The $500 Billion AI Bet That Could Turn to Junk

The AI Infrastructure Euphoria Is Meeting Its Credit Rating

There is a version of the AI infrastructure story where Oracle is a hero: a legacy enterprise giant that bet big on the future, landed a starring role in the $500 billion Stargate project alongside OpenAI, and rewarded its shareholders for the faith. Then there is the version that is actually playing out right now.

JP Morgan is struggling to syndicate $38 billion in loans connected to the Stargate project without pushing yields high enough to spook the market. Oracle's debt sits at a BBB rating, which is investment grade, but barely. S&P has formally flagged the company's AI spending as a financial risk. And the word nobody wants to say out loud is now being said quietly: junk.

As Neeta Bidwai laid out in this week's episode, Oracle went from riding high in Q3 of 2025 to a position that should concern anyone holding the stock or the debt.

Why This Is Bigger Than One Company's Balance Sheet

Oracle is not some scrappy startup that overpromised and underdelivered. It is one of the most established names in enterprise technology. The fact that its aggressive push into AI infrastructure is now a credit concern is a signal worth taking seriously about the economics of building AI at scale.

The Stargate project represents the kind of capital commitment that looked visionary when AI euphoria was at its peak. What it looks like when financing costs rise and lenders demand higher yields is a different story entirely. The investors JP Morgan is trying to bring in are looking at that BBB rating and calculating how quickly it could slip, and whether they want to be holding paper when it does.

There is a broader pattern here. PitchBook data cited in the episode shows that AI represented 65% of annual deal value in 2025, totaling $222 billion, but that number is concentrated in a small number of mega deals. Corporates participated in 68% of AI deal value, an all-time high, and the episode notes this appears to be driven by both genuine opportunity and fear of missing out. Fear of missing out is not a balance sheet strategy.

The Ellison Wildcard

The Oracle story has a subplot that is genuinely strange. Larry Ellison personally guaranteed $40 billion connected to Paramount's bid to acquire Warner Bros. Discovery from Netflix. That proxy campaign has already stumbled, and the tender offer deadline has been pushed to February 20th. But the more interesting problem is the link it creates between Oracle's financial health and a major media M&A battle. If Oracle's creditworthiness continues to deteriorate, the knock-on effects reach well beyond cloud infrastructure.

This is what overleveraged optimism looks like when the cycle turns. One big bet creates exposure that bleeds into entirely unrelated industries.

Apple Just Made OpenAI's Problem Worse

Separate from the Oracle situation, the episode surfaced another data point that reshapes the competitive picture for AI services. Apple has chosen Google's Gemini to power the AI replacement for Siri, explicitly passing over OpenAI. Apple and Google reportedly described Gemini and Google Cloud infrastructure as the superior option currently available.

For a company trying to establish itself as the default AI layer across enterprise and consumer products, losing Apple to a competitor is not a minor setback. Siri is installed on hundreds of millions of devices. If Gemini becomes the AI engine behind it, that is a distribution advantage that compounds over time.

What Comes Next

The AI infrastructure buildout was always going to require a reckoning with basic financial reality. The question was when, and what the trigger would be. Oracle's debt pressure suggests the reckoning is arriving earlier than the most optimistic projections assumed.

For investors, the episode's framing is useful: AI deal value is real, but it is concentrated, and the companies carrying the most debt to fund that concentration are now the ones drawing scrutiny. The infrastructure layer of the AI economy is not immune to the same dynamics that have plagued every previous technology buildout, from fiber optic cables in the late 1990s to data centers in the 2010s.

The fish-gill microplastics filter that closed the episode is a useful contrast. Small, cheap, installable at the source, and solving a genuine problem. Sometimes the most durable infrastructure is the kind that does not require a junk bond to build.