One year ago this week, Oracle hit $345.72 a share and Larry Ellison spent about twelve minutes as the richest man on Earth on the strength of a slide deck about remaining performance obligations. Thursday, Oracle reports Q1 FY27 earnings. The stock closed Tuesday around $159. Do the math yourself, I'll wait — it's down more than 50% from the number that made every AI-capex bull's group chat light up like a slot machine, and it's still, somehow, being framed by half of Wall Street as a buying opportunity. Bernstein's got a $325 target on a stock trading at $159. Morgan Stanley just bumped theirs to $210. UBS trimmed to $245 and called that a haircut. These are not forecasts. These are hostages reading a script written under different macro conditions, refusing to update because updating means admitting the thesis needs surgery, not a Band-Aid.
Here's the thesis, for anyone who's forgotten, because forgetting is apparently the house style: Oracle would spend $50 billion — later revised to $55.7 billion, because guidance is now more of a starting bid than a number — building data centers for OpenAI and friends, fund a chunk of it with debt, and eventually the $638 billion remaining-performance-obligations backlog would convert into actual cash flowing back the other direction. RPO jumped $85 billion in a single quarter. Almost all of the growth, Oracle told everyone, came from AI contracts where customers were prepaying for GPUs or buying the chips themselves and handing them to Oracle to operate, supposedly reducing the company's own financing burden. It's a clever structure. It is also, functionally, a company telling you it doesn't need to raise as much money because its customers are pre-funding the thing they'll eventually rent back from it — which is either brilliant capital efficiency or an admission that nobody involved wants to be the one holding the balance sheet when the music stops, so everyone's quietly passing the instrument around the circle.
And now Thursday arrives with the options market pricing an 11.2% move on a stock that's already down a quarter this year, against a backdrop that makes the setup almost cruel. Friday's CPI print lands the day after earnings — the same CPI that Warsh has all but promised will determine whether the Fed hikes on September 16th. So Oracle gets to report into a market that's simultaneously deciding whether borrowing costs are about to get more expensive at the exact moment Oracle needs those costs to stay cheap, because a company running negative free cash flow while promising $80-100 billion in fiscal 2027 capex does not want the 10-year sitting at 4.8%, does not want oil above $90 pushing inflation prints hot, and does not want a hawkish Fed chair fresh off a Jackson Hole speech about "underlying inflation" using this exact CPI number as his casus belli.
Nobody's saying it plainly so let's say it plainly: Oracle is the AI capex trade's most exposed nerve ending, and it's reporting directly into the week the Fed decides how much that nerve is going to hurt. This isn't Nvidia, sitting on 70-plus percent gross margins and actual pricing power, able to absorb a bad quarter with a shrug. This is a company whose entire bull case depends on converting paper backlog into cash before the debt comes due, in a rate environment that is actively getting worse, not better, while it's trying to do it. The stock's 52-week range — from $345.72 to $114.50 — isn't volatility, it's a seismograph reading of exactly how fast sentiment can invert on a company that hasn't materially changed its underlying business in either direction. Same backlog story. Same capex story. Same OpenAI deal. The number moved $230 a share because the narrative moved, and the narrative is about to get tested against inflation data instead of vibes.
And that's really the whole AI-capex complex in miniature right now, isn't it. Nobody's disputing the demand — the RPO is real, the contracts are real, the data centers are being poured in concrete as we speak. What's actually in question, the thing that keeps getting dressed up as a "financing" story when it's really a rates story wearing a trench coat, is whether the capital structure underneath all of it survives a Fed that's about to spend a week deciding whether 4.8% ten-year paper becomes 5%-plus ten-year paper. Every hyperscaler running this playbook — Oracle most nakedly, because Oracle's the one who needed the debt market most — is implicitly betting that money stays cheap enough long enough for the backlog to become revenue before the bill comes due. That bet gets repriced Thursday and again Friday and again on the 16th, three separate times in nine days, and Oracle is the company standing closest to the blast radius with a $55.7 billion capex number in one hand and a negative free cash flow figure in the other, hoping the analysts with the $325 targets turn out to be right about the timing.
They might be. Bull cases built on genuine demand curves do eventually work out, and Oracle's demand curve is not fictional — $638 billion of backlog doesn't materialize from nothing. But "eventually" is doing an enormous amount of work in a capital structure this leveraged, at a moment when the cost of "eventually" is being repriced in real time by a Fed chair who's spent the last two weeks all but announcing he's done being patient. Thursday's number will tell you whether Oracle's story is still intact. Friday's number will tell you whether the Fed thinks it can afford to let it stay that way.
