SemiAnalysis estimates that NVIDIA has the capacity to support up to 46 gigawatts of AI data-centre capacity with its backstops and guarantees, against 240 gigawatts it says is coming. In a clip it posted on X on 11 October 2026, the research firm says: “The lending market has to evolve to take risks outside of backstops. It’s really the biggest question of our time.”
The clip comes from SemiAnalysis’s Ep. 037, published on 9 October. In the full episode, the speaker frames the 46 gigawatts as “not our forecast” and says of the hyperscalers: “We haven’t done it for the hyperscalers yet”. The text of the X post leaves both caveats out.
The post opens: “The AI buildout needs more capital than backstops can cover.” It then quotes the episode on who lends: “There’s bank lenders, insurance lenders, they don’t want to take a lot of risk. You’ve got private equity, they want to take a lot of risk. You’ve got private credit, they’re in the middle.”
On the structure, the post quotes: “What you do is you tranche risk. If there are losses, the equity will take the loss first, then the B piece, and then part of the A2. The loss has to be so severe to affect the senior tranche, which is how you can create different risk profiles.”
On the scale, it quotes: “We think NVIDIA has the capacity to support up to 46 gigawatts. That’s still not enough. There’s 240 gigawatts that’s gonna come in.” When we read the post on 11 October, it had about 18,800 views.
SemiAnalysis’s own gigawatt figures, each from the source and date shown: about 6.5 GW backstopped by NVIDIA today by its count (11 September), up to 46 GW NVIDIA could support (Ep. 037, 9 October; ‘not our forecast’) and 240 GW it says is coming (X clip, 11 October).
Business Pill · TRANCHES
A one-minute explainer of tranches: one pool of money cut into slices that take losses in a set order, so careful and bold lenders can fund the same project. It teaches the general idea with made-up numbers and says nothing about any company in this story.
The key insight: As we read it, SemiAnalysis is making two separate points. One is a size gap: by its estimates, the backstops NVIDIA could offer cover a fraction of the capacity it expects. The other is a structure: lenders with different appetites for risk would take different slices of the same loans, with equity taking the first loss. Both are SemiAnalysis’s analysis, and it calls the 46 gigawatts “not our forecast”.
Who Is Speaking
The episode’s description says Daniel Nishball “kicks off our new Compute Capital Markets group by following the money through Nvidia’s $497B backstops”, with Jordan Nanos bringing the rental data. In the auto-generated captions, the passage with the 46 gigawatts is spoken to Nanos (“here’s the kind of conclusion of this, Jordan”), which is why we read the speaker as Nishball.
The captions say the tranche structure is a sketch for NVIDIA’s $500 billion capital partnership: “this is what we think it could look like”. In that sketch, the B piece “doesn’t have any residual value guarantee”, and NVIDIA gives “a 25% residual value guarantee in this 500 billion partnership”.
Where the 25% Comes From
NVIDIA’s chief executive Jensen Huang announced the programme in an article on X on 10 August 2026: partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR (financing platforms) “to establish independent financing platforms designed to mobilize over $500 billion of third-party capital to support the buildout of AI infrastructure over time.”
He wrote that “NVIDIA may provide a residual-value support mechanism for up to 25% of an opportunity, assessed carefully on a project-by-project basis.” He also wrote that the $500 billion figure “represents aggregate third-party capital that these platforms are designed to mobilize over time” and is not NVIDIA revenue.
SemiAnalysis’s 11 September newsletter makes the same tranching argument in writing: “We think that the funds could tranche the lending to create different pools of risk, with the first loss being taken by riskier equity/mezzanine style lending and the last loss taken by more senior creditors.”
The order of losses in the tranche structure SemiAnalysis sketches for NVIDIA’s $500 billion capital partnership, from its X clip, Ep. 037 and its 11 September article. No tranche sizes were given. The 25% residual-value support is from Jensen Huang’s 10 August article.
What the Written Analysis Adds
The 11 September article, “Nvidia’s Backstop Universe – Heads I Win, Tails Who Loses?”, gives a smaller figure for today: “By our count, Nvidia today backstops ~6.5 GW of capacity, most of which is not yet built.” It also says its datacenter model has Microsoft, Meta, AWS and Oracle leasing roughly 15 gigawatts of third-party capacity in 2026, and that it expects them to lease more than 35 gigawatts by 2028.
