Platformonomics TGIF #105: October 24, 2025

Did OpenAI Find the Money This Week?

Announcing spending plans is easy. Paying for them is hard. Especially when the funding gap is hundreds of billions of dollars.

Some interesting details from a WSJ story this week on OpenAI’s recent dealmaking frenzy:

The resulting game of financial one-upmanship has tied the fates of the world’s biggest semiconductor and cloud companies—and vast swaths of the U.S. economy—to OpenAI, essentially making it too big to fail. All of them are now betting on the success of a startup that is nowhere near turning a profit and facing a mounting list of business challenges.

As strategies go, being “too big to fail” is light on agency.

Nvidia ended up signing an agreement to lease up to 5 million of its chips to OpenAI, costing $350 billion by today’s standards.

This is the first time I’ve seen the NVIDIA-OpenAI deal described as GPU leases versus purchases. There have been some GPU lease trial balloons, but this is more definitive. Not positive for NVIDIA if they are becoming a buy-now, pay-later company.

As part of the deal, Nvidia is also discussing guaranteeing some of the loans that OpenAI plans to take out to build its own data centers, people familiar with the matter said—a move that could saddle the chip giant with billions of dollars in debt obligations if the startup can’t pay for them. The arrangement hasn’t been previously reported.

The deal grows even more circular, but also an admission OpenAI can’t raise debt on its own?

We’re still looking for an external injection of several hundred billion dollars to make OpenAI’s current slate of ambitions work, even if they are “too big to fail”.

What Oracle didn’t say is how it expects to pay for the very expensive expansion of its network that will be needed to generate such returns. Powering AI workloads first requires pricey chips from companies like Nvidia and AMD, and the components to run them in data centers. The company’s capital expenditures exceeded its operating cash flow for the first time since 1990 in its latest fiscal year that ended in May.

That is likely just the start of it. Wall Street expects Oracle’s negative free cash flow to continue for the next three fiscal years, with cash burn for the period totaling nearly $29 billion by the end of fiscal 2028, according to consensus estimates from Visible Alpha.

Oracle’s ability to live up to its lofty projections will depend not just on the growth of customers like OpenAI and Elon Musk’s xAI, but also its own ability to scale up its networks to serve those businesses. That won’t be easy, or cheap. In a report last month, Morgan Stanley’s debt analysts projected a “sizable uptick in new bonds” from the software giant. And that is even after an $18 billion bond sale last month that will only cover about one-quarter of the company’s cash needs through 2028, the analysts said.

Simply put, 4.5GW * $40 billion/GW = ~$180 billion in CAPEX. Claiming you’re “asset pretty light” doesn’t actually make that number go away.

He’s not the hero San Francisco deserves, but the one it needs right now?

(Dark Benioff started as a Biden meme reference, but we’re switching to Batman).