DEEP DIVE — A Year Ago Tomorrow, Oracle Had the Best Day in Its History. Thursday It Reports With Twice the Backlog and Half the Stock Price.

By Lily Caruso · Tuesday, September 8, 2026 · filed 9:30 PM ET; fact-checked and revised 9:45 PM ET — TrendyVest. Not investment advice. TrendyVest and its writers may own securities discussed here.

Fact-check notes are printed at the bottom, at full size.** --- On September 10, 2025, Oracle reported a quarter in which revenue grew 12 percent, and the stock rose more than a third in one session, to a record close of $326.90 — not because of the revenue, but because of a number most people had never looked at before. Remaining performance obligations: the contracts signed and not yet delivered. They had gone from about a hundred billion dollars to $455 billion in a year.

The crowd understood immediately what it was looking at. A company that had sold the next five years of its cloud before it had built it. Tomorrow is the anniversary.

Thursday, at 4:05, Oracle reports again. The backlog is $638 billion now. The stock closed today at $162.52, half the record, after rising 15 percent in four sessions.

And the question the market has been asking all year — first in a rising panic, then in a rising hope — isn't whether the contracts are real. It's who pays for the buildings, and whether the customer on the other side of half of those contracts can pay for what it ordered. I spent the day inside the filings, the June call, the rating reports and the arithmetic.

Here's what I found, in the order I'd want it if I owned the stock, and I'll tell you where I'm inferring and where I'm reading. **THE NUMBERS THE COMPANY ALREADY GAVE YOU.** Oracle guides. That's rarer than it should be for a company whose earnings days move the stock by 8 to 36 percent, and it means the first thing to do with Thursday is to read the guide as the floor the company built for itself. For the quarter that ended August 31, management said on June 10: total revenue up 27 to 29 percent; cloud revenue — infrastructure plus applications — up 58 to 64 percent in dollars; non-GAAP earnings of $1.72 to $1.76 a share.

Consensus, per this week's roundups, is about $19.14 billion of revenue, up 28 percent, and $1.74 of earnings, up 18 percent from last year's $1.47. Now the arithmetic the guide implies. This part is mine, and it's labeled.

A year ago the quarter had $7.2 billion of cloud revenue: $3.3 billion of infrastructure, $3.9 billion of applications. Applications grew 10 percent in the fourth quarter. If they grow 10 percent again, the 58-to-64 percent cloud guide leaves infrastructure — OCI — somewhere between $7.1 and $7.5 billion.

That's 115 to 128 percent growth from a year ago, and 22 to 30 percent growth from the $5.8 billion it printed one quarter ago. Read that again, because it's the whole quarter in a sentence: the company's own guide, taken at face value, requires OCI to add somewhere between $1.3 and $1.7 billion of revenue in ninety days. That's the number Thursday is about.

Not $19 billion. Seven. | The quarter, in one table | Q1 FY26 (actual) | Q4 FY26 (actual) | Q1 FY27 (guide / consensus) | |---|---|---|---| | Total revenue | $14.9B, +12% | $19.2B, +21% | +27–29% guide; ~$19.14B consensus | | Cloud (IaaS + SaaS) | $7.2B, +28% | $9.9B, +47% | +58–64% guide (USD) | | OCI (infrastructure) | $3.3B, +55% | $5.8B, +93% | ~$7.1–7.5B, desk arithmetic from the guide | | Non-GAAP EPS | $1.47 | $2.11 | $1.72–1.76 guide; ~$1.74 consensus | | RPO | $455B, +359% | $638B, +363% | not guided; up $85B q/q last quarter | The other numbers management has already told you. Fiscal 2027 revenue of $90 billion — 33.5 percent growth on the $67.4 billion just reported, 34 percent in constant currency.

