, date, structure (equity, warrant, backstop, lease guarantee, capacity buyback), and whether the proceeds are reported or expected to return as Nvidia GPU purchases. Produce a table and a money-flow diagram description, labeling each item as confirmed, reported, or rumored, with sources.
Nvidia has made multiple equity investments, staged commitments, capacity backstops/lease guarantees, and related financing arrangements with frontier AI labs (OpenAI, xAI, Anthropic, Perplexity) and neocloud/GPU cloud providers (CoreWeave, Lambda, Crusoe, Nscale, Nebius, and others). These deepen demand for Nvidia GPUs while providing capital or risk mitigation to customers building infrastructure. Many involve explicit or implicit expectations that proceeds fund Nvidia hardware purchases or capacity that drives GPU demand.[1][2]
Arrangements range from straightforward equity stakes to complex structures like residual-value lease guarantees (e.g., OpenAI Ohio project) and unsold-capacity purchase commitments (e.g., CoreWeave). Nvidia has publicly stated certain deals (especially OpenAI guarantees) are not "circular financing," as customers remain primarily responsible for payments.[3][4] Most data comes from company announcements, SEC filings (e.g., S-1s, 8-Ks, 10-Qs), and reputable reporting (Reuters, Bloomberg, The Information); some earlier LOIs did not fully materialize or were superseded.
AI Labs
- OpenAI: Nvidia participated in a confirmed $30 billion equity investment (Feb 2026 funding round at $730 billion pre-money valuation, alongside SoftBank $30B and Amazon $50B+). This replaced/superseded a September 2025 LOI for up to $100 billion staged equity (tied to GW-scale deployments of Vera Rubin systems). Additionally, in August 2026, Nvidia provided up to $105 billion in residual-value lease guarantees/backstops for OpenAI’s Ohio data center campus (Pike County, developed by SB Energy; covers initial ~4.25 GW phase of an up-to-8 GW project; 20-year leases; Nvidia also invested $1.5 billion in SB Energy). OpenAI will pay leases primarily, with Nvidia stepping in only on default/insolvency (reimbursable by OpenAI). Proceeds and capacity support massive Nvidia GPU deployments (e.g., 10+ GW ambitions overall). Confirmed via announcements and filings; some elements reported in detail via financial disclosures.[5][6][7][4]
- xAI: Nvidia committed up to $2 billion equity in the confirmed $20 billion funding round (closed Jan 2026; other investors included Valor, Qatar Investment Authority, etc.). Structured partly via SPV/debt for GPU purchases (e.g., GB200/Colossus projects); Nvidia also involved in related infrastructure financing. Proceeds directly tied to Nvidia GPU acquisitions and data center buildout. Confirmed.[8][9][10]
- Anthropic: Nvidia committed up to $10 billion equity (announced/planned Nov 2025), executed via $10 billion purchase of non-voting Class N shares in a $26 billion concurrent private placement at IPO price (Nov 2026 S-1 filing; alongside Google $10B, AMD $2.5B, others). Tied to strategic partnership and up to ~1+ GW Nvidia compute commitments. Confirmed via SEC filings and announcements.[11][12][13]
- Perplexity: Nvidia participated in earlier rounds (e.g., Series B ~2024 with undisclosed amount; also noted in subsequent extensions). In August 2026, talks reported for a new equity investment in a round valuing the company at >$30 billion (up from ~$20 billion prior). No confirmed close on the latest round as of late 2026 data. Reported/rumored for the new round; earlier stakes confirmed.[14][15]
Neoclouds (GPU/AI Cloud Providers)
- CoreWeave: Equity investments include early (~$100M range in 2023) plus confirmed additional $2 billion in Jan 2026 (Class A shares at $87.20/share, increasing stake significantly to ~11%). Backstop: Confirmed $6.3 billion initial order (Sep 2025) for unsold/residual capacity through April 2032 (under master services agreement; Nvidia buys what CoreWeave cannot sell to third parties). Partnership includes early access to Rubin/Vera hardware and support for >5 GW buildout by 2030. Equity proceeds and backstop enable GPU purchases and capacity expansion. Confirmed via filings and announcements.[16][17][18]
- Lambda: Equity participation in Series D (~Feb 2025, $480M round at $2.5B post-money valuation; undisclosed Nvidia amount). Capacity/lease backstops include ~$1.5 billion in GPU/server lease-back deals (making Nvidia a major customer). Nvidia-linked financing (e.g., $926M leveraged loan for chip purchases, Aug 2026). Proceeds and structures tied to Nvidia GPU acquisitions. Confirmed/reported via announcements and reporting.[19][20][21]
- Crusoe: Equity participation across rounds, including Series E (Oct 2025, $1.375B) and Series F (Sep 2026 initial close of anticipated $3.9B at $30.9B post-money valuation; Nvidia among investors with Founders Fund, Mubadala, Valor, etc.). Focus on vertically integrated power + Nvidia GPU clusters; proceeds support GPU buys and >6 GW contracted capacity. Confirmed.[22][23]
- Nscale: Convertible note/equity commitments in Sep 2026 $3.36B pre-IPO round (Nvidia initially ~$1B, later reported up to $2B total in convertibles/notes; tranche closing ~Nov 16, 2026). Includes warrant tied to Texas net lease facility (Oct 2025). Additional lease/guarantee support and ~$1.2B capacity deals reported. Proceeds for Nvidia GPU-powered builds (contracts with Anthropic, Microsoft, etc.). Confirmed via filings and announcements.[24][25][26]
- Nebius: Confirmed $2 billion equity investment (March 2026; ~8-9% stake via shares/pre-funded warrant at ~$94.94/share). Strategic partnership for >5 GW Nvidia systems by end-2030, with engineering collaboration on AI factories, inference, and hardware (Rubin, Vera, etc.). Confirmed.[27][28]
Other notes: Nvidia has broader guarantees/backstops for AI clouds (e.g., ~$3.5B land/power/shell or capacity-related, plus larger aggregates reported in filings up to $100B+ range across ecosystem). Some neoclouds have GPU-backed debt or customer contracts that indirectly loop back via Nvidia hardware. Earlier or smaller deals (e.g., with Together AI or others) exist but are less detailed publicly.[1][29]
Summary Table of Key Arrangements
| Company | Date | Amount | Structure | GPU/Compute Link | Status | Key Sources |
|---|---|---|---|---|---|---|
| OpenAI | Feb 2026 | $30B equity | Equity in funding round | Expected GPU purchases; 10+ GW ambitions | Confirmed | Announcements, filings |
| OpenAI | Aug 2026 | Up to $105B | Lease residual value guarantees (Ohio campus) | Capacity for Nvidia systems | Confirmed (filings) | SEC/docs, Reuters/Bloomberg |
| xAI | Jan 2026 | Up to $2B equity | Equity in $20B round (SPV/debt elements) | Direct GPU buys (Colossus etc.) | Confirmed | Bloomberg, company |
| Anthropic | Nov 2025/2026 | $10B equity | Equity (non-voting shares in IPO placement) | ~1+ GW Nvidia compute commitments | Confirmed (S-1) | SEC filing, Reuters |
| Perplexity | 2024+; Aug 2026 talks | Undisclosed earlier; new round TBD | Equity stakes | Runs on Nvidia chips | Confirmed earlier; reported new | Reuters/The Information |
| CoreWeave | 2023 + Jan 2026 | ~$100M+ early; +$2B | Equity (multiple tranches) | Rubin/Vera access; >5 GW target | Confirmed | Reuters, filings |
| CoreWeave | Sep 2025 | $6.3B initial | Unsold capacity purchase commitment (to 2032) | Backstop enables more GPU deployments | Confirmed (8-K) | CoreWeave filing, Reuters |
| Lambda | Feb 2025 + 2026 | Undisclosed equity; ~$1.5B leases; debt support | Equity + lease-back + financing | GPU purchases and lease-backs | Confirmed/reported | Reuters, Bloomberg |
| Crusoe | 2025-2026 | Participant in $1.375B+ and $3.9B rounds | Equity | Nvidia GPU clusters; power integration | Confirmed | Company announcements |
| Nscale | Sep 2026 (close ~Nov) | ~$1-2B convertibles/notes | Convertible notes/equity in pre-IPO round; warrants | GPU-powered contracts (Anthropic etc.) | Confirmed | Filings, announcements |
| Nebius | Mar 2026 | $2B equity | Equity (shares/warrant) | >5 GW Nvidia systems by 2030 | Confirmed | Company, Reuters |
Notes on table: Amounts often "up to" or initial tranches; structures can combine elements. GPU link inferred from context (nearly all explicitly involve Nvidia hardware deployments). Status based on public confirmations vs. reporting. Not exhaustive of every small or indirect tie.
Money-Flow Diagram Description
Nvidia capital (equity or guarantees) flows to AI labs/neoclouds → recipients use funds (or enabled financing) to purchase/lease Nvidia GPUs and build data centers/power infrastructure → labs/neoclouds generate revenue by renting capacity to end customers (hyperscalers, enterprises, or other AI firms) or using it internally → revenue supports further expansion or repayments, creating a feedback loop where Nvidia hardware demand is sustained or accelerated.[1][2]
Key flows and labels:
- Direct equity loops (confirmed): Nvidia → OpenAI/xAI/Anthropic/CoreWeave/Nebius/etc. (equity cash) → back to Nvidia via GPU sales.
- Backstop/guarantee flows (confirmed/reported): Nvidia guarantees (e.g., OpenAI leases or CoreWeave unsold capacity) reduce lender risk → enables cheaper/more debt for builds → more GPU orders. Nvidia pays only on defaults (often reimbursable); capacity deals give Nvidia utilization rights.
- SPV/debt structures (e.g., xAI, some neoclouds): External financiers + Nvidia support fund GPU buys; chips collateralized; rentals repay debt.
- Capacity/lease circularity (reported/confirmed elements): Nvidia buys/rents back capacity it helped finance, providing demand floor while monetizing its own chips indirectly.
- Implications for competition/entry: High barriers via capital access and hardware preference; new entrants face "circular" financing advantages for Nvidia-aligned players but regulatory/execution risks around concentration and guarantees. Nvidia’s ecosystem role amplifies its market position but invites scrutiny on financing circularity.
Data is current as of November 2026 reporting; some commitments are multi-year/staged and subject to conditions (delivery, credit, financing contingencies). Additional primary filings would further verify exact terms.
