Oracle’s the poster child for the market’s AI capex concerns, and also for investors’ growing impatience with the temporal gap between enormous outlays and returns on those investments.

To me, what matters are the longer run margins on AI-related and AI-adjacent spending, not so much the timing of the payoff.

The sooner the better, sure, but this is an arms race, not a race against time. Or if the hyper-scalers are racing the clock, it’s in the context of competition to woo customers and create the best product and service offerings.

Simply put: The hyper-scalers aren’t going to go under if all the speculative AI spending doesn’t accrue fast enough to the top and bottom lines.

That said, the timeline on AI-adjacent revenue can’t be completely indeterminate. After all, the money’s being spent in the here and now and increasingly, that money’s borrowed.

Oracle, which sold $18 billion in debt in September, is lining up nearly $40 billion in new data center financing, and analysts expect the company to borrow as much as $30 billion per year through 2028. It already has more than $100 billion in debt outstanding, anomalous for an IG tech company.

Those debt dynamics are behind the much ballyhooed blowout in Oracle’s CDS, which kissed 140bps on Thursday, when the stock plunged following an earnings release which failed to dispense with concerns around runaway spending.

I’ve been over (and over and over) the CDS hysteria in recent weeks. The overarching point of that coverage is that in most cases — i.e., in the case of blue-chip companies — the widening doesn’t reflect actual default concerns, but rather hedging to account for a shift in the debt dynamics of America’s biggest tech firms.

A meaningful share of the demand for single-name CDS likely emanates from banks. Oracle’s borrowing is essentially just construction loans, and lenders want some sort of protection in case things go awry.

As one fixed income trader told Bloomberg last week, “It should be a shock to no one that [banks] may be exploring hedging or risk transfer mechanisms” in light of richly-valued firms borrowing heavily to fund a highly speculative venture.

Still, the fact that one of Oracle’s co-CEOs (this week’s report was the first post-Catz earnings update) had to reiterate the company’s commitment to maintaining its IG rating was notable. (“If you have to say it…”)

There’s the simple chart of the share price reaction. I’ve highlighted the rather stark juxtaposition with last quarter’s post-earnings reaction.

Recall that the September rally (highlighted in green) briefly made Larry Ellison the richest person on Earth. Elon Musk’s since regained the crown and Larry’s gonna take a hit this week, that’s for sure. As of late Thursday morning, the market cap loss for Oracle was $90 billion.

Cash burn’s obviously accelerating (it was ~$10 billion or so over the quarter) amid escalating capex which exceeded $12 billion, far more than analysts expected and up more than 200% YoY, as shown below.

The chart also shows cloud bookings, which rose another 15% or so to $523 billion. That’s a lot of commitments, but there are concerns (to put it politely) around concentration risk.

Oracle’s part of what I’ve described as history’s most complicated Venn diagram, which is to say it’s embedded in the circular web of deals spun by OpenAI and Nvidia, both of which are customers.

Oracle was outright lampooned in the Financial Times’s legacy blog, “Alphaville,” last month for its over-reliance on OpenAI, exemplified by Sam Altman’s already infamous five-year, $300 billion spending commitment. As Alphaville caustically put it, Oracle’s “throwing everything it can at supporting its one big customer in exchange for an IOU.”

Simple math suggests OpenAI accounts for 57 cents out of every dollar of Oracle’s backlog. Oracle’s leveraging itself to ensure it can deliver the computing capacity Altman promised to buy, but there are no guarantees he’ll come up with the money.

OpenAI’s total spending commitments are nearly $1.5 trillion. That’d be a lot for any company, let alone one with just ~$20 billion in current revenue. And then there’s Alphabet’s Gemini coup, which prompted Altman to issue a “code red” in an alarmed memo to employees this month.

What’s bad for OpenAI is necessarily bad for Oracle. That’s what I meant in October, when I mused repeatedly about the perils of these companies binding their fates such that the success of one depends upon the success of the others.

Hopefully, you now have something like a comprehensive understanding of where the problem(s) might come in for Oracle, and why the stock was down sharply on Thursday despite a 34% YoY increase in realized cloud sales and a 68% jump in infrastructure revenue.

Co-CEO Clay Magouryk said company analysts’ projections for Oracle’s total, multi-year spending on the data center buildout are likely overstated. “We expect we will need less if not substantially less than that amount,” he said, of the $100 billion consensus.

And yet, Oracle said capex for this year alone will likely approach $50 billion (its fiscal year ends in May). That was more than 40% higher than the company projected just three months ago.

Remember: Oracle doesn’t enjoy the same sort of financial heft and cushion as the other hyper-scalers, all of which boast balance sheets generally viewed as impregnable.