Fiscal 2027 non-GAAP earnings of $8.05, up 18 percent. Gross capital spending of $90 to $95 billion this fiscal year, offset by $20 to $25 billion of customer prepayments to a net cash outlay of about $70 billion. Clay Magouyrk, the co-chief executive who runs the infrastructure side, said on the call that Oracle signed "$67 billion in AI infrastructure contracts during the quarter." Hilary Maxson, the chief financial officer: *"We expect to raise around $40 billion in debt and equity in our fiscal year '27,"* including the $20 billion at-the-market equity program announced in February, and no additional debt in calendar 2026.

And she gave the schedule that turns a backlog into a business: 12 percent of the $638 billion recognized within twelve months, another 34 percent over the following two years, with both "expected to accelerate." Twelve percent of $638 billion is $77 billion. Fiscal 2027 revenue is guided to $90 billion. On the company's own schedule, the backlog is most of next year's sales. **WHAT THURSDAY HAS TO SHOW.** Five things, in the order the tape will read them. *One — OCI at seven billion.* If infrastructure prints above $7.1 billion, the guide was honest.

If it prints $7.5 billion or above, the crowd will assume the second-half acceleration is real and start modeling fiscal 2028 in earnest. If it prints under $7 billion with the cloud guide still met — which can happen if applications surprise — the market will read it as capacity arriving late, and capacity arriving late is the whole risk in one line. *Two — the gigawatt.* This one the company set up itself. Management said in June that Oracle delivered more than 1.2 gigawatts of capacity in all of fiscal 2026, and that first-quarter delivery would "approach" a gigawatt.

Oppenheimer's Brian Schwartz, at a $275 target, wrote this week that a gigawatt in the quarter "could support higher FY27 guidance." Mizuho's Siti Panigrahi, at $320, put it more sharply: a year's worth of capacity in one quarter is acceleration. Watch for the number on the call. Capacity is revenue with a lag; a gigawatt now is next year's seven billion. *Three — RPO, and how much of it is one customer.* The backlog rose $85 billion between the third and fourth quarters.

S&P's July note said OpenAI is approximately half of it. Any new name of size dilutes that — Oracle's own February financing release listed AMD, Meta, Nvidia, OpenAI, TikTok and xAI as "our largest Oracle Cloud Infrastructure customers" — and the market will pay more for a $650 billion backlog with three anchors than for a $700 billion backlog with one. *Four — the cash, and how much stock got sold.* Guggenheim's John DiFucci, whose $400 target is the highest on the Street, spent his preview telling investors to check how much of the $20 billion equity program was executed in the quarter, because the company "may need up to $40 billion this year." Free cash flow was negative $23.7 billion in fiscal 2026 on $32 billion of operating cash flow, and S&P expects the deficit to widen to about $42 billion a year. The quarter's capex, the quarter's operating cash flow and the count of new shares are the three lines that decide whether the financing plan is a plan or a hope. *Five — the margin.* Maxson said in June that fiscal 2027 gross margin "will step down because of data center ramp timing and mix," and that infrastructure margins improve as sites reach "full contractual revenue contribution." The October analyst meeting put AI-infrastructure gross margins at 30 to 40 percent against 65 to 80 for enterprise work.

As OCI becomes the business, the blended margin falls by construction. The question is whether it falls faster than the revenue rises. In the fourth quarter it didn't: non-GAAP operating income grew 22 percent on 21 percent revenue growth.

Thursday tells you whether that holds at seven billion of OCI. And then the tape's own verdict. The options market is pricing roughly an 11 percent move, which on today's close is a range of about $145 to $180.

Two prints bracket the year: a year ago, plus 36 percent in a day; in June, a record quarter that was fined 8.5 percent by the next close, because the capex number was $55.7 billion and the financing was $40 billion. The statute this desk writes by says a beat gets paid only when the crowd hasn't already spent it. The crowd spent 15 percent in the last four sessions on GPT-6 Astra, three price targets and an anniversary.

