When I write about the AI buildout, I tend to reference Oracle’s September 2025 earnings report more often than Nvidia’s Q1 2023 update.

That’s certainly not because Oracle’s report a year ago was actually more consequential for the AI narrative than Jensen Huang’s famous “guide heard ’round the world.”

Rather, it’s because, as SocGen’s equity derivatives team put it in their latest update, Oracle’s September 2025 report “stands out as the moment the market realized AI [will require] a global infrastructure buildout lead[ing] to severe shortages.”

While editorializing around that release 11 months ago, I called Oracle’s press release “a dream come true for investors champing at the bit to bid up companies with a claim on AI-cloud hybrid revenue.” The shares rose almost 40% the following day, briefly making Larry Ellison the richest person on the planet.

The fact that the stock plummeted over the ensuing six months only underscores why that release was so important.

Just days after the earnings announcement, Oracle sold $18 billion in bonds, kicking off a borrowing binge that saw the hyper-scalers rack up $250 billion in debt between them. Ungenerous types have since suggested Oracle started an arms race with four companies against which it had no chance of competing — at least not without imperiling its balance sheet.

Thereafter, Oracle became the poster child for debt-funded AI investments and, critically, free cash flow burn. The figure below’s a reminder that the “before and after” snapshot for capex at Oracle is night and day. Apples to oranges, even. You have to use a log scale.

To this day, Oracle’s treated, fairly or not, as an outlier among the hyper-scalers: A company which can’t really afford to play the game it started a year ago next month.

That game began to manifest in 2026 in outsized — sometimes wild — top- and bottom-line results from the beneficiaries of the AI capex splurge. Company analysts quickly marked up expectations for those companies, mostly semi names, which in turn contributed to the largest forward profit boom in history outside of recession rebounds.

One consequence of all that was an increase in earnings uncertainty, which helps explain very high single-stock vol. “Aggressive capex has led to unprecedented expectations of earnings growth, but more importantly for volatility, uncertainty around earnings has also ramped up materially,” SocGen’s Jitesh Kumar and Vincent Cassot remarked.

The figure on the left, above, gives you a sense of how Oracle changed the game on the profit expectations front, while the figure on the right shows you the shift in the relationship between big tech earnings uncertainty and single-stock vol.

“The market now seems to be concerned about higher uncertainty and single-stock volatility has moved up [accordingly],” SocGen went on, adding that the uncertainty “has its origins in both left-and right-tailed risks, but over the past few years, the right-tail risks have dominated.”

That latter bit’s key: It often feels as though the frequency of melt-ups is increasing, and as discussed here and here, the most recent example of call skew steepening (i.e., the burst of demand for OTM upside “protection” that accompanied the push to new records on the US benchmarks) was anomalous.

In the same note, Kumar and Cassot casually remarked that the most you can lose on unlevered stock longs is 100%, which explains why, during booms and bubbles, “the expansion of the returns envelope often happens on the upside.”