Nvidia is due to take a further $1.0 billion of Nscale’s convertible notes or shares on or around 16 November, two days from now, according to the AI cloud company’s amended IPO filing of 12 November: “The NVIDIA Sale will close on or around November 16, 2026.” Nvidia’s board told CNBC on 13 November that it is in the process of selecting a search firm for its chief executive role (our explainer).
It is one of several funding commitments Nvidia has made to AI clouds. Under a credit-support and revenue-sharing model Nvidia set out on 1 July, Sharon AI and Firmus agreed to deploy up to 210,000 Nvidia GPUs on 432 megawatts, by our sum of the companies’ announcements.
Business Pill · Concentration Explained
An evergreen Business Engineer explainer on concentration, where a few holdings account for most of the value. It teaches the general idea only and does not discuss Nvidia, Nscale or the AI cloud deals.
The key insight: As we read the documents together, Nvidia funds AI clouds through three instruments with different clocks: a $1.0 billion Nscale note due on or around 16 November, credit support for up to 210,000 GPUs at Sharon AI and Firmus, by our sum, and $3.5 billion of lease guarantees in its 10-Q.
Nscale’s S-1/A says that in September 2026 it agreed subscriptions of $3.36 billion: $2.36 billion of unsecured convertible loan notes issued to several investors, including NVIDIA, and “a further $1.0 billion of unsecured convertible loan notes or Non-Voting Shares (as applicable) that will be issued to NVIDIA”. That is about 30% of the round, by our arithmetic. The notes convert automatically when the offering completes, the filing says (our earlier piece on the round).
The same filing says Nscale issued a warrant instrument to NVIDIA on 14 October 2025 “in connection with a net lease agreement for a facility in Ward County, Texas”, and that NVIDIA became a holder of more than 5% of its share capital through the Series B financing.
GPUs covered by the two AI cloud deals Nvidia named under its credit-support and revenue-sharing model: Firmus, up to 170,000 at a 360 MW campus in Batam, and Sharon AI, up to 40,000 on 72 MW in Australia. Total is our sum of the companies’ announcements.
The Credit-Support Model
Nvidia set out the model in a 1 July blog post by Colette Kress, then its finance chief and now its interim chief executive, and Raj Mirpuri. AI clouds sell Nvidia-powered cloud services, “with NVIDIA earning both standard productrevenue and a share of the cloudrevenue on the supported capacity”. The post named Sharon AI and Firmus as among the first companies to work with Nvidia on it.
Sharon AI, listed on Nasdaq, said in a 12 June release filed with the SEC that its six-year agreement covers 72 megawatts in Australia and up to 40,000 Grace Blackwell GB300 GPUs, with economics aligned “through a revenue-sharing and credit-support model”.
Firmus said its partnership runs through 2034 and covers up to 170,000 Nvidia accelerators at a 360-megawatt campus in Batam, Indonesia. It expects US$25 billion to US$30 billion from committed offtake agreements in the first six years, about $4.2 billion to $5.0 billion a year, by our arithmetic.
What the 10-Q Already Shows
Nvidia’s 10-Q for the quarter to 26 July lists $3.5 billion of guarantees on AI cloud partners’ data centre leases, within $108.5 billion of maximum guarantee exposure, as our guarantees piece and our backstop piece reported. On the earnings call Kress said “we get paid twice”: “once on the hardware sale, and again through the share of rental revenue”, Fortune reported.
The New York Times reported on 13 November that no loans have closed under the $500 billion AI financing effort Nvidia announced with six lenders in August. The Information reported on 12 November that Kress, as finance chief, pushed back on a proposed $250 billion credit backstop on behalf of OpenAI, our backstop piece noted.
Other AI clouds are leaning on their customers. CoreWeave said it expects about $9 billion of prepayments and deposits in the next three months, and Nebius $17 billion to $20 billion of customer prepayments in 2027 (our neocloud funding piece).
The Structural Read
As we read the Sharon AI and Firmus terms, Nvidia is paid twice on supported capacity, from the hardware sale and a share of cloudrevenue, so its return on that support rises and falls with how fully those GPUs are rented.
As we read the Nscale filing, the NVIDIA Sale is the one commitment here with a published closing date, which makes its completion an observable data point during Nvidia’s chief executive search.
Colette Kress and Raj Mirpuri, NVIDIA blog, 1 July 2026
“Emerging AI companies historically have had limited access to capital-intensive infrastructure, with even long-term commitments insufficient to unlock financing for compute.”
Three Implications
FOR AI CLOUDS Sharon AI and Firmus announced credit-support deals for up to 210,000 Nvidia GPUs on 432 MW, by our sum of their releases.
FOR NSCALE’S IPO The $1.0 billion NVIDIA Sale is about 30% of Nscale’s $3.36 billion round and closes on or around 16 November, its S-1/A says.
FOR NVIDIA’S SHAREHOLDERS The 10-Q lists $3.5 billion of guarantees for AI cloud partners within $108.5 billion of maximum exposure, as our earlier piece reported.
The framework’s central line: “The $1T capex cycle is being underwritten by a $1T concentrated demand commitment from two counterparties whose combined free cash flow is deeply negative.”
As we read it, Nvidia’s credit support widens the set of buyers beyond those two counterparties by lending its own balance sheet to smaller AI clouds. The framework does not discuss Nscale, Sharon AI, Firmus or this week’s filing.
Nscale’s $3.36 billion convertible financing, per its S-1/A of 12 November: $2.36 billion of notes issued to several investors including NVIDIA, and a further $1.0 billion to be issued to NVIDIA on or around 16 November, about 30% of the round by our arithmetic.
What Is Not Established
Neither Sharon AI nor Firmus has disclosed the amount of credit support Nvidia provides or the share of cloudrevenue it receives. Nscale’s filing leaves the offering price blank, so the number of shares the NVIDIA Sale becomes is not yet set. None of these documents ties the commitments to Nvidia’s leadership change.
We read the filings and releases named here; Nvidia, Nscale, Sharon AI and Firmus were not contacted for this piece.
This Week In The AI Supercycle
AI Lost Its Central Banker, the Lab Oligopoly, and the Future of Open Source
This week’s issue: the Nvidia leadership change, read as the AI build-out losing its central banker: who now finances the GPU landlords, backstops the data centres and sponsors open models, and why two labs are becoming the bank.
Nvidia’s next funding step for an AI cloud has a date: $1.0 billion into Nscale on or around 16 November, the filing says. Behind it sit credit-support deals covering up to 210,000 GPUs at Sharon AI and Firmus, by our sum, and $3.5 billion of lease guarantees for AI cloud partners in the 10-Q, while the $500 billion lender effort has closed no loans, the New York Times reported. The Business Engineer’s weekly issue maps these among the roles Nvidia played across the AI stack.
A note on sourcing. This analysis combines Nscale’s S-1/A, Nvidia’s 1 July blog post, Sharon AI’s SEC-filed release and Firmus’s release, re-read on 14 November, with five of our earlier pieces, each built on its primary source or named outlet. Sums and shares are ours. This piece makes no claim about any company’s conduct.
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 (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