Recent Findings Supplement (November 2026)
Nvidia has significantly expanded its role as financier and backstop provider for AI infrastructure since mid-2026, shifting emphasis toward lease/credit guarantees and capacity commitments (especially post-August 2026) alongside equity stakes. This creates circular flows where Nvidia capital or risk support enables GPU purchases or capacity that generates Nvidia revenue, though many guarantees phase in later (e.g., 2029+) and are conditional. Focus here is strictly on publicly reported items with new details or updates after May 19, 2026 (primarily August–November 2026 reporting on deals from ~August onward or updated portfolio figures). Older deals (e.g., early 2026 equity) are noted only if recent reports add material context or confirmation.[1][2]
No comprehensive public “map” of every arrangement exists; data comes from Nvidia filings/earnings commentary, S-1 filings, and news citing sources. Many neocloud details aggregate under “AI clouds.” Rumors/talks (e.g., Perplexity) remain unconfirmed as closed. Post-Jensen Huang resignation (November 2026), interim leadership signals more disciplined future commitments.[3]
OpenAI Arrangements (Primarily Guarantee-Focused New Developments)
Nvidia’s largest single exposure ties to OpenAI via a major data center project, structured as phased credit support rather than direct equity in the recent period. Proceeds/usage expected to flow back via exclusive Nvidia hardware on long-term leases.[1]
- SB Energy PORTS-Pike Ohio campus (OpenAI leases): Up to $105B credit support/lease guarantee (part of $118B total disclosed guarantees). Signed ~August 17, 2026 (phased; first exposure fiscal 2029; declines as OpenAI pays over 20 years). Covers ~4.25 GW (nine buildings; option for more). Campus to host Nvidia infrastructure exclusively under 20-year leases to OpenAI. Confirmed in Nvidia Q3 2026 earnings/10-Q (November 2026); previously discussed at ~$250B and scaled back. No payments made as of Oct. 25, 2026. Confirmed.[1][4]
- Equity stake: $30B in OpenAI’s ~$110–122B funding round (pre-money ~$730B; closed ~March 2026 at higher valuation). Recent reports (post-May) confirm Nvidia’s participation alongside SoftBank/Amazon and strategic compute partnerships. Confirmed (via company and news updates).[5]
- Related capacity/compute: Multi-GW dedicated inference/training capacity on Nvidia systems (e.g., via clouds); part of broader ecosystem. Proceeds tied to GPU purchases.
Implications for competition: This backstop lowers OpenAI’s financing risk for massive scale but ties hardware choices tightly to Nvidia; competitors (e.g., AMD) face higher hurdles entering similar large-scale deals without comparable balance-sheet support.
Anthropic Arrangements (Equity via IPO Placement + Compute Ecosystem)
Nvidia maintains a prior equity commitment while adding to it via IPO-related purchase; Anthropic’s provider deals indirectly support Nvidia GPU demand.[6]
- IPO concurrent private placement: $10B purchase of non-voting Class N shares at IPO price (part of $26B placement; Nvidia alongside Google at $10B each). Announced/confirmed via amended S-1 filing (November 2026). Builds on November 2025 commitment (up to $10B at ~$350B valuation). Confirmed.[6][7]
- Prior/ongoing equity: Up to $10B commitment (November 2025, ~$350B valuation). Recent valuations/markups noted in November 2026 reports (potential paper gains). Confirmed.
- Compute ecosystem: Anthropic agreements with Nvidia-powered providers (e.g., SpaceXAI, Nscale, Lambda) totaling hundreds of billions in committed spend; strategic partnership for hardware optimization. Reported/confirmed via filings.
Implications: Strengthens Nvidia’s position as Anthropic scales pre-IPO; indirect GPU demand via third-party clouds diversifies exposure beyond direct sales.
xAI Arrangements
Limited new public details post-May 2026. Nvidia previously invested in xAI (stake later tied to SpaceX shares, valued ~$17B in recent reports). One November 2026 report references Nvidia as strategic investor in an xAI round, but specifics (amount/date) overlap with earlier periods and lack fresh confirmation. No major new guarantees or equity reported in results. Sparse/reported at best.[8]
Perplexity Arrangements (Investor Status + Rumored Follow-On)
Nvidia confirmed as existing investor with ongoing commercial ties (e.g., NIMATRON coalition); new round talks reported but unclosed.[9]
- Existing investment: Nvidia is an investor (confirmed on-record by Perplexity CEO, November 2026). Earlier participation in rounds (e.g., Series B onward). Confirmed.
- Potential new round: Talks for Nvidia to lead/join round at >$30B valuation (reported August 2026 by The Information/Reuters). Not announced as closed by mid-November 2026. Reported/rumored (talks stage).[10]
Implications: Equity supports Perplexity’s growth and Nvidia ecosystem (open models, compute); unclosed round leaves uncertainty on further capital commitment.
Neoclouds (CoreWeave, Nscale, Lambda, Crusoe, Others) – Guarantees, Capacity Buys, and Equity
August 2026 saw broad backstops for “AI clouds/neoclouds”; specific updates in November 2026 filings/news. Nvidia acts as investor, supplier, capacity buyer, and partial guarantor—creating revenue loops via GPU sales and unused-capacity purchases.[11]
- CoreWeave: $2B equity (January 2026; recent reports affirm ongoing). $6.3B agreement to buy unsold/ unused capacity (through ~2032). Nvidia also supplier and partner for multi-GW buildout. Recent Q3 results (November 2026) highlight relationship amid CoreWeave’s growth. Confirmed.[12]
- Nscale: $2B commitment (doubled from $1B) to >$3B convertible financing (closes ~November 2026; per amended IPO filing November 13, 2026). Additional: $1.2B lease deal; $860M guarantee for Texas site. Debt rose sharply. Confirmed (filing).[11]
- Lambda: Nearly $20B in lease guarantees (reported November 2026). Reported.[13]
- General AI clouds/neoclouds (Aug 2026 disclosures): $3.5B land/power/shell guarantees; $36B purchase of unused capacity; $20B data-center leases; $32B total commitments (sharing rental revenue; down slightly in latest quarter). Part of $118B guarantees. Confirmed (Nvidia filings).[1]
- Broader platform: August 2026 MOU with six financial firms (Apollo, BlackRock, etc.) for platforms targeting >$500B third-party capital; Nvidia backstop up to $125B (25%). Deals still under discussion. Reported/confirmed structure.[14]
- Crusoe and others: Mentioned in Nvidia’s startup portfolio/backed companies (PitchBook data in recent reports); no specific new post-May amounts disclosed. Reported in aggregate.
Implications for entry/competition: Neoclouds rely on Nvidia capital/risk support for rapid scaling; post-Huang, more scrutiny on new deals favors established players or those with independent financing. Capacity buybacks reduce idle-risk but tie revenue directly to Nvidia.
Summary Table of Key Post-May 19, 2026 Reported Items
Focus on new/updated details (amounts in USD; structures as described; labels per sources; many tie back to GPU purchases/capacity).
- OpenAI (SB Energy Ohio): $105B max guarantee (lease/credit support); ~Aug 2026; phased 2029+; Confirmed (Nvidia filings).[1]
- OpenAI equity: $30B; round closed ~Mar 2026 (reports post-May); Confirmed.
- Anthropic IPO placement: $10B equity purchase; Nov 2026; at IPO price; Confirmed (S-1).[6]
- Nscale convertible: $2B (of >$3B total); ~Sep/Nov 2026; Confirmed (filing).[11]
- Lambda leases: ~$20B guarantees; reported Nov 2026; Reported.[13]
- Neoclouds aggregate (guarantees/capacity): $3.5B + $36B + $20B + $32B commitments; Aug 2026 disclosures; Confirmed.
- Financial firms MOU backstop: Up to $125B; Aug 2026; Reported.
- Perplexity new round: Talks at >$30B val (amount unspecified); Aug 2026 reports; Reported/rumored (unclosed).[10]
- xAI: Limited/no major new specifics; prior stake noted in recent valuations; Sparse.
Money-Flow Diagram Description
Nvidia equity/guarantees/MOUs → AI labs (OpenAI/Anthropic) or neoclouds (CoreWeave/Nscale/etc.) secure financing/build data centers → These entities purchase/lease Nvidia GPUs/systems or commit capacity (direct or via providers) → Revenue/cash flow returns to Nvidia (chip sales, capacity buybacks, or shared rentals); guarantees reduce default risk, enabling larger scale. Key loops: OpenAI Ohio leases (Nvidia hardware exclusive) + neocloud unused-capacity purchases + Anthropic provider deals. Total ecosystem commitments cited in analyst estimates ~$400B+ range (including supply), with ~$118B guarantees as of Nov 2026. Flows are conditional and phased; not all proceeds are one-for-one GPU buys but strongly aligned.[15]
Caveats: Exact individual amounts for some neoclouds aggregate; not all “proceeds return as GPU purchases” explicitly quantified per deal (inferred from structures). Future deals may face tighter terms. Sources primarily Nvidia disclosures and reputable news (e.g., Reuters, TechCrunch, filings). For full verification, review latest 10-Q/S-1 documents. Additional primary filings would strengthen specifics on unclosed talks.
Report 2 Research what Nvidia disclosed in its Q3 fiscal 2027 earnings release, 10-Q, and call about strategic investments, non-marketable equity holdings, purchase and supply commitments, guarantees, cloud-service agreements, and customer concentration. Compare these with prior quarters and note the changes in tone and wording. Add analyst and accounting commentary on the adequacy of the disclosures, and flag anything that is a publicly estimated figure rather than a reported one.
Nvidia’s Q3 FY2027 results (quarter ended October 25, 2026, reported November 17, 2026) include expanded disclosures on off-balance-sheet commitments, guarantees, and equity investments in its 8-K earnings release, CFO commentary exhibit, and supplemental financial information. The full 10-Q has not yet been filed (expected in the coming weeks, with a December 4, 2026 SEC deadline), so granular customer concentration details (typically in Note on concentrations or revenue by geography/platform) are not available in the latest filing.[1][2]
These disclosures build on prior quarters with larger absolute numbers, more explicit conditionality language, and a rename of certain agreements, reflecting Nvidia’s role in financing AI infrastructure buildout amid rapid growth (Q3 revenue $115.8 billion, Data Center $107.6 billion).[3][4]
Supply, Capacity, and Total Forward Commitments
Nvidia disclosed $390 billion in supply and capacity commitments as of October 25, 2026 (primarily multi-year memory procurement to secure components), up sharply from $279 billion as of July 26, 2026 (Q2 FY2027). Total future commitments across categories reached $518 billion, including supply/capacity ($390B), cloud service agreements ($28B), data center leases not yet commenced ($74B), equity investments ($20B subject to contingencies), and capital expenditures ($6B).[5][6]
- The increase signals confidence in sustained demand but highlights reliance on suppliers for scaling Rubin and other platforms.
