Nvidia's $50 Billion Bet: How Chip Maker Finances Its Own Demand

Nvidia has deployed nearly US$50 billion into artificial intelligence laboratories that serve as major buyers of its chips, while securing pledges for more than $500 billion in additional funding. During an earnings call on August 26, Colette Kress, Nvidia's chief financial officer, disclosed to analysts that revenue from these backed laboratories will represent approximately one-quarter of the company's business in the coming year.

The structure has drawn comparisons to circular financing, a term Nvidia itself raised before analysts brought it up. Kress acknowledged on the earnings call that the company understands the magnitude of backing it provides and recognizes that some observers would characterize it as circular financing. However, she indicated Nvidia views the arrangement in a different light.

The mechanics work as follows: Nvidia invests capital into an AI laboratory. That laboratory deploys the funds or leverages credit access that Nvidia's involvement provides to construct a data centre. Nvidia processors fill the facility. The resulting purchase translates into Nvidia revenue. As Nvidia's valuation and cash reserves expand, the company invests again.

What Nvidia has committed

Kress herself disclosed the specifics. She announced that Nvidia has established partnerships with six investment organizations—Apollo Global Management, BlackRock, Blackstone, Brookfield Asset Management, Goldman Sachs, and KKR—to establish financing vehicles designed to mobilize more than $500 billion in external capital for laboratory infrastructure development.

Additionally, Kress stated that Nvidia has secured land, electrical capacity, and construction resources through SB Energy exclusively for Nvidia hardware deployment. The initial phase will deliver 4.25 gigawatts of capacity and serve OpenAI. Kress estimated OpenAI's total and prospective Nvidia compute requirements at approximately 12 gigawatts through 2030. For a second unnamed laboratory, Nvidia will furnish credit backing for nearly two gigawatts of capacity.

Nvidia's financial disclosures characterize these partnerships as contingent on final agreements, indicating that definitive contracts remain unsigned. The $500 billion figure represents a stated objective rather than committed funds.

Nvidia is also extending its brand to smaller cloud service providers. According to Kress, the company pledges to lease a portion of a provider's infrastructure itself, furnishing the provider's creditors with assured revenue to underwrite lending. In return, Nvidia receives a percentage of earnings the provider generates above that baseline. Kress characterized this as a dual revenue stream for Nvidia—once through equipment sales and again through rental payments.

Why Nvidia rejects the circular financing label

Kress offered three explanations, each warranting consideration alongside the financial figures.

Third-party lenders independently evaluate each transaction on its particulars, she stated, and Nvidia itself is not originating loans. The processors Nvidia delivers reach customers meeting investment-grade standards or supported by investment-grade entities. Should a customer encounter financial distress, the equipment can be reallocated to another purchaser, a factor Kress cited as limiting Nvidia's financial exposure.

Kress also articulated why the laboratories require this assistance. These entities face computational demand that surpasses their financial capacity, she explained. As emerging organizations lacking the extended service agreements and credit histories that conventional lenders demand before financing data centre projects, their expansion is constrained not by customer appetite or technological capability, but by computing access.

The principal vulnerability involves a scenario where one of these laboratories cannot meet its obligations. Nvidia would forfeit both the transaction and its investment simultaneously. Kress countered that the equipment would find another buyer—a proposition that remains valid only when demand outpaces available supply, which Nvidia maintains is the current market condition.

Vivek Arya of BofA Securities questioned Jensen Huang regarding how the company reconciles financing laboratories developing proprietary processors, citing OpenAI's Jalapeño chip. Huang responded that Nvidia provides a platform deployable across any cloud environment throughout an AI system's operational span, whereas competing processors are optimized for specific applications. Regarding the investment strategy, he remarked that his sole regret was not committing more resources and doing so earlier.

The agent assumption underneath it all

Kress informed Morgan Stanley's Joseph Moore that an agent requires between 15 and 100 times the computational resources of a person operating the identical system. Huang stated his conviction that AI transitioned to being predominantly agentic within the previous month. Nvidia furnished no supporting data for this assertion.

Drawing on this premise, the company projected $108 billion in quarterly revenue and indicated preliminary expectations for approximately 70% annual expansion through January 2028, a projection Kress attributed to supply constraints rather than demand limitations.

Kress also flagged that memory costs are accelerating beyond Nvidia's initial projections. She guided gross margins to 74% for the current quarter, with a projected floor of 71% to 72% in the fourth quarter, attributing memory shortages substantially to the AI infrastructure expansion.

Nvidia will report financial results again on November 17. Kress did not identify which laboratory is receiving the credit backing for nearly 2 gigawatts.