Jensen Huang Bought Into OpenAI. He Called It His Last Chance.

Jensen Huang just gave OpenAI $30 billion. He also said it might be the last time he gets the chance.

Nvidia's chief executive called the investment what might be his final opportunity to invest in "a consequential company like this" before OpenAI goes public, according to Reuters. That's not how you talk about a bet you're confident in. That's how you talk about a relationship you're trying to lock in before the terms change.

OpenAI has been telling investors a different story about its infrastructure ambitions. It slashed its spending target from $1.4 trillion to $600 billion through 2030 — still enormous, but the gap between those numbers tells you something about the physics of what was being promised versus what can actually be built. OpenAI told investors in February that it is now targeting roughly $600 billion in total compute spend by 2030, down from the $1.4 trillion that CEO Sam Altman discussed publicly in a November 2025 post on X, per multiple reports citing the statement.

The physical constraints are where this story gets real. Building a one-gigawatt data center — enough power to run roughly 750,000 U.S. homes — costs around $50 billion and takes three to ten years from start to finish, according to Walid Saad, an engineering professor at Virginia Tech. OpenAI has claimed its first one-gigawatt facility will be ready before the end of 2026. That timeline is the kind of thing that sounds like a commitment until you ask an engineer.

The Abilene, Texas story tells you who is actually building at scale. A complex in Abilene being constructed for OpenAI and Oracle will supply 2.1 gigawatts when complete, per AP News. Crusoe, the construction firm handling the project, is still completing six more buildings for OpenAI and Oracle due by the end of this year. But Microsoft separately took over a 900-megawatt data center project in Abilene that OpenAI had declined to pursue, according to AP News. Microsoft is building what OpenAI walked away from.

On the Nvidia side, the deal structure deserves scrutiny. MIT Sloan professor Michael Cusumano described the Nvidia-OpenAI arrangement as "kind of a wash" — Nvidia puts $100 billion into OpenAI stock, OpenAI turns around and buys $100 billion or more in Nvidia chips, per TechCrunch. That framing doesn't sound like a venture investment thesis. It sounds like vendor financing.

Jensen Huang called it a one-time opportunity, implying Nvidia is getting in before the window closes. "The opportunity to invest $100 billion in OpenAI is probably not in the cards," Huang told CNBC. "It might be the last time it invests in the artificial intelligence startup before it could go public." If that's true, it means Nvidia sees more value in being locked into OpenAI as a private company than as a public one — before an IPO introduces different dynamics, different shareholders, and different expectations about what the relationship is worth.

The infrastructure numbers only make sense if the revenue numbers land. OpenAI generated $13.1 billion in revenue in 2025 and is projecting more than $280 billion by 2030, per CNBC. The $600 billion commitment is the load-bearing wall of that projection — buy the compute, run the models, capture the users, prove the revenue. If the data centers don't get built on schedule, none of the downstream numbers work. OpenAI does not currently own any data centers, and may not for the foreseeable future, according to people familiar with the matter — it buys cloud capacity from Oracle, Microsoft, and Amazon. That means the infrastructure story is also a relationship management story, one where OpenAI's partners have their own priorities and their own limits.

The $110 billion funding round that closed in early 2026 — $30 billion from SoftBank, $30 billion from Nvidia, $50 billion from Amazon, at a $730 billion pre-money valuation, per OpenAI's own announcement — is the largest single capital formation event in the history of the technology industry. Jensen Huang's decision to describe his $30 billion piece as a final-chance investment rather than a vote of confidence is the most interesting thing anyone involved in that round has said about it.

The data center is the story. Not the press release about the data center — the actual building of it, or the failure to build it, or the partner who walked away from it and the competitor who picked it up. That is where the IPO thesis meets the physical world.