- Cloud service agreements and leases support R&D (e.g., chip engineering, open models like Nemotron/Cosmos/GR00T, autonomous vehicle software); leases run up to 20 years and may be assignable to third parties.[5]
- Compared to Q2: The supply line more than doubled quarter-over-quarter in recent periods (from ~$119B earlier to $279B then $390B), driven by memory deals.[7]
For competitors or new entrants: This underscores Nvidia’s ability (and need) to lock in supply chains at massive scale; rivals would face similar or greater challenges without equivalent balance-sheet strength or demand visibility. The figures are reported contractual commitments, not estimates.
Neocloud (Formerly AI Cloud) Agreements
Commitments under neocloud agreements totaled $32 billion as of October 25, 2026 (down from $36 billion as of July 26), scheduled roughly as $6B in FY2028, $7B in FY2029, $6B in FY2030, $5B in FY2031, and $8B thereafter.[5][8]
- These are conditional: They take effect only if neoclouds secure financing to purchase infrastructure, deploy/operationalize capacity, and obtain Nvidia acceptance. If contingencies are not met timely, amounts/timing can change or commitments can be reduced/eliminated.[5]
- Nvidia earns revenue on upfront infrastructure sales and participates in upside rental revenue share above a minimum (per call remarks by interim CEO/CFO Colette Kress).[8]
- Wording change: Renamed from “AI cloud agreements”; commentary and call emphasize conditionality more explicitly than prior periods. Kress clarified on the call that guarantees/commitments activate only after financing, construction, and readiness—delays mean the backstop “doesn’t exist anymore.”[9]
Analyst view: This language addresses concerns about Nvidia acting as financier for under-capitalized neoclouds; adequacy is viewed as improved by contingencies, though the scale still draws scrutiny as potential off-balance-sheet exposure.[9]
Guarantees and Credit Support
Maximum gross guarantee exposure rose to $118 billion (from $108.5 billion in Q2), with no payments made to date. Breakdown includes ~$105 billion for credit support on land/power/shell at SB Energy’s PORTS-Pike campus in Ohio (4.25 GW across nine buildings, 20-year leases to OpenAI, phasing in starting ~FY2029 and declining with payments); $9.5 billion in power/systems credit support; and $3.5 billion for neocloud land/power/shell.[6][10]
- Guarantees are conditional (e.g., activate building-by-building when ready for service; tied to partner performance/defaults).
- Compared to prior: Up modestly, with added power/systems detail; tone in commentary and call stresses triggers and risk mitigation more than before.[5]
Analyst/accounting commentary: The $118B figure (far exceeding liquid assets of ~$53.5B in cash/marketable securities) is highlighted as Nvidia functioning as the “AI boom’s banker,” but contingencies and phased activation limit immediate risk. Some note it reflects bottlenecks in powered facilities rather than chips alone. Disclosures are seen as reasonably detailed for a large accelerated filer, though full 10-Q footnotes may add more.[9]
Equity Investments (Strategic and Non-Marketable Holdings)
Marketable equity securities stood at $44.219 billion and non-marketable equity securities at $61.65 billion as of October 25, 2026 (non-marketable up from $47.898 billion prior quarter; total equity holdings ~$109.7 billion).[11][12]
- Q3 net gains from equity securities: $1.8 billion ($2.8B public gains offset by $1.0B non-marketable net loss, including $1.8B impairment). Cumulative unrealized gains/losses tracked in supplemental info.
- Focused on frontier labs, infrastructure financiers, and private companies (subject to contingencies); $20B in equity investment commitments in the forward table.
- Growth driven by new investments and valuation increases (e.g., public stakes like Intel/SpaceX; private in AI labs/neoclouds). Specific large stakes (e.g., ~$30B OpenAI) are analyst/public estimates, not line-item reported figures.[13]
Changes vs. prior quarters: Portfolio has grown dramatically (from low billions pre-2025/2026 to >$100B), with more reclassifications (e.g., xAI to SpaceX) and impairments noted. Tone remains focused on strategic alignment with ecosystem partners.
Customer Concentration and Related Disclosures
Detailed concentration (e.g., % from individual direct customers >10%) is typically in the 10-Q and not yet available. From the CFO commentary and call: Hyperscale revenue $60.1 billion (up 23% sequentially); AI Clouds, Industrial & Enterprise (ACIE) $47.5 billion (up 18% q/q, nearly tripled y/y). Both segments grew with Rubin ramp; hyperscalers drove more of the sequential DC increase this quarter.[14][15]
- Prior 10-Qs (e.g., Q2): One direct customer ~16% of revenue; top three ~44% in 1H FY2027; five customers ~70% of receivables. Hyperscale often ~50-55% of DC revenue.[16]
- Call emphasized balanced growth across platforms and strong demand visibility, with no China DC compute assumed in guidance.
Overall analyst view on disclosures: Expanded conditionality language and tables enhance transparency on risks (e.g., financing dependencies), addressing “banker” concerns, though the sheer scale prompts questions about execution and balance-sheet leverage. No major red flags on adequacy noted in initial coverage; figures are company-reported (not estimates), except for inferred stake sizes.[9]
These elements illustrate Nvidia’s ecosystem strategy—locking supply, backing partners selectively, and investing strategically—while highlighting growing (but conditional) financial entanglements with the AI buildout. Full 10-Q will likely provide more on concentrations, fair value measurements, and risk factors.
Recent Findings Supplement (November 2026)
Nvidia reported Q3 FY2027 results (ended Oct. 25, 2026) on Nov. 17, 2026, via 8-K/earnings release and detailed CFO commentary (Colette Kress, interim CEO/CFO), with the 10-Q expected in coming weeks.[1][2]
Revenue reached a record $115.8 billion (+103% YoY, +20% QoQ; beat $108B ±2% guidance), driven by Data Center ($107.6B, +110% YoY). Q4 guidance is $130B ±3% (extra week; no China Data Center compute assumed). The results mark the first under interim leadership following Jensen Huang’s resignation.[3][3]
Key focus areas show scaled balance-sheet support for the AI ecosystem alongside more explicit contingency language and a cautious tone on future commitments.
Strategic investments and non-marketable equity holdings expanded sharply, with Nvidia acting as a major investor in frontier AI labs and related infrastructure. Marketable equity securities reached $44.219 billion and non-marketable $65.528 billion (total ~$109.7 billion) as of Oct. 25, 2026. Nvidia invested $58 billion in public/private equities in the first nine months of FY2027 ($16 billion in Q3 alone), versus $5 billion in the prior-year period.[4][5]
- Cumulative unrealized gains on non-marketable equity securities: $9.9 billion; cumulative unrealized losses/impairments: $2.2 billion. Q3 included a $1.8 billion impairment on non-marketable securities and $2.88 billion in gains on publicly traded equities (net equity securities gain ~$1.8 billion overall).[6]
- Future equity investment commitments: $20 billion (focused on frontier labs, infrastructure financiers, and private companies), part of the broader $518 billion total commitments table.[2]
- Portfolio growth reflects both new investments and valuation increases; prior Q2 FY2027 non-marketable holdings were lower (~$47.9 billion at end of July 2026).[7]
This positions Nvidia deeper in the ecosystem it supplies, creating both upside (valuation gains, strategic alignment) and downside (impairment/liquidity risk). For competitors or entrants, it highlights a high bar: matching Nvidia’s data/relationship advantages or its willingness to deploy capital alongside hardware.
Purchase and supply commitments rose materially to secure long-term capacity, primarily memory, while cloud-service agreements and data-center leases grew with multi-year structures. Supply and capacity commitments reached $390 billion (up from $279 billion in Q2), mostly multi-year memory supply. Total future commitments (supply/capacity + cloud services + leases not commenced + equity investments + capex) hit $518 billion as of Oct. 25, 2026.[2][8]
- Breakdown by fiscal year (in billions): Supply/capacity $390 total (heavily front-loaded through 2029); cloud service agreements $28 total; data-center leases not commenced $74 total (many 20-year terms, some assignable to third parties, commencing FY2027–2033); equity investments $20; capex $6.[2]
- Cloud-service agreements and data-center lease commitments combined: ~$102 billion (up from $74 billion prior quarter).[3]
- Additional neocloud agreements (formerly “AI cloud agreements”): $32 billion maximum (down slightly from ~$36 billion prior; revenue share on upside above minimum after upfront hardware sales).[9][10]
New supply-chain deals (e.g., with Kinsus, Samsung Electro-Mechanics, SPIL for substrates/packaging) were announced alongside results. These commitments signal sustained demand visibility but create long-term obligations (multi-year, not immediate debt). Entrants must navigate similar supply constraints or partner structures; the scale underscores Nvidia’s moat in locking capacity ahead of demand.
Guarantees expanded to a $118 billion maximum gross exposure, with detailed contingencies and a prominent OpenAI-linked item; new wording clarifies activation conditions. Maximum gross guarantee exposure: $118 billion (up from ~$108.5 billion prior quarter). Of this, $105 billion supports land/power/shell for ~4.25 GW across nine buildings at SB Energy’s PORTS-Pike campus in Ohio (20-year leases exclusively for NVIDIA infrastructure to OpenAI; activates building-by-building when ready, first potentially FY2029).[2][11]
- Other components: ~$3.5 billion for neocloud land/power/shell; ~$9.5 billion for power/systems credit support. No payments made to date under these guarantees.[2]
- Explicit new language: Neocloud commitments “take effect only if the neoclouds secure financing to purchase the data center infrastructure products and the related compute capacity is deployed, made operational, and accepted by us. … If they are not timely resolved, the amount and timing of our commitments may be affected, and certain commitments could be reduced or eliminated.”[9][9]
- Kress emphasized on the call that guarantees depend on financing/execution and that Nvidia is “rigorously evaluating future investments, commitments, and guarantees” with “disciplined” and “selective” support, prioritizing land/power/shell readiness and buybacks.[12]
This represents contingent, off-balance-sheet exposure tied to partner performance. Disclosures provide year-by-year tables and breakdowns, improving transparency versus prior periods. Analysts noted the contingent nature as a positive clarification that could limit “circular financing” risks, though the absolute scale ($118B guarantees + $518B commitments) drew attention as Nvidia effectively bankrolling ecosystem buildout.[11][11] For market participants, these figures flag execution and credit risks in partner-dependent growth; adequacy appears reasonable given the detail, but monitoring actual drawdowns or reductions will be key.