My colleague would say that's room being closed before the print. He'd be right. **THE CUSTOMER.** Here is the part that keeps me up. The contract that made the backlog — roughly $300 billion over five years, about 4.5 gigawatts, "slated to begin in 2027" — is with OpenAI, and spread evenly it runs at something like $60 billion a year once it starts.

OpenAI's annualized revenue was about $25 billion in February. By August, according to reporting summarized by Value Add VC and attributed to Bloomberg, The Information and investor updates, it was about $40 billion, after an enterprise surge in July. That's a real change since I started this piece, and I'm printing it because it cuts against my worry: the gap between what OpenAI makes and what it has promised Oracle is now one and a half times, not two and a half.

But the rest of the structure hasn't changed. OpenAI closed a $122 billion funding round on March 31 at an $852 billion valuation — $50 billion from Amazon, $30 billion each from Nvidia and SoftBank — and its internal forecast, as reported, is a loss of about $14 billion this year, on compute spending Greg Brockman put at roughly $50 billion in testimony. A customer that loses money buying your product, and has to keep raising money to keep buying it, is a customer whose ability to pay is somebody else's balance sheet.

That isn't an accusation. It's a description. And S&P, when it cut Oracle to BBB-minus on July 9 — one notch above junk — named it: OpenAI at about half of the backlog; capex at $90 to $95 billion against a $60 billion earlier forecast; leverage headed to the mid-4x range this fiscal year; the cash deficit headed to $42 billion.

By mid-July Oracle's ten-year bonds yielded about 6.5 percent, closer to where junk trades than to the average triple-B. The five-year credit default swap was near 219 basis points in late July, the widest in years. Long-term debt is $122.3 billion, up from $85.3 billion a year earlier; lease obligations $26.6 billion, up from $11.5 billion.

Moody's carries a negative outlook. Against that, the facts that argue the other way, because they're real too. Abilene — the flagship Stargate campus — went live in September 2025 with the first 200 megawatts across two buildings, per Data Center Dynamics, and IntuitionLabs' April update has it operational at 1.2 gigawatts on Oracle's cloud, with sites under construction in Michigan, Wisconsin, Wyoming and Pennsylvania.

OpenAI is diversifying its suppliers — Broadcom silicon, AMD, AWS — which is bad for Oracle's share and good for OpenAI's survival, and Oracle needs the second more than the first. And the company says it "now serves all of the top five AI models," which means the concentration is a choice about who signs the biggest contracts, not a shortage of alternatives. **THE MONEY, PLAINLY.** Oracle is spending like a hyperscaler without a hyperscaler's balance sheet. Every credit analyst has used some version of that line since June, and it's accurate.

Fiscal 2026 capex was $55.7 billion, from $21.2 billion the year before. Fiscal 2027's gross number is $90 to $95 billion. The company raised $43 billion of debt and $5 billion of equity in fiscal 2026 and plans about $40 billion more this year, half of it stock — the $20 billion program is about 4 percent of a $468 billion market capitalization at today's price, and it's being sold into a stock that's down 17 percent on the year and 50 from its high.

The February plan said all of this "reflects Oracle's commitment to maintaining an investment-grade rating." Five months later the rating was one notch from losing that word. What I'd hold onto: the business underneath the financing works. OCI grew 93 percent last quarter.

Operating cash flow grew 54 percent, to $32 billion. The company hit the $18 billion OCI target it set a year ago almost to the dollar — $18.1 billion. The plan is not failing.

The plan is expensive, and it's being funded by lenders and shareholders on the expectation that a money-losing customer pays $60 billion a year starting sometime in 2027. Both of those sentences are true at once, and the stock price is the market's attempt to hold them both in one hand. **TWELVE MONTHS: WHAT HAS TO HAPPEN BY NEXT SEPTEMBER.** By this time next year Oracle will have reported all of fiscal 2027. The company says $90 billion of revenue; the October analyst meeting implies OCI of about $34 billion, up 88 percent.