The article puts the overall investment need at “~$11T of cumulative capex from CY24 through CY29”. On its own estimate of how much more NVIDIA could support, it says: “To be clear, this does not represent our base case or a definitive forecast for where we think Nvidia will go with its various backstops.”
SemiAnalysis’s materials use different totals for what NVIDIA has already committed. The article counts $530 billion of “gross off-balance sheet guarantees” in NVIDIA’s 10-Q; the episode description speaks of “Nvidia’s $497B backstops”. Our own read of the 10-Q found three tables that add up to $530.5 billion by our arithmetic, and the filing does not use the term off-balance sheet.
What SemiAnalysis Is Building
The firm’s product page for Compute Capital and Markets, marked “Coming soon”, says it will “map every funding, offtake, backstop, guarantee, credit-enhancement and off-balance-sheet arrangement across the AI complex” and lists “500+ debt instruments across hundreds of issuers”.
The Structural Read
The 46 gigawatts is a ceiling SemiAnalysis puts on NVIDIA’s capacity to back projects, not a count of what it has backed. Its 11 September article put what NVIDIA backstops today at about 6.5 gigawatts, “most of which is not yet built”.
As we read it, the tranche structure moves the question from NVIDIA’s balance sheet to lenders’ risk appetite. NVIDIA’s own description of its support is narrower: residual-value support of up to 25% of an opportunity, assessed project by project, which Jensen Huang wrote is designed to complement independent underwriting.
SemiAnalysis also says the exercise has not yet been done for the hyperscalers. Its article describes Microsoft, Meta, AWS and Oracle as leasing far more third-party capacity than NVIDIA backstops.
SemiAnalysis, clip posted on X, 11 October 2026
“The lending market has to evolve to take risks outside of backstops. It’s really the biggest question of our time.”
Three Implications
LENDERS In SemiAnalysis’s sketch, banks and insurers sit in senior tranches that are hit only if losses are severe, while private equity takes the first loss and private credit sits in the middle.
NVIDIA Jensen Huang wrote that NVIDIA’s residual-value support is limited to up to 25% of an opportunity and is designed to complement, not replace, independent underwriting.
READERS OF THE NUMBERS SemiAnalysis’s totals for what NVIDIA has already committed differ between its article ($530 billion) and its episode description ($497 billion); the 46 and 240 gigawatt figures are its estimates.
The analysis puts it this way: “the capex race is no longer a corporate spending cycle. It is a sovereign-scale financing problem dressed in tech-company clothing — and the financial architecture is starting to invert.”
As we read it, SemiAnalysis’s clip describes one form that financial architecture could take: a chip supplier offering limited residual-value support, and lenders with different risk appetites taking different tranches of the debt behind AI data centres.
What Is Not Established
Neither the clip, the captions we read nor the free portion of the article explains how SemiAnalysis arrived at 46 gigawatts or over what period the 240 gigawatts would arrive. The episode says the same exercise has not yet been done for the hyperscalers, whose leases the article describes as far larger than NVIDIA’s backstopped capacity.
The captions are auto-generated, and we did not read the paid portion of the article. The 46 and 240 gigawatt figures are SemiAnalysis’s estimates, not company disclosures.
Business Engineer Framework
The AI Capex Map & The State of AI Hyperscalers
A Business Engineer analysis of how the AI capex cycle is financed: who commits the money, where it concentrates and how the financial architecture is changing.
SemiAnalysis estimates that NVIDIA could support up to 46 gigawatts of capacity with backstops and guarantees, against 240 gigawatts it says is coming, and argues that lenders will have to take risk outside those backstops, through tranches in which equity takes the first loss. In the full episode it calls the 46 gigawatts “not our forecast”.
A note on sourcing. On 11 October 2026 we read SemiAnalysis’s post on X and the description and auto-generated captions of its Ep. 037 (the passage from 74:00 to 78:30), the free portion of its 11 September newsletter article, its Compute Capital and Markets page, and Jensen Huang’s 10 August article on X. We did not read the paid portion of the article. The gigawatt and dollar estimates are SemiAnalysis’s unless marked as our arithmetic. Nothing here is a forecast, and nothing here is financial or investment advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur, HackerNoon and Search Engine People. Find him on LinkedIn and Substack. See how we source.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