Customer concentration remains elevated, with four direct customers each accounting for 11–13% of revenue over recent periods (combined ~48% in 9M FY2027); disclosures use unnamed “Customer A/B/C/D” labels consistent with prior quarters.[13] Receivables are also concentrated (e.g., two customers at 18% and 17% of the $81 billion balance).[13]
This pattern (top customers rising from lower single-digit shares in earlier years to 30–60%+ ranges recently) continues without material change in wording or disclosure format. No new named customers or shifts in tone were highlighted in the Q3 materials. High concentration amplifies risks from any single buyer’s capex shifts or financing issues, reinforcing the importance of the neocloud and guarantee structures for diversification.
Overall tone in the CFO commentary and call shifted toward greater caution and selectivity compared to prior quarters, with new explicit contingency language on neoclouds and emphasis on disciplined capital allocation. Prior materials referenced similar OpenAI-related guarantees (~$105B) and rising commitments but lacked the same level of “take effect only if…could be reduced or eliminated” phrasing or post-leadership framing of rigorous evaluation.[11][10] Analyst and commentary reaction has been largely positive on disclosure depth (tables, breakdowns, contingencies) while flagging the banker-like role and scale of exposures as noteworthy; no widespread critiques of inadequacy emerged in initial coverage, though the figures invite ongoing scrutiny of realization versus maximums.[11][11] All headline commitment, guarantee, and holdings figures are company-reported (not estimates); some analyst aggregates (e.g., combined exposures) are interpretive.
These updates reflect Nvidia’s response to explosive demand via balance-sheet tools, with evolving risk disclosures. Competitors or new entrants face a capital- and relationship-intensive environment where matching such commitments or contingencies is challenging. Further 10-Q details or subsequent filings could provide additional line-item granularity.
Report 3 Investigate the reported Nvidia backstop talks (about $250B) for OpenAI compute, including the structure, counterparties (Oracle, Microsoft, CoreWeave, SPVs), and the status of the reporting. Analyze whether the OpenAI CEO change alters the talks, and cover the reactions of investors, partners, and lenders. Cover related deals: the AMD warrant agreement, Oracle and Microsoft contracts, and TSMC's reported interest in OpenAI and Anthropic stakes.
Nvidia scaled back a proposed $250 billion credit backstop for OpenAI’s massive Ohio data center project to a signed $105 billion residual-value guarantee covering the initial 4.25 GW phase, structured as a last-resort shortfall cover rather than direct financing or lease payments.[1][2]
The mechanism works as follows: SB Energy (SoftBank subsidiary) builds, owns, and operates the PORTS-Pike campus in Pike County, Ohio (former uranium site), leasing it to OpenAI for 20 years. Nvidia provides credit support/guarantees on land, power, and shell (LPS) infrastructure for the first phase (with an option for more), capped at $105 billion aggregate. This activates only after buildings are completed and ready for service (first expected FY2029), and only triggers on specific events like OpenAI default or insolvency—after SB Energy attempts re-leasing or sale. Nvidia also invests $1.5 billion in SB Energy and secures exclusivity as the compute provider (full-stack DSX platform) for the site. OpenAI indemnifies Nvidia for any payouts. The original July 2026 $250 billion proposal (covering a full ~10 GW campus with >$500 billion total project cost, plus separate chip financing talks) was halved amid investor concerns over risk exposure; shares fell ~5% on initial reports.[3][4][5]
- The guarantee is conditional and phased; it declines as OpenAI makes lease payments and can terminate if OpenAI achieves investment-grade credit. No payments have been made as of late 2026.[6]
- Broader Nvidia guarantees reached $118 billion by Q3 FY2027 (including $105 billion for PORTS-Pike, plus neocloud and other support), part of ~$286 billion+ in named commitments.[7]
- This enables OpenAI (unrated/private) to secure favorable debt/lease terms by leveraging Nvidia’s balance sheet, while locking in long-term Nvidia chip demand across multiple generations.
For competitors or entrants: This creates a high barrier via exclusivity and vendor-financed infrastructure; new players must either match balance-sheet strength or find alternative financing without such backstops. It highlights circular financing risks but also Nvidia’s leverage in the ecosystem.
Counterparties center on SB Energy (developer/owner), with OpenAI as tenant and indirect involvement from Microsoft/Oracle/CoreWeave via broader OpenAI cloud and compute deals; SPVs or special structures are not prominently detailed beyond SB Energy’s role and Nvidia’s equity stake.[2]
SB Energy handles build-own-operate under the 20-year OpenAI lease. Nvidia’s involvement includes the guarantee plus $1.5B equity in SB Energy (joining SoftBank and OpenAI as investors). Power involves U.S. government control/funding under trade deals, with Commerce Secretary input. OpenAI’s other infrastructure includes Microsoft Azure (core partner), Oracle Cloud (significant backlog exposure, with analysts noting heavy OpenAI weighting in Oracle’s ~$664B backlog), and CoreWeave (neocloud with rising debt, e.g., ~$29.8B by end-2025, and Nvidia capacity deals).[8][9]
- Nvidia has separate neocloud commitments (e.g., up to $36B cloud purchases, later trimmed) that support operators like CoreWeave filling halls with its chips.[7]
- No major SPV-specific details emerged in reporting; structures emphasize direct guarantees and equity rather than complex vehicles. Oracle and Microsoft contracts provide additional OpenAI capacity but compete with or complement the Ohio site.
Implications: Concentration risk is high—OpenAI is a linchpin customer for multiple providers. Lenders gain comfort from Nvidia’s backing but face scrutiny on circularity (guarantees ultimately support Nvidia chip sales).
Nvidia’s leadership transition (Jensen Huang’s November 2026 resignation amid personal scandals, replaced by interim CEO/CFO Colette Kress) has introduced caution but not derailed the OpenAI deal; OpenAI’s Sam Altman remains CEO with no material change affecting the talks.[10]
Kress reportedly pushed back on the original $250B proposal as CFO (reworking it downward) and favors a more conservative approach than Huang’s aggressive dealmaking. Post-transition, she clarified on the Q3 earnings call that guarantees are conditional (only after financing, construction, and acceptance; “doesn’t exist” if projects stall) and emphasized capital discipline—without signaling a pullback. Altman publicly stated the AI industry “will be fine” due to strong demand/revenue growth, praising Huang’s vision while noting Nvidia’s scale and trusted guarantees; he expressed empathy but declined to opine on Huang’s return. OpenAI had a brief April 2026 Altman ouster (quickly reversed); stability holds as of November 2026.[11][12]
- Nvidia shares and related stocks (e.g., neoclouds like CoreWeave, IREN, Nebius) saw volatility around the Huang exit and earnings, with some declines tied to backstop concerns.[7]
- Reactions: Investors marked credit risk (wider spreads possible for weaker names); partners like OpenAI, Meta, and AWS continue engagement; lenders appreciate clarification but watch for reduced appetite. Altman’s comments helped reassure on demand.
Implications: Continuity is likely under Kress (“business as usual” with discipline), but aggressive expansion of guarantees may slow. OpenAI leadership stability supports deal momentum.
Related deals reinforce ecosystem ties: AMD’s performance-based warrants with OpenAI (and Meta), heavy Oracle/Microsoft OpenAI contracts, and TSMC’s November 2026 board-approved stakes.[13]
- AMD warrants: October 2025 OpenAI deal for up to 6 GW Instinct GPUs (starting ~1 GW MI450 shipments H2 2026) includes warrant for up to 160M shares at $0.01 (vesting on purchase milestones, stock-price targets up to $600, technical/commercial conditions; expires 2030). Similar Meta deal (Feb 2026). Q3 2026: ~$1.07B non-cash charge as shipments ramp; potential dilution if fully exercised (~16% combined stake).[13][14]
- Oracle/Microsoft: Oracle has substantial OpenAI exposure in its backlog; Microsoft remains a core Azure partner. These complement (or compete with) Nvidia-backed sites.
- TSMC stakes: November 10, 2026 board approval for up to $500M OpenAI preferred stock and $500M Anthropic shares (at IPO price, part of larger $26B placement including Nvidia, Google, etc.). Signals confidence in demand while deepening supplier-customer financial links; small relative to TSMC’s capex.[15][16]
Overall status (as of Nov 19, 2026): The downsized Nvidia-OpenAI guarantee is signed and disclosed (SEC filings confirm terms); talks evolved from ambitious proposal to executed, conditional structure amid pushback. No collapse post-leadership shifts; ecosystem deals (AMD, TSMC) advance in parallel. Risks include execution delays, credit events, or policy shifts, but strong underlying AI demand supports continuation.[11]
For market participants: Monitor Nvidia’s next disclosures on guarantee utilization, neocloud financing outcomes, and any Kress-led adjustments. High circularity amplifies systemic exposure but accelerates buildout for those aligned with Nvidia/TSMC supply chains.
Recent Findings Supplement (November 2026)
Nvidia scaled its reported $250B OpenAI backstop proposal to a conditional $105B lease guarantee (executed ~August 2026), with the full structure clarified in November 2026 earnings as highly conditional on project completion.[1][2]
The deal supports SB Energy’s (SoftBank-linked) PORTS-Pike campus in Ohio for ~4.25 GW across nine buildings under 20-year leases to OpenAI (exclusive Nvidia chips). Nvidia’s maximum exposure across all AI infrastructure guarantees reached $118B (including $9.5B power/systems and $3.5B land/power/shell for neoclouds), up modestly from prior quarter disclosures. No guarantee payments have been made.[3][4]
- Activation is phased and conditional: Each building’s guarantee takes effect only after independent financing is secured, construction completes, power is available, Nvidia compute is installed/operational/accepted, and the facility is ready for service (first expected FY2029). Exposure then declines over the lease term as OpenAI pays.[2][3]
- Interim CEO Colette Kress (former CFO) emphasized on the Nov 17/18 earnings call that guarantees “don’t exist” if pre-completion delays or challenges arise; this was framed as a clarification of existing terms, not a policy shift, though new disclosure language was added.[1][5]
- Broader context includes a ~$500B MOU/platform with financial firms (Apollo, BlackRock, Brookfield, KKR, etc.) for third-party capital mobilization, where Nvidia may backstop up to ~25% ($125B) on a case-by-case basis; talks continue but no deals closed yet, with expectations for 2027+.[6][7]
Implications for competitors/entrants: The conditional structure limits Nvidia’s construction/early-stage risk while still de-risking completed assets for lenders, potentially favoring larger or better-capitalized projects. New entrants or smaller neoclouds face higher hurdles for similar support.