The backlog schedule says $77 billion of the $638 billion converts to revenue. The financing plan says $40 billion of new capital; S&P says the cash deficit is $42 billion. So the twelve-month test is three questions with dates on them.

Does OCI hold above the guide each quarter — September, December, March, June? Does the $40 billion turn out to be enough, or does the company come back for more? And does the rating hold — because the difference between BBB-minus and BB-plus is the difference between a bond most investment-grade mandates can hold and one many of them can't.

The path I'd expect: the revenue is delivered, because it's contracted; the capital raise ends up larger than $40 billion, because capex has been raised at every disclosure since June 2025; and the rating survives, because the agencies have said they're watching cash flow and OCI revenue, and OCI revenue is the thing that's arriving. That's a company that makes its numbers and dilutes its holders to do it. Whether that's a good stock depends entirely on the price you paid. **TWENTY-FOUR MONTHS: THE YEAR THE CONTRACT STARTS.** By September 2028, fiscal 2028 will be in the books, and fiscal 2028 is the year the OpenAI contract turns on.

A year ago Safra Catz laid out the path — OCI revenue of $18 billion, then $32 billion, $73 billion, $114 billion, $144 billion — and said most of it was "already secured within reported RPO." The October meeting raised those numbers, to $34 billion, $77 billion, $129 billion and $166 billion by CreditSights' account, and put the fiscal 2028 growth at an implied 126 percent. From $34 billion to $77 billion is a step of $43 billion in a single year, and nearly all of that step is one customer switching on. So the twenty-four-month question isn't about Oracle's execution, which has been good.

It's whether a company with $40 billion of revenue this summer becomes a company that can pay $60 billion a year in 2027 and 2028 — by growing, by raising, or by renegotiating. If it does, Oracle's fiscal 2028 is the best year in the history of enterprise software and the stock is cheap here at 20 times forward earnings. If the payments slip, the backlog is still real but the schedule isn't, the capex was spent on time for revenue that arrives late, and the balance sheet is one notch above junk with $95 billion of new buildings on it.

I don't know which. Nobody does, including the two companies. What I can tell you is that the second outcome is the one the price already reflects — a stock at half its high on a backlog that doubled — and the first is the one it doesn't. **THE ROWS.** *Thursday.* **60% that Q1 FY27 OCI revenue is at or above $7.0 billion.** The desk's arithmetic from the company's cloud guide with applications at plus 10 percent.

Falsifier: applications grow faster than 10 and carry the cloud guide with OCI under $7 billion. Settlement: Oracle's September 10 release, the cloud-infrastructure line. **55% that RPO reported Thursday is at or above $700 billion.** Last quarter added $85 billion; a $62 billion add gets there. Falsifier: no new anchor contract in the quarter.

Settlement: Oracle's release. **60% that Oracle closes Friday, September 11 below Thursday's close.** The statute. Fifteen percent in four sessions is the crowd spending the beat before it prints, and the last record quarter was fined 8.5 percent. Falsifier: OCI above $7.5 billion, a gigawatt confirmed on the call, and a raised fiscal-year guide — all three.

Settlement: the NYSE official close, two sources; a stamp doesn't count. *Twelve months.* **60% that fiscal 2027 revenue, reported in June 2027, is at or above $90 billion.** It's contracted. Falsifier: capacity arrives late. Settlement: Oracle's Q4 FY27 release. **65% that Oracle announces capital raising beyond the $40 billion fiscal-2027 plan — new debt, equity, or an enlarged program — by September 30, 2027.** Capex has gone up at every disclosure since June 2025.

Falsifier: fiscal 2028 capex is guided flat or down. Settlement: Oracle's own releases and filings. **60% that S&P still rates Oracle investment grade — BBB-minus or better — on September 30, 2027.** The agencies said they're watching OCI revenue and cash flow; the revenue is arriving. Falsifier: fiscal 2028 capex guided above $100 billion with the cash deficit widening past S&P's $42 billion.

