Nvidia's Capital Web: From Chip Salesman to the AI Boom's Financial Architect

Nvidia expands AI infrastructure financing, trims OpenAI guarantee, and holds stakes in Intel and SpaceX to secure demand.

There's a telling symmetry in Nvidia's latest moves. The company that built its fortune selling the silicon brains of artificial intelligence is now quietly becoming the financial plumbing that keeps the entire AI ecosystem humming — underwriting power plants, guaranteeing data-center loans, and holding equity stakes in the very companies that buy its chips.

The most striking evidence arrived this week in two separate disclosures. First, Nvidia revealed through its 13F filing that it holds a portfolio worth $63.4 billion as of June 30, up from $18.4 billion previously. The bulk of that — roughly $30 billion in Intel and $21 billion in SpaceX — represents a strategic bet on its own demand chain rather than a passive investment portfolio. The Intel position alone has ballooned from a $5 billion investment twelve months ago, while the SpaceX stake originated from an investment in xAI that was converted into shares of the space company following their merger, though it has since slipped to around $17 billion in value after the stock fell from its early-June IPO levels.

A Guarantee Trimmed, Not a Retreat

At the same time, Nvidia has sharply reduced its financial guarantee for the joint data-center project with OpenAI in Ohio, cutting its commitment from $250 billion to under $120 billion. The Wall Street Journal reported Thursday that Nvidia scaled back its investment obligation for the first phase of the project — roughly 5 gigawatts of a campus ultimately planned for 10 gigawatts and more than $500 billion in total cost. Reuters indicated the revised deal could be signed as early as this weekend.

The temptation is to read this as a loss of confidence in OpenAI. That would be a misreading. Nvidia remains embedded in the SB Energy-developed infrastructure and committed back in September 2025 to supply at least 10 gigawatts of its own systems, with potential participation of up to $100 billion. The guarantee reduction looks less like an exit and more like a renegotiation of who bears what risk.

That interpretation gains weight when set alongside the financing platform Nvidia announced on August 10 with Apollo Global Management, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR — an alliance designed to mobilize more than $500 billion for AI infrastructure. CEO Jensen Huang has said Nvidia would backstop up to 25 percent of that sum, roughly $125 billion. Goldman Sachs is reportedly in talks with additional investors to join the initiative.

Huang's own framing captures the logic: "In AI, compute is revenue." Nvidia isn't shedding risk so much as redistributing it — moving away from a single massive Ohio guarantee toward a broader structure carried by third-party capital. The company has also reportedly been in discussions to invest up to $3 billion in SB Energy, SoftBank's renewable-energy subsidiary, as part of the Ohio project's credit-support negotiations. That follows news that Nvidia is considering a similar $3 billion investment in Lancium, the developer behind the Stargate data center in Texas, which could go public in 2027.

The Software Track Runs in Parallel

While the capital flows into concrete and turbines, Nvidia continues advancing its software ambitions. Reuters reported this week that the company is developing Nemotron 4, a new family of open AI models, with the largest version expected to exceed one trillion parameters and a possible launch targeted for late autumn. The company has also introduced Nemotron 3.5 Lightning, designed for tasks including code review, tool use, and monitoring security alerts.

The partnership channel is active as well. IBM and Together AI announced a multi-year, $240 million agreement to build a large AI cluster on IBM Cloud using Nvidia HGX B300 systems and Spectrum-X networking technology, initially deploying around 2,000 Nvidia Blackwell 300 chips. It's another illustration of how deeply Nvidia's hardware is now embedded in third-party infrastructure without the company having to operate data centers itself.

What the Market Makes of All This

The stock's reaction to this flurry of activity has been measured rather than euphoric. Shares closed Friday at €194.74, down 0.4 percent on the day, though the 30-day picture shows a 5.1 percent gain and the year-to-date return stands at 22 percent. The stock sits 3.8 percent below its 52-week high of €202.50 reached in May — and a comfortable 39 percent above its yearly low.

Skeptics have their case. Investor Steve Eisman warns that the entire AI boom rests on just two customers — OpenAI and Anthropic — which together account for roughly 70 percent of AI-related revenue at Microsoft, Amazon, Alphabet, and Oracle. Michael Burry has positioned himself publicly against Nvidia and semiconductor stocks. Goldman Sachs has pegged unstarted hyperscaler leasing commitments at around $1 trillion. Insider selling at Nvidia totaling $767.2 million over 90 days is hard to ignore entirely.

The market will render its next verdict on August 26, when Nvidia reports quarterly earnings. Analysts expect revenue around $91.9 billion, while management has guided to a midpoint of $91.0 billion.

What's becoming clear is that Nvidia has crossed a threshold. It no longer merely sells chips; it finances the power plants, data centers, and credit structures that make those chips useful. Whether that makes the company a more resilient player in the AI economy or simply a more exposed one is the question the coming quarters will answer. For now, the market seems to be treating the reordering of Nvidia's financial architecture with equanimity — not alarm.