Nvidia’s leadership transition (Jensen Huang resignation ~Nov 12, 2026; Colette Kress as interim CEO) reinforces a more conservative stance on large backstops, with Kress having reportedly opposed/reduced the original OpenAI proposal.[8][9]
Huang’s exit followed reports of extramarital affairs. Kress (who pushed back on the $250B figure, leading to the split/scaled $105B version) is expected to apply stricter evaluation to future commitments, prioritizing land/power/shell availability and selectivity as other capital sources emerge.[10][1]
- OpenAI CEO Sam Altman remains in place with no reported change; a brief April 2026 dismissal was quickly reversed, and he continues active leadership (e.g., Nov 18 interviews). No evidence the Nvidia transition has altered OpenAI-side talks.[11][12]
- Reactions: Nvidia stock had dropped ~5% on initial $250B reports (wiping ~$250B market value); neocloud stocks (e.g., CoreWeave) fell on Kress news amid repricing concerns. Partners/lenders continue platform discussions but may demand higher spreads or more independent financing for riskier deals. Altman stated the AI industry “will be fine.”[6][13]
Implications: Reduced risk appetite at Nvidia could slow marginal neocloud or high-risk projects while accelerating consolidation toward hyperscalers or well-backed players. Existing commitments (including the OpenAI guarantee) are expected to hold.
Related deals show continued circular financing patterns with accounting and execution updates.[14]
- AMD warrants (OpenAI/Meta): OpenAI deal (2025) covers up to 6 GW AMD Instinct GPUs with warrants for ~160M shares (~10% potential stake, $0.01 exercise). Nov 3/4, 2026 earnings: AMD recorded a $1.07B non-cash charge on warrant fair-value changes as MI450 shipments ramp; vesting tied to milestones (initial 1 GW binding, full at 6 GW plus stock-price thresholds up to $600) and technical/commercial conditions. Additional OpenAI capacity deals announced (e.g., Nov 2026 TensorWave).[14][15]
- Oracle/Microsoft contracts: OpenAI maintains multi-billion commitments (e.g., Oracle Abilene capacity accepted; Microsoft Azure as major provider with tens of billions in incremental deals). Some reports note Microsoft winding down older Oracle AI compute rentals on prior-gen Nvidia chips; Oracle raising significant capital ($20B stock/bonds) amid large backlog. No major new Nvidia-tied announcements.[16][17]
- TSMC stakes: Nov 10, 2026 board approval for up to $500M OpenAI preferred stock and $500M Anthropic common shares (at IPO price), as part of Anthropic’s $26B private placement (other participants include Nvidia, Google, AMD, Microsoft). Anthropic IPO priced Nov 18; trading began Nov 19. Signals supplier confidence amid spending concerns.[18][19]
Implications: Warrant structures and equity stakes deepen vendor-customer alignment but create dilution/accounting volatility (e.g., AMD charges). TSMC’s dual bets highlight supply-chain entanglement. Entrants must navigate similar equity-linked or guaranteed financing to compete on scale.
No significant new regulatory/policy updates or independent research publications on these specific deals appeared in recent sources. All figures are in USD.
Report 4 Research Anthropic's compute commitments and financing: the Broadcom roughly $63B chip deal and its SPV structure, the Nscale deal that is not yet funded, Google TPU and Amazon arrangements, Nvidia and Microsoft involvement, Nvidia's interest in the Anthropic IPO, and TSMC's reported plans. Identify who bears the risk (SPV lenders, Broadcom, Anthropic), what remains unfunded, and the IPO-related incentives. Present the findings as a structured deal table with sources.
Broadcom $63B SPV Deal (Project North Star / Northstar Compute LLC): A newly formed SPV purchases Broadcom XPU AI accelerators and networking equipment (tied to Google TPU designs) and leases them to Anthropic, with deliveries starting in calendar 2027. The SPV is financed by ~$63B in committed debt/equity from Bank of America, Citigroup, and Morgan Stanley (first-lien arrangers, senior tranche ~$43.5B or ~69% per Reuters sources) plus Blackstone (sole committed second-lien provider). Broadcom provides residual value support equal to the first-lien amount, structured around a formula (e.g., covering gaps below 72.5% of outstanding lease obligations versus resale value of racks, with peak exposure cited around $43B undiscounted, declining over ~5 years).[1][2][3]
Anthropic’s S-1 discloses aggregate ~$161.2B in five-year master equipment lease rent obligations (excluding interest) tied to this and related Broadcom/TPU capacity, classified as finance leases that are generally non-cancelable except on default; obligations are secured by the leased equipment and guaranteed by Anthropic PBC. Anthropic may also issue up to $42B in convertible notes to Broadcom or a designated partner.[4][5]
Risk allocation: Senior lenders (BofA/Citi/MS) are protected by Broadcom’s backstop on asset value/residual risk up to the first-lien size; Blackstone’s junior tranche bears more exposure after the guarantee. Broadcom bears the primary downside on hardware value depreciation (a key AI chip obsolescence risk). Anthropic bears ongoing lease payment obligations regardless of usage in the broader structure.[6][7]
Nscale Deal: Up to ~$44.6B (six-year agreement signed August 2026) for Nvidia GPU-powered compute capacity (initially ~460 MW at Monarch campus in West Virginia, with potential for a second similar ~$45B building under negotiation; Anthropic holds rights of first refusal on additional buildings). Fees accrue only as capacity is delivered and accepted.[8][9]
This remains largely unfunded as of mid-November 2026 filings: Nscale has no binding financing commitments for the required data center/GPU buildout (needs tens of billions more beyond ~$1B already committed to the site); it must use “best efforts” to obtain qualifying financing or face potential termination rights for Anthropic without liability in some cases. Nscale’s total debt reached ~$5.5B by end-September 2026, with warnings of needing additional capital for 2027 obligations amid widened losses. Nvidia has been in talks to guarantee or support related debt.[10][11][12]
Google TPU/Cloud Arrangements: Minimum commitments of ~$114.5B–$121.3B (updated in later S-1 amendments) for TPU-based hosting/compute under infrastructure service agreements (take-or-pay style, payable regardless of usage with true-up mechanisms; non-cancelable except for uncured breach by Google). This covers periods into 2033 and is separate from but linked to the Broadcom SPV hardware leases for additional TPU capacity (earlier agreements referenced ~3.5 GW starting 2027). Anthropic estimates a further $125B–$150B for additional TPU compute delivered 2028–2031.[13][14]
Amazon Arrangements: Minimum $110B commitment (April 2026 agreement) for AWS compute including custom Trainium chips over ~10 years (to April 2036), take-or-pay with true-up, non-cancelable except uncured breach.[15]
Microsoft Arrangements: Minimum commitments of ~$44.9B–$49.2B (amended) for Azure compute capacity (Nvidia GPU-powered elements noted in broader ecosystem), take-or-pay style into 2033, with tranche-specific delivery protections.[14]
Nvidia and Broader Involvement: Nvidia GPUs power multiple third-party deals (e.g., portions of SpaceXAI up to $120.5B, Nscale, Lambda ~$41.3B). Nvidia has strategic supply ties and previously pledged up to $10B investment (partially drawn). It is buying ~$10B in Anthropic shares via a concurrent $26B private placement (non-voting Class N stock at IPO price) alongside Google (~$10B), AMD ($2.5B), Broadcom/Micron/Microsoft (~$1B each), and TSMC ($500M).[16][17]
TSMC Plans: TSMC (key foundry supplier for Nvidia and others; flagged as a concentration risk in Anthropic’s filing) is investing $500M in the IPO private placement for common shares at the IPO price. Its board also approved up to $500M for OpenAI preferred stock.[18][19]
IPO-Related Incentives and Context: Anthropic’s IPO (pricing targeted around November 18, 2026; range $70–$78/share implying ~$1.4T+ valuation) includes a large concurrent private placement where suppliers (Nvidia, Google, TSMC, Broadcom, etc.) anchor with $26B in equity purchases. This aligns incentives for continued supply/support while providing Anthropic capital; suppliers gain equity upside tied to Anthropic’s success (and thus compute demand). Earlier Nvidia pledges and ongoing GPU ecosystem deals reinforce this.[20][21]
Overall Unfunded/Outstanding Elements: Nscale capacity buildout/financing is the clearest unfunded item (Anthropic termination rights protect it). Broader TPU estimates ($125–150B) and some AMD tranches sit outside core contractual floors. Total named/estimated compute commitments across S-1 disclosures exceed $600–900B (varying by inclusion of estimates and updates), dwarfing Anthropic’s revenue run-rate.[22]
Structured Summary Table of Key Deals (approximate figures as of late 2026 filings; sources stacked where multiple reports align):
- Broadcom SPV/TPU Leases: $63B financing SPV; ~$161.2B lease rent to Anthropic. Status: Committed/financed with Broadcom backstop. Risk: Broadcom (residual ~$43.5B senior), lenders (junior), Anthropic (payments). Sources: Bloomberg/WSJ/Reuters announcements + S-1.[1][23]
- Nscale (Nvidia GPUs): Up to ~$44.6B (possible +$45B). Status: Signed but unfunded (Nscale seeking capital). Risk: Primarily Nscale (financing/default). Sources: Bloomberg/FT/Nscale filings.[8][24]
- Google TPU/Cloud: Min. ~$121B (est. +$125–150B additional). Status: Committed, take-or-pay. Risk: Anthropic (payments). Sources: S-1 analyses.[13]
- Amazon (Trainium/AWS): Min. $110B. Status: Committed, take-or-pay. Risk: Anthropic. Sources: S-1.[15]
- Microsoft (Azure): Min. ~$45–49B. Status: Committed. Risk: Anthropic. Sources: S-1.[14]
- Nvidia Ecosystem + Equity: Indirect via partners + $10B IPO placement (part of $26B total). Status: Ongoing supply + equity commitment. Risk/Incentive: Aligned via equity and demand. Sources: S-1 amendments + placement disclosures.[16]
- TSMC: $500M IPO placement (supplier). Status: Committed. Risk/Incentive: Equity alignment. Sources: Reuters/WSJ/board approvals.[19]
These structures allow Anthropic to scale compute without full upfront balance-sheet burden while shifting some hardware/residual risks to suppliers like Broadcom and financing entities. Nscale remains the most contingent element. All figures derive from contemporaneous announcements and Anthropic/Nscale SEC filings as of November 2026.
