$3 billion. Invested in data centers. Not a loan—equity.

On August 16, according to The Information and Wall Street Journal, NVIDIA is in talks to invest up to $3 billion in SoftBank subsidiary SB Energy to support its Ohio data center development for OpenAI. The investment comes in two tranches: half upon project signing, half upon SB Energy IPO.

Note the key players: NVIDIA puts up the money, SoftBank provides the land, OpenAI brings the demand. Three companies joining forces, using financial leverage to scale AI compute infrastructure.

A Three-Way Partnership, Each With Their Own Math

The structure of this deal is fascinating—each party gets something different.

NVIDIA's math: selling chips isn't enough anymore; they're now directly investing in data centers. Previously NVIDIA's model was "you build the data center, I sell you GPUs"—simple, straightforward, cash and carry. But as data center investments grow larger, customers face mounting pressure to spend tens of billions upfront on facilities. NVIDIA's answer: I'll help you fund it, you pay it back over time with my chips as collateral, and we both profit from equity upside.

This isn't the first time. Just on August 11, NVIDIA teamed up with six Wall Street giants—Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR—to launch a $500 billion AI infrastructure financing platform. The $3 billion SB Energy investment is another instance of this financial playbook.

SoftBank's math: SB Energy was originally a clean energy company, now crossing over into AI data centers. They have land, energy capacity, and construction experience—but they lack customers and AI industry connections. Partnering with NVIDIA + OpenAI means securing both chip supply and the biggest customer simultaneously. The data center has a committed tenant and revenue stream before it's even built, pulling IPO valuation straight up.

OpenAI's math: compute demand grows by the day. Building data centers yourself is too slow and capital-intensive; renting isn't flexible enough. Having SB Energy as landlord and NVIDIA as chip supplier is like "moving into a furnished place"—no need to spend tens of billions building facilities yourself, just pay for what you use, scale fast, keep the balance sheet clean.

Compute Infrastructure Enters the Financialization Era

The larger significance of this deal: the investment logic of AI compute infrastructure has shifted from corporate capex to long-term asset investment.

Previously, data centers were built by internet companies and cloud vendors using their own capital expenditure, depreciated over time. Not anymore—data centers are becoming an asset class that generates stable long-term cash flow, like real estate, highways, or power grids—financeable, securitizable, held long-term by insurance companies and pension funds.

NVIDIA leading the $500 billion financing platform is exactly about this: packaging GPUs as assets, data centers as assets, compute services as assets, then selling them to long-term capital on Wall Street. In this paradigm, AI compute supply is no longer constrained by tech companies' balance sheets—it taps into the infinite ammunition of global capital markets.

Of course, this raises new questions. When compute infrastructure becomes financialized, will return-seeking capital force compute vendors to accelerate expansion? Will there be overcapacity? A compute bubble? These questions don't have answers yet. But one thing is certain: the AI compute arms race has escalated from competition among tech companies to a game of global capital.

See you tomorrow.