Settlement: S&P's published rating, confirmed by a second outlet. *Twenty-four months.* **45% that fiscal 2028 OCI revenue, reported in June 2028, is at or above $73 billion — the number Safra Catz gave a year ago.** It requires the OpenAI ramp to start on schedule and at scale. Falsifier: a renegotiated or delayed start. Settlement: Oracle's Q4 FY28 release. **60% that Oracle has not been downgraded to junk by either S&P or Moody's on September 30, 2028.** Falsifier: the customer can't pay on schedule and the buildings are already built.

Settlement: the agencies' published ratings, two sources. *Grading notes.* Thursday's rows grade on the release and Friday's close. The twelve- and twenty-four-month rows are logged to the tracker tonight with their settlement sources; they grade on company documents and agency ratings, not on reports of them. **A NOTE ON THE STOCK ITSELF.** $194.91 on December 31. $201.26 on June 10, before the fourth-quarter print, and $184.10 the next day. $114.99 on July 24, the low close of the year, two weeks after the downgrade. $141.32 a week ago and $162.52 tonight — down 17 percent on the year, down 50 from the record, up 41 from the low, up 15 in four days. Two feeds disagree by 29 cents on today's close, which is a small thing that this desk prints anyway.

Thirty-one analysts, 27 of them at Buy, average target $257, from $145 at the bottom to $400 at the top; Morgan Stanley, at Neutral, raised its target to $210 this week. The forward multiple is about 20 times. That is not an expensive stock for a company guiding to 33 percent growth.

It's an expensive stock for a company whose growth is financed by a customer that hasn't paid yet, and the market can't decide which of those it's holding. Thursday won't settle that. It will tell you whether OCI printed seven billion, and that's enough to trade on. — Lily Caruso **FACT-CHECK NOTES, PRINTED INSIDE THE COLUMN.** The first draft was filed at 9:30 and checked against second sources afterward.

What changed, at full size: *OpenAI's revenue.* The first draft used "$25 billion in February" and built a two-and-a-half-times gap on it. Reporting summarized by Value Add VC, attributed to Bloomberg, The Information and investor updates, has annualized revenue at about $40 billion by August after an enterprise surge in July. The column now carries both figures and the smaller gap, and says so in the text, because a fact-check that only tightens the case for the writer's worry isn't a fact-check. *The gigawatt.* The first draft credited the "extra gigawatt" thesis to Oppenheimer and Mizuho.

The June call, per ERP Today's account, had management itself saying Oracle delivered more than 1.2 gigawatts in fiscal 2026 and expected first-quarter delivery to approach a gigawatt. The analysts were reading the company's guide; the column now says so. *Abilene.* The first draft said the campus is operational at 1.2 gigawatts on one source. It now carries two: Data Center Dynamics' September 2025 report of the first 200 megawatts across two buildings, and IntuitionLabs' April 2026 update at 1.2 gigawatts. *The June reaction.* TheStreet's account had the after-hours drop at 7.4 percent; ERP Today's had it near 10.

The column prints the number that settled — the next day's close, down 8.5 percent on Yahoo's history — and nothing else. *The bond yield.* Added: The Next Web's July 16 report that Oracle's ten-year bonds yielded about 6.5 percent after the downgrade. *The RPO schedule.* The 12-percent-in-twelve-months and 34-percent-over-two-years figures were one source at 9:30; Yahoo Finance's account of the call carries the same numbers. *The $67 billion.* Added from Yahoo's call coverage: Magouyrk's statement that Oracle signed $67 billion in AI infrastructure contracts in the fourth quarter. *The margin quote.* The first draft used a paraphrase from TheStreet. ERP Today carries Maxson's wording — "will step down because of data center ramp timing and mix" — which is what's printed now. *Still single-sourced, and labeled.* The CDS level, the debt and lease figures, and the Moody's outlook are from TheStreet's August 14 report; the 30–40 percent and 65–80 percent gross-margin ranges are from CreditSights' account of the October meeting; the OpenAI share of RPO is S&P's estimate as reported by MLQ and The Next Web. --- *This is TrendyVest's analysis and opinion — for informational purposes only, not investment advice or a recommendation to buy or sell any security. TrendyVest and its writers may own securities discussed here.