Recent Findings Supplement (November 2026)
Broadcom $63B SPV financing (announced October 26-27, 2026) and related TPU leases form the largest recent structured deal. An SPV buys Broadcom XPU accelerators and networking gear, then leases them to Anthropic (deliveries starting 2027) under an earlier Google/Broadcom TPU compute agreement. Banks (BofA, Citi, Morgan Stanley) arrange first-lien debt (~$43.5B / ~69% of total); Blackstone provides committed second-lien. Broadcom supplies residual-value support/backstop equivalent to the first lien, creating estimated peak contingent exposure of ~$43B (or up to ~$43.5B) after deployment—effectively shielding senior lenders while exposing Broadcom to default/recovery shortfalls.[1][2][3]
This sits alongside ~$161.2B in aggregate 5-year Broadcom equipment lease rent (excluding interest) disclosed in Anthropic’s S-1 filings. An earlier ~$35B tranche (with Apollo/Blackstone) had Broadcom backstop exposure estimated at ~$29B.[4][5]
Nscale ~$45B (or $44.6B) deals remain largely unfunded as of mid-November 2026. August 2026 agreement covers first Monarch campus building (West Virginia, Nvidia Vera Rubin GPUs, ~460MW, six-year term, online late 2027). October 2026 talks advanced for a second ~$45B building (right of first refusal on a third). Nscale’s IPO prospectus and updates confirm no binding financing secured for the first site; Anthropic can terminate without liability if financing deadlines are missed. Nscale must raise substantial additional capital (site needs ~$30B+ more beyond ~$1B committed; total capex commitments rose sharply) and faces penalties or termination rights.[6][7][8]
Google TPU/Amazon arrangements are among the largest non-cancelable commitments in Anthropic’s October–November 2026 S-1 filings. Google/TPU-related: at least ~$114.5–121.3B (non-cancelable, true-up if underspent) plus Broadcom leases (~$161.2B rent) and estimated additional $125–150B for 2028–2031 capacity. Amazon: at least $110B (Trainium and other, non-cancelable over ~10 years to 2036). These form core fixed obligations alongside Microsoft (~$44.9–49.2B Azure, non-cancelable).[9][10][11]
Aggregate compute/infrastructure commitments reached ~$895–900.9B (as of/after Sept. 30, 2026), with ~$70.7B due in the following 12 months and ~$451.6B by end-2030. Nvidia GPU exposure includes SpaceXAI (up to $120.5B, largely cancellable on 90-day notice after early 2027), Microsoft, Nscale, and Lambda deals.[12][12]
Nvidia, Microsoft, and others participate in Anthropic’s IPO via a $26B private placement (at IPO price) alongside the public offering. Buyers: Nvidia and Google $10B each; AMD $2.5B; Broadcom, Micron, and Microsoft $1B each; TSMC $500M. IPO priced at $83/share on Nov. 18, 2026 (above $70–78 range; upsized to ~980M shares); trading began Nov. 19 under ticker ANT, raising ~$81.3B gross (largest on record). Suppliers gain equity upside and help ensure Anthropic can meet near-term payments amid ~$900B commitments.[13][14]
TSMC plans include a $500M stake in Anthropic (at IPO price) as part of a $1B total commitment (with OpenAI). Board approval Nov. 10, 2026; Anthropic’s filings flag Taiwan/TSMC concentration risk for advanced chips.[15][16]
Risk allocation summary: SPV senior lenders (banks/Blackstone) have Broadcom’s backstop as primary protection (Broadcom bears ~$43B peak exposure). Nscale financing risk falls on Nscale (Anthropic can exit). Many Nvidia-linked deals (e.g., SpaceXAI) carry cancellation rights. IPO equity purchases by suppliers create alignment incentives tied to Anthropic’s ability to service commitments.[3][17]
Structured Deal Summary (post-May 2026 announcements/filings only)
- Broadcom SPV / Leases: $63B SPV financing (Oct 2026); ~$161.2B lease rent. Structure: SPV buys/leases Broadcom XPUs; banks + Blackstone debt; Broadcom residual backstop (~$43B exposure). Status: Announced/arranged; deliveries 2027+. Risk: Broadcom primary backstop. Sources: WSJ/Bloomberg/NeoTeo filings.[1][2]
- Nscale Compute: ~$45B (Aug 2026, first building) + ~$45B talks (Oct 2026, second). Nvidia GPUs, Monarch WV. Status: Signed/talks; unfunded (Nscale seeking financing; termination rights). Risk: Nscale financing/execution. Sources: Bloomberg/FT/Nscale prospectus.[6][7]
- Google TPU / Broadcom + Amazon: Google ~$114.5–121.3B+ (non-cancelable); Amazon ≥$110B (non-cancelable). Part of multi-chip pool. Status: Amended/expanded in S-1. Risk: Fixed obligations. Sources: Anthropic S-1/A (Oct–Nov 2026).[9]
- Nvidia/Microsoft Ecosystem: SpaceXAI ≤$120.5B (cancellable); Microsoft ≥$44.9–49.2B; others (Nscale/Lambda). Status: Ongoing via clouds/providers. Risk: Partial cancellation exposure. Sources: S-1 filings.[11]
- IPO / Strategic Equity: $26B placement (Nvidia/Google $10B each, etc.); public raise ~$81.3B at $83/share (Nov 18 pricing). TSMC $500M Anthropic stake. Status: Priced/trading Nov 19. Incentives: Supplier alignment on ~$900B commitments. Sources: Anthropic/TSMC announcements.[13][14]
Total disclosed commitments ~$895–900.9B (latest S-1); ~$70B+ near-term payments. Much of the Nvidia portion carries more flexibility than Google/Amazon fixed deals. No major regulatory updates or new research publications identified in the results beyond these commercial and filing developments.
Report 5 Estimate the roughly $1 trillion in AI lab compute commitments (OpenAI, Anthropic, xAI, Meta, others) by counterparty (Oracle, Microsoft, AWS, CoreWeave, Google, Broadcom, AMD, Nvidia). Estimate how much is take-or-pay or non-cancellable versus conditional or contingent on funding, using 10-K/10-Q remaining performance obligations, lease disclosures, and credit-rating agency analysis. Estimate how much of the reported revenue of Nvidia, Oracle, CoreWeave, AMD, and Broadcom depends on a few AI labs, and show your assumptions and ranges explicitly. Include the Meta equity raise with Nvidia and AMD participation.
OpenAI and Anthropic alone account for well over $2T in headline multi-year compute and infrastructure commitments (primarily 2025–2035 or to 2036), far exceeding the query’s “roughly $1T” framing, which likely reflects an earlier or narrower subset of signed deals.[1][2][3] OpenAI’s commitments center on a mix of hyperscaler cloud capacity, custom silicon development/leases (Broadcom), GPU purchases, and specialized cloud (Oracle, CoreWeave). Anthropic’s are detailed in its amended S-1 (~$895–901B in purchase commitments and leases as of late September/early November 2026 disclosures). Meta adds another ~$100B+ across multiple providers, with xAI/SpaceXAI appearing in both as a provider and buyer. These figures are headline/contract values (minima, “up to” caps, or reported totals); actual cash outlays depend on delivery schedules, acceptance, and lab revenue growth to fund them.[4][5]
Breakdown by counterparty (approximate aggregates across major labs, with ranges reflecting “up to”/minima and partial overlaps):
- Microsoft (Azure): ~$295B+ (OpenAI incremental $250B Azure services commitment from Oct 2025 restructuring/expansion; Anthropic at least $44.9–49.2B minimum through ~2033).[1][2]
- Oracle: ~$300B (primarily OpenAI multi-year capacity, e.g., Stargate-related or Abilene-scale deals starting ~2027; Oracle’s own RPO reached $664B by Aug 31, 2026, with OpenAI cited as a significant but not sole portion).[1][6]
- AWS/Amazon: ~$250B+ (OpenAI expanded to ~$138B total including $100B addition; Anthropic $110B minimum through 2036).[7][2]
- Google Cloud: ~$121B+ minimum from Anthropic (through 2033, take-or-pay elements) plus additional TPU estimates of $125–150B (2028–2031, outside core table).[2][8]
- CoreWeave: ~$60B+ (OpenAI cumulative ~$22.4B across expansions; Meta ~$35B total including ~$21B expansion; Anthropic multi-year undisclosed but material).[9][10]
- Broadcom: ~$500B+ (OpenAI $350B custom silicon/10 GW deployment; Anthropic $161.2B TPU leases).[1][2]
- AMD: ~$110B+ (OpenAI $90B; Anthropic >$20B Helios/custom; Meta multi-GW partnership with warrants).[1][2]
- Nvidia: Direct chip commitments harder to isolate (~$100B cited for OpenAI in one model) but substantial via guarantees/backstops (~$105B+ for OpenAI Ohio campus leases) and neocloud capacity agreements; Nvidia is more a chip supplier with equity stakes and residual guarantees than a pure compute renter.[1][11]
Roughly 40–60%+ of the largest named commitments (especially Anthropic’s hyperscaler and Broadcom deals, totaling hundreds of billions) are structured as take-or-pay or non-cancellable (payable “regardless of actual usage,” with true-up payments if spend falls short; limited cancellation only for provider breach or specific undelivered tranches).[2][12] The balance is more conditional: subject to delivery/acceptance of capacity, counterparty financing milestones, or terminable on notice (e.g., Anthropic’s SpaceXAI up to $120.5B cancellable on 90 days’ notice after early 2027; portions of Lambda/Nscale deals billed only on delivery; some OpenAI/Oracle ramps tied to construction). Hyperscaler 10-Q lease disclosures (Microsoft $364.6B leases not yet started; Oracle $288B data center leases; Amazon $164.9B; Meta ~$356B) and credit agency analysis of off-balance-sheet obligations provide supporting context for the buildout scale, though labs’ own filings (Anthropic S-1) give the most granular counterparty and cancellability detail.[5][4] Many deals include equity investments or financing facilities from providers back to labs (circular element), which can make commitments more binding or provide funding support.
Revenue dependence on a handful of AI labs is high for specialized players and material (but diversified) for larger ones; single-lab concentration is a noted risk in filings and analysis. Oracle’s $664B RPO (as of Aug 31, 2026) has OpenAI as a substantial portion (estimates around half in some analyses); CoreWeave’s ~$132.5B backlog has OpenAI (~$22B contracts) and Meta as key anchors (OpenAI/Meta together a large share of its named deals).[13][7] Broadcom and AMD have significant exposure via custom/GPU deals with OpenAI/Anthropic/Meta (one analysis puts Nvidia at ~39% of frontier lab contracted compute share, Broadcom ~24%, AMD ~15%, with labs driving much of the growth).[14] Nvidia’s data center revenue (dominant portion of its totals) benefits broadly from AI buildout but is less concentrated on any single lab; it also holds guarantees and backstops tied to labs/neoclouds. Assumptions here include that “dependence” reflects contract/backlog attribution rather than recognized revenue to date (many deals ramp later); ranges are wide because not all providers break out lab-specific revenue, and actual realization depends on lab funding and usage.[15]
Meta’s ~$97.5B equity raise (upsized Oct 2026 offering, priced at $625/share with strong demand) included direct participation from Nvidia ($3B) and AMD ($1B), part of broader supplier interest up to $5B aggregate (not all binding); this deepens chip partnerships (e.g., expanded Nvidia Vera Rubin/Muse collaboration, AMD Helios testing and up to 6 GW deals) while funding Meta’s own multi-GW AI infrastructure.[16][16][17] This fits the pattern of circular financing where chip providers invest in (or guarantee for) customers who then commit to their hardware/cloud.