The implied OCI range is the desk's arithmetic from the company's guide and an assumption about applications growth, labeled as such; the $60 billion a year is the contract's headline value divided evenly over five years, as IntuitionLabs presents it; OpenAI's share of RPO is S&P's estimate as reported; the Meta arrangement was reported as under discussion in April and Meta appears as a named customer in Oracle's February release. Sources: Oracle's Q4 FY26 release, June 10, 2026 (Q4 revenue $19.2B, +21%; cloud $9.9B, +47%; OCI $5.8B, +93%; applications $4.1B, +10%; non-GAAP EPS $2.11; FY26 revenue $67.4B; OCI $18.1B, +77%; operating cash flow $32.0B, +54%; free cash flow −$23.7B; debt raised $43B, equity $5B; RPO $638B, +363%, +$85B q/q, $75B prepaid/customer-supplied; Q1 FY27 guide of +27–29% revenue, +58–64% cloud in USD, EPS $1.72–1.76; FY27 revenue $90B and EPS $8.05); Oracle's Q1 FY26 release, September 9, 2025 (revenue $14.9B, +12%; cloud $7.2B; OCI $3.3B, +55%; RPO $455B, +359%; EPS $1.47; the Catz OCI path of $18B, $32B, $73B, $114B, $144B); the June 10 call as reported by TheStreet (net capex ~$70B; ~$40B financing; no additional 2026 debt; −7.4% after hours), ERP Today (1.2 GW delivered in FY26; Q1 delivery approaching 1 GW; the Maxson margin quotes; the Magouyrk "look small" quote; ~10% after hours), Yahoo Finance (12%/34% RPO conversion; FY27 +34% constant currency; the $67 billion quarter; the Sicilia quote) and MLQ (gross capex $90–95B; $20–25B customer offset; the Maxson financing quote); Oracle's February 1, 2026 financing release ($45–50B gross for calendar 2026; ~50/50 equity and debt; $20B ATM; mandatory convertible preferred; the customer list; the investment-grade language); TheStreet, August 14, 2026 (long-term debt $122.3B from $85.3B; total liabilities $218.7B; leases $26.6B from $11.5B; S&P BBB− in July; Moody's negative outlook; 5-year CDS ~219bp); MLQ's and The Next Web's accounts of the July 9 S&P downgrade (OpenAI ~half of RPO; capex $90–95B vs $60B; FOCF deficit ~−$42B; leverage mid-4x; ten-year bonds ~6.5%); CreditSights' takeaways from the October 2025 analyst meeting (FY30 revenue $225B; OCI raised by $2B, $4B, $15B, $22B; OCI ~74% of revenue; 75% five-year CAGR; implied FY28 growth 126%; AI-infrastructure gross margins 30–40%, enterprise 65–80%); IntuitionLabs' April 2026 update (OpenAI contract ~$300B over five years, ~4.5 GW, beginning 2027; Abilene 1.2 GW operational; sites in Michigan, Wisconsin, Wyoming, Pennsylvania; "all of the top five AI models"); Data Center Dynamics, September 30, 2025 (Abilene's first 200 MW live across two buildings); Value Add VC's summary of OpenAI reporting (~$25B ARR February–May; ~$40B by August; ~$14B 2026 loss as an internal forecast; the $122B March 31 round at $852B with Amazon $50B, Nvidia $30B, SoftBank $30B) and Presenc's commitments tracker (Brockman's ~$50B 2026 compute spend); the co