Key assumptions and ranges: Figures aggregate public announcements, S-1/10-Q disclosures, and analyst models (e.g., Tunguz for OpenAI vendor split); many are “up to” or estimated totals not fully reconciled in filings, so actual minima are lower and maxima higher. Delivery is phased (often 2027+ ramps), with ~$200B+ of Anthropic commitments noted as subject to termination/conditions in earlier disclosures (updated totals reflect evolution). Revenue dependence estimates are directional (high single-lab or few-lab concentration for Oracle/CoreWeave/Broadcom elements) and could shift with diversification or cancellations. Additional research on latest 10-Qs or specific credit analyses would refine conditional portions. Overall, the ecosystem shows strong commitments but material execution/funding risk concentrated among a few labs and providers.
Recent Findings Supplement (November 2026)
Meta's October 2026 equity raise includes direct participation from chip suppliers Nvidia and AMD, providing fresh capital for its AI buildout while deepening hardware partnerships.[1][2]
Meta priced $25 billion in Class A common stock (at $625/share, upsized from $22.5B) and $12.5 billion in mandatory convertible preferred depositary shares on October 27, 2026, with a planned $60 billion at-the-market (ATM) program for non-voting Class C shares starting in Q1 2027 (total potential ~$97.5 billion). Nvidia committed $3 billion and AMD $1 billion (Broadcom had indicated non-binding interest as part of a combined $5 billion ceiling, described as non-binding). Proceeds support general corporate purposes, including AI infrastructure amid 2026 capex guidance of $135–145 billion. The deals coincide with expanded deployments: Meta intends to broaden Nvidia Vera Rubin GPUs/CPUs, Spectrum-X networking, and BlueField-4; AMD Helios racks are installed and under testing for broader rollout.[3][4]
This structure mixes immediate and deferred equity (forward sales settle by Nov 19, 2027) with supplier equity stakes that align incentives on hardware roadmaps. For competitors, it highlights how AI labs can tap public markets and supplier balance sheets rather than pure debt or internal cash, potentially pressuring pure-play neoclouds or smaller labs without similar access.
Anthropic's amended S-1 filings (October–November 2026) provide the most granular recent disclosure of lab-side commitments, totaling ~$900.9 billion in purchase commitments and leases (substantially all compute/infrastructure) as of late September/early November updates, against an ~$81 billion revenue run-rate.[5][5]
Key counterparty breakdown (minimums or stated figures; some "up to"):
- Broadcom (TPU equipment leases): ~$161.2 billion (5-year terms, non-cancellable except lessor default; rent starts on "ready for service").
- Google Cloud: At least ~$121.3 billion minimum (take-or-pay/"regardless of actual usage," true-up for shortfalls; Apr 2026–Jul 2033).
- SpaceXAI (xAI-related): Up to $120.5 billion (delivery-based; cancellable by either party on 90 days' notice after initial periods ending ~Jan/Mar 2027).
- Amazon (AWS): At least $110 billion minimum (take-or-pay, true-up; May 2026–Apr 2036).
- Microsoft (Azure): $47.2–49.2 billion (mostly non-cancellable except uncured breach or undelivered tranches; Nov 2026–mid-2033).
- Nscale: Up to ~$44.6 billion (fees accrue on delivery/acceptance; into 2033; some financing conditions).
- Lambda: ~$41.3 billion (~$34.9 billion terminable; into 2034; delivery-based).
~ $186 billion of the total is subject to termination rights or conditions (e.g., counterparty financing or delivery milestones), down slightly from prior filings even as the headline rose. Additional estimated $125–150 billion for extra TPU capacity (2028–2031 deliveries, not in the main total) and other deals (e.g., AMD hardware >$20 billion, Fluidstack) sit outside the core figure. Leases not yet commenced and post-Sept commitments contribute heavily; ~$451.6 billion expected by end-2030.[6][5]
This data (from Anthropic's IPO registration amendments) is firmer on the buyer side than most seller disclosures and shows a mix of locked-in hyperscaler take-or-pay deals versus more conditional neocloud/GPU capacity. Implications for entrants: Labs face extreme leverage to utilization/revenue growth; suppliers with firm take-or-pay exposure (e.g., Google, Amazon, Broadcom) have stronger downside protection than those reliant on delivery milestones.
OpenAI's reported multi-year Oracle commitment (~$300 billion over ~5 years, ~4.5 GW, slated to start 2027) remains a focal point of counterparty concentration risk, referenced in recent Oracle filings and analyst commentary.[7][8]
Oracle's remaining performance obligations (RPO) reached $664 billion as of Aug 31, 2026 (up sharply YoY; ~13% expected in next 12 months, ~37% in months 13–36, etc.). Analysts (including S&P references) estimate ~$300 billion (~45%) tied to OpenAI. Oracle also disclosed ~$248 billion in future lease commitments (substantially data centers/cloud capacity, 15–19 year terms, commencing through FY2028/2029). The Oracle-OpenAI deal is part of broader Stargate-related efforts but carries execution risk given OpenAI's cash burn and funding needs.[9][9]
Oracle's RPO growth outpaces near-term revenue conversion, highlighting lease-duration risk on the supplier side. For market participants, this underscores how a few labs drive outsized backlogs at infrastructure providers, with limited public detail on cancellation terms.
CoreWeave and AMD show notable revenue/backlog dependence on a handful of AI labs or hyperscalers, per recent 10-Q/10-K disclosures, while Nvidia's direct AI chip share has declined amid diversification.[10][11]
- CoreWeave: Microsoft accounted for ~67% of FY2025 revenue (diversifying in 2026 quarters, with top customers at 30–45% ranges in recent periods but still concentrated). OpenAI contracts total up to ~$22.4 billion (multiple 2025–2026 deals). RPO/backlog ~$132.3 billion as of Sep 30, 2026 (~98% of revenue from committed contracts). Debt ~$35–50 billion.[12][13]
- AMD: Issued warrants tied to OpenAI/Meta chip deals (up to ~10% stakes possible on milestones), triggering a $1.07 billion non-cash Q3 2026 charge as shipments ramp. Meta equity participation ($1B) and OpenAI commitments add exposure.[14]
- Nvidia: Share of contracted AI compute infrastructure for frontier labs fell to ~39% (from 83% historical chip revenue share), with Broadcom (~24%), Google TPU (~16%), AMD (~15%), and Trainium (~6%) rising. Data center revenue remains dominant but customer concentration disclosures emphasize broader ecosystem.[11][15]
- Broadcom: Prominent via large Anthropic TPU lease exposure.
Overall ranges and assumptions (post-May 2026 data only): Aggregated trackers (e.g., aicomputedeals.com as of mid-Nov 2026) show OpenAI deals totaling hundreds of billions to ~$1.6T headline across ~21 deals and Anthropic ~$549B across 15, with combined lab commitments often cited in the $1T+ range when including Meta/xAI/etc. (many overlapping or estimated). Firmness varies: hyperscaler deals (Google, Amazon, Microsoft, Broadcom) skew take-or-pay/non-cancellable (~$440B+ in Anthropic alone); neocloud/GPU deals more conditional on delivery/financing (~$186B+ terminable in Anthropic case). Oracle RPO concentration and CoreWeave debt amplify supplier-side risks. Meta's raise adds a circular equity element. These figures derive from S-1s, 10-Q/10-K RPO/lease notes, and press on specific contracts; actual recognized revenue depends on utilization, delivery, and funding. Ranges reflect "up to"/minimum distinctions and analyst estimates (e.g., Oracle-OpenAI split). Additional research on full 10-K details or credit agency reports would refine conditional portions.[16][17]
New data since May 2026 centers on Anthropic's detailed S-1 disclosures, Meta's supplier-backed raise, updated RPO/backlog figures, and concentration analyses—shifting focus from headline announcements to execution, firmness, and balance-sheet implications.
Report 6 Present the strongest arguments critics make comparing Nvidia to GE Capital, Lucent/Nortel-era telecom vendor financing, and the dot-com bubble, drawing on named analysts, short sellers, and accounting commentary. Present the bull rebuttals (end-user demand, inference revenue growth, Nvidia's cash generation, differences in structure and scale) with evidence. This is the disconfirming-evidence prompt: also identify failure modes (funding drying up, an IPO window closing, GPU depreciation, power constraints, lab revenue shortfalls, concentration contagion), historical analogies with measurable outcomes, and signposts that would show who is right.
Nvidia's financing activities—mobilizing third-party capital and providing guarantees tied to customer purchases of its GPUs—have drawn comparisons to GE Capital's evolution into a risky lender, Lucent/Nortel's vendor financing that inflated telecom sales before write-downs, and dot-com-era circular or hype-driven revenue recognition.[1][2][3]
Short sellers and analysts such as Jim Chanos (Enron short), Michael Burry, Culper Research (Christian Lamarco), and commentary referencing BIS reports highlight these parallels. They point to Nvidia's stakes in customers (e.g., OpenAI, CoreWeave, xAI, neoclouds), guarantees (up to ~$118B recently disclosed, with a large OpenAI-linked component), and arrangements where investments or backstops enable chip purchases.[4][5][3]
Critics' Core Arguments
- GE Capital parallel (Gilder Report and related analysis): Nvidia is shifting from pure chip designer toward financing infrastructure via partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman, and KKR to mobilize >$500B in third-party capital. Critics note potential skin-in-the-game (reports suggest ~25% on some deals) and residual value guarantees that turn depreciating GPUs into "financeable assets," risking balance-sheet exposure if customers default or tech obsolesces, akin to GE Capital's growth from product financing into a systemic risk for GE.[1][6]
- Lucent/Nortel vendor financing: Lucent committed ~$8.1B (or up to $7B+ in some tallies) in customer credit/guarantees by 2000 (~24% of revenue); Nortel and Cisco added billions more. This boosted reported sales as cash-strapped telcos bought equipment with vendor money, only for loans to sour post-bubble (Lucent provisions of $2.2B in 2001; stock collapse from ~$84 to pennies; Nortel bankruptcy 2009). Nvidia's ~$110B+ direct investments/guarantees plus GPU-backed debt (e.g., $100B OpenAI tranche structure, CoreWeave deals) represent far larger relative exposure (~67% of ~$165B LTM revenue vs. Lucent's scale), with "circular" elements via equity stakes enabling purchases. BIS explicitly drew the telecom parallel for AI upstream vendors financing downstream demand.[2][7][3][8]
- Dot-com bubble and accounting concerns: Overvaluation driven by hype and circular flows (Enron-style off-balance-sheet or round-tripping concerns raised by Burry/Chanos); revenue from related-party or financed deals that may not reflect sustainable end-user demand. Chanos noted investments in loss-making entities to drive chip orders; Burry flagged suspicious recognition and a "protection racket" dynamic. Recent scandal (Huang ouster amid China exposure allegations flagged earlier by Culper) amplified distrust.[4][9]
These views gained traction amid Nvidia's stock decline (~35% from Oct 2026 high), with shorts realizing ~$24B in paper profits.[10]
Bull rebuttals emphasize real end-user demand from hyperscalers and enterprises, accelerating inference workloads, Nvidia's capital-light model with massive cash generation, and structural differences (third-party capital, rapid pay terms, diversified growth beyond pure vendor financing).[11]
Bull Case Evidence
- End-user demand and inference growth: Q3 FY2027 (ended Oct 2026) revenue hit $115.8B (+103% YoY), data center $107.6B (+110% YoY). Hyperscaler revenue ~$60.1B (+71% YoY); AI clouds/industrial/enterprise ~$47.5B (+195% YoY, contributing more incremental growth). Vera Rubin ramp and Blackwell Ultra cited; sovereign AI and physical AI/edge also expanding. Guidance ~$130B for Q4 (no China data center assumption). Broader cloud backlog cited at $2.5T.[12][13][14]
- Cash generation and scale: Operating cash flow strong (Q3 ~$48.3B, up significantly YoY; TTM/annual figures in the $100B+ range in recent periods). Free cash flow robust (Q3 ~$43.8B). High margins (~75% gross) and capital-light model (low capex relative to cash flow) support buybacks/dividends while funding selective investments. Net debt low; balance sheet can absorb risks better than 1990s peers.[15][16]
- Structural differences: Nvidia emphasizes third-party capital mobilization (not primarily its own balance sheet); customers pay in ~53 days (vs. multi-year vendor loans historically); guarantees often conditional (e.g., post-construction, customer financing secured first) and declining over leases. Recent tightening under interim CEO Colette Kress ("disciplined/selective," "very strong look" at neocloud support; some commitments reduced or clarified). Scale, profitability, and real AI adoption (training + inference) differ from dot-com speculation or telecom overbuild.[17][18]
Failure modes include funding/IPO windows closing amid higher rates or risk aversion, rapid GPU depreciation amid tech cycles, power/grid constraints delaying deployments, lab/startup revenue shortfalls or defaults triggering guarantees, and concentration/contagion from hyperscaler/neocloud reliance or related-party exposures.[19]
Key Risks and Historical Context
- Power constraints: US large-load (data center/AI) queues ~175 GW (AIxEnergy model expects ~44 GW realized; 75% speculative/blocked). PJM capacity shortfalls, ERCOT queues hundreds of GW, transformer lead times 2.5–5 years, interconnection medians ~61 months. Local opposition/moratoria widespread. Goldman flags power as "forever bottleneck."[20][21][22]
- Financing and contagion: Recent Nvidia tightening on guarantees/neocloud support; broader AI firm circular investments noted by BIS as risk if end demand slows. GPU-backed loans and neocloud debt (e.g., CoreWeave leverage) vulnerable to utilization shortfalls.
- Depreciation and demand shortfalls: Next-gen platforms accelerate obsolescence; inference may fragment share over time.
- Historical outcomes: Lucent revenue collapsed post-2000 (~$38B peak to $8B); massive write-downs, stock wipeout, Alcatel merger. Nortel bankruptcy 2009 after similar financing-fueled sales. Dot-com infrastructure vendors saw prolonged pain when capital dried up.[3][23]
Signposts that would favor critics (bubble/financing unwind) vs. bulls (sustainable growth): Slowing sequential revenue or guidance misses; rising bad debt provisions or guarantee payouts; hyperscaler capex pauses or cancellations; power-related deployment delays showing in utilization metrics or customer commentary; AI startup/lab funding/IPO drought or defaults; China exposure realizations or regulatory hits; margin compression from mix or competition. Conversely, sustained >50–70% YoY growth with broadening customers, stable/high utilization driving inference revenue, and financing deals performing without losses would support the bull view. Recent earnings beat amid leadership change provides one near-term data point, but financing scrutiny and macro/power factors remain key watches.[17]
This analysis draws on named sources and recent filings/earnings as of mid-November 2026. Markets and fundamentals evolve rapidly; cross-reference primary SEC filings and company disclosures for the latest.
Recent Findings Supplement (November 2026)
Nvidia's August 2026 launch of a $500B+ third-party AI infrastructure financing platform with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR has intensified comparisons to GE Capital's customer-financing model.[1][2]
Critics argue this shifts Nvidia from chip seller to de facto banker, risking balance-sheet exposure if GPU collateral values or customer revenues falter. George Gilder highlighted how GE Capital began as legitimate customer support for expensive equipment but ballooned into a risk that nearly sank GE, warning Nvidia's initiative could distort its self-understanding despite third-party capital.[1] Bank of America estimated Nvidia's commitments at ~$70B equity stakes and $230B guarantees as of August 2026.[2][3]
- The platform targets AI factories as an "investable asset class" with repeatable institutional financing; Nvidia limits its own guarantees to a maximum of 25% in some structures.[4]
- By November 2026, no capital had been deployed under the pledges, and deals were still in discussion.[5]
- This creates a mechanism where Nvidia's profits and guarantees help lower customer cost of capital, but critics see circularity if end demand disappoints.
For competitors or entrants: The structure raises the bar for alternative chip or infrastructure plays, as Nvidia leverages its cash flow and relationships to crowd in capital; challengers must demonstrate independent financing viability or superior economics without similar backstops.
Short sellers and analysts (including George Noble and Jim Chanos) have explicitly likened Nvidia's customer financing, equity stakes, and guarantees to Lucent/Nortel-era telecom vendor financing and Cisco's dot-com practices.[6][7][8]
Noble warned in November 2026 that Nvidia helps finance the same ecosystem buying its hardware, creating a dangerous feedback loop akin to Cisco's vendor financing that contributed to an 89% stock drop when customers failed.[6] Lucent's stock fell from ~$84 to $0.55 amid customer defaults on financed equipment; Nortel had similar ~$4B+ exposures.[9][7]
- Nvidia's accounts receivable reached $63.1B (up from prior levels) with collection times lengthening to 60 days (from 45), and guarantees hit ~$108.5B (including $105B tied to an OpenAI Ohio campus).[10][11]
- A BIS report (cited in November coverage) noted ~30% of AI investment deals by value involved AI firms on both sides (55% of incoming investments from other AI firms), flagging risks of artificially inflated demand and potential volatility if real revenue slows.[12]
- GPU depreciation debates center on collateral value: Nvidia cites hyperscaler shifts to 5-6 year schedules, while bankers reference 3-4 year useful lives.[13]
Implication: Entrants face heightened scrutiny on customer credit quality and must avoid or transparently manage similar circular exposures; the scale (Nvidia's guarantees exceed historical telco examples relative to revenue) amplifies systemic concerns if concentrated neocloud or lab counterparties weaken.
Bulls counter that Nvidia's model differs structurally due to massive end-user demand, inference-driven revenue growth, and exceptional cash generation, with third-party capital and stronger counterparties mitigating risks.[14]
Gene Munster argued in a November 2026 debate that the economics are sound—roughly $250B input could yield $420B in returns at 70% margins—unlike dot-com circular ad deals that collapsed when capital dried up without real buyers.[14] Nvidia's Q3 fiscal 2027 results (reported November 17-18, 2026) showed revenue doubling year-over-year to $115.8B (data center $107.6B), with strong free cash flow (~$43.7B+ in the quarter) supporting buybacks and dividends.[15][16]
- Guidance for Q4 at ~$130B beat expectations; management emphasized disciplined review of arrangements post-leadership change.[16]
- Hyperscalers (stronger credits) are prioritized over riskier neoclouds; real inference and training demand underpins growth, with projections of continued 70-75%+ annual revenue increases in coming periods.[2]
- Valuations have compressed sharply (forward P/E around 11x post-selloff), which bulls like UBS and BofA view as overly pessimistic given earnings momentum.[17]
For market participants: Evidence of sustained hyperscaler capex and actual token/inference utilization would validate the bulls; Nvidia's cash machine enables ecosystem support without the same fragility as 1990s vendors.
Recent developments post-May 2026 highlight failure modes materializing around funding access, customer concentration, and leadership transition.[18]
Jensen Huang's November 2026 resignation amid scandal accelerated a shift under interim CEO/CFO Colette Kress toward "disciplined" and "rigorous" scrutiny, with expectations of pivoting financing away from neoclouds toward larger hyperscalers.[2][19] Neocloud bonds (e.g., CoreWeave yields spiking to ~15%, Nebius to >12%) reflect tightening credit as Nvidia support signals wane.[18]
- GPU power constraints and depreciation risks remain live (useful life debates affect lending collateral).[13]
- Lab/neocloud revenue shortfalls could trigger guarantee payouts; concentration in a few counterparties amplifies contagion.
- Historical outcomes: Lucent/Nortel saw massive losses, stock collapses (99%+ in Lucent's case), SEC issues, and eventual acquisition/bankruptcy; Cisco dropped sharply but survived with a different trajectory.[9][7]
Signposts favoring critics include further AR/FCF divergence, new guarantee draws, neocloud distress or defaults, hyperscaler capex slowdowns, or slower-than-guided growth. Signposts favoring bulls include stable or rising inference metrics, successful third-party deployments, and sustained or accelerating revenue beats with contained credit exposure.[20]
Nvidia's November 2026 earnings showed resilience but underscored the financing pivot; the stock's ~30-40% drop from October peaks reflects these tensions predating and amplified by leadership news.[21]
Overall, the debate centers on whether Nvidia's cash generation and demand reality differentiate it sufficiently from historical vendor-financing failures, or if the scale of circular support creates similar vulnerabilities. Recent tightening provides a live test case.