Oracle is down two-thirds. Its order book grew 363%. Both are true.

Oracle $ORCL closed July 27 at $119.97 — within a few percent of its 52-week low, and 65% below its 52-week high of $345.72. On X it has become the exhibit for the end of the AI trade: the first big AI casualty, the proof the buildout was a bubble.

Six weeks before that low, Oracle filed a press release reporting the largest order book relative to revenue that any large software company has ever disclosed: $638 billion of Remaining Performance Obligations, up 363% year over year.

If you trade headlines, those two facts cannot coexist. If you read filings, they coexist easily — and the way they coexist is the most useful lesson of this earnings season. This is not a defense of the stock and not a price call; we never make those. It is what the documents say, arranged so you can decide what the market is actually repricing.

What the June 10 filing says

Oracle's Q4 and FY2026 results (fiscal year ended May 31, 2026, reported June 10):

• RPO $638 billion, up 363% year over year, up $85 billion sequentially from $553 billion at the end of Q3. RPO is contracted future revenue — commitments customers have signed but Oracle has not yet delivered.
• Q4 total revenue $19.2 billion, up 21%. Q4 cloud revenue $9.9 billion, up 47%. Cloud infrastructure (IaaS) revenue $5.8 billion, up 93%.
• FY2026 total revenue $67.4 billion, up 17%; FY IaaS $18.1 billion, up 77%. GAAP EPS $5.83, up 34%.
• Record operating cash flow $32.0 billion, up 54%.
• Free cash flow negative $23.7 billion.
• FY2027 guidance confirmed at $90 billion of total revenue — 34% growth on the year just ended — with Q1 cloud revenue guided up 57% to 64%.
• The Oracle Multicloud AI Database grew 404% in Q4, which the release calls Oracle's fastest growing business ever.

Revenue accelerating, order book compounding, cash from operations at a record. And free cash flow deeply negative. Hold both.

What the market is actually repricing

Here is the part the bubble-obituary posts skip. The same filing describes, precisely, how this buildout is financed:

• In fiscal 2026 Oracle raised $43 billion in debt and $5 billion in equity.
• In fiscal 2027 it expects to raise approximately $40 billion more in debt and equity, including a previously announced $20 billion at-the-market equity program. It states it does not expect to issue additional debt in calendar 2026.
• Of the giant AI contracts driving RPO, $75 billion is prepaid by customers or covered by customer-supplied GPUs — which, in Oracle's words, "substantially reduces the amount of capital Oracle must raise."

Now put Oracle's funding model next to the buyers we covered in our Mag-7 series. Meta $META generated $115.8 billion of operating cash flow in FY2025 and guides 2026 operating income above 2025 even at $125–145 billion of capex — the buildout is funded from operations. Alphabet $GOOGL raised $49.6 billion of equity in June from a position of strength — operating income up 30%, buybacks paused by choice — with the stated purpose of "capital expenditures to scale AI infrastructure and global compute" (see our Alphabet Q2 2026 results breakdown). Microsoft $MSFT grew operating income faster than revenue during its heaviest capex cycle ever.

Oracle is running the same buildout with a different balance sheet: roughly $88 billion of new debt and equity across two fiscal years, negative free cash flow, and a growth rate that has to stay extreme for the math to compound comfortably. The market has cut the price of that model by two-thirds while barely touching the cash-funded builders.

That is the honest read of 2026 so far: the market is not repricing AI demand. It is repricing leverage in the AI supply chain. Demand, as filed — 93% IaaS growth, $638 billion of signed obligations, a 404% product line — has only accelerated.

What would have to be true for each side

For the bear case to be right from here, the filed numbers that matter are these: RPO converting to revenue slower than the market expects (RPO is recognized only as Oracle delivers and customers consume), concentration risk in a handful of very large AI contracts, and funding costs on the next $40 billion. Negative free cash flow is a fact; whether it is an investment or a warning depends entirely on conversion.

For the bull case to be right, watch the quarterly RPO-to-revenue conversion rate, the Q1 FY2027 print against the 27% to 29% revenue guide, and whether the $75 billion of customer-prepaid and customer-supplied hardware keeps growing — because every dollar a customer prepays is a dollar of buildout risk transferred off Oracle's balance sheet.

Neither side needs an opinion columnist. Both sides need the June 10 exhibit and a calendar.

The transformation on offer

Once you separate demand evidence from funding structure, this whole tape gets easier to read. Stocks funded by their own cash flow are being judged on returns. Stocks funded by markets are being judged on access to capital. Same buildout, two very different risk books — and the filings tell you exactly which book each company is running.

Small first step: open Oracle's June 10 press release and read the two paragraphs under "Remaining Performance Obligations" and "Capital Investment Program and Capital Funding." They take ninety seconds and replace a hundred threads.

Sources: ORCL Q4 FY2026 press release, June 10, 2026 (acc 0001193125-26-265848, sec.gov); ORCL Q3 FY2026 press release, March 10, 2026 (acc 0001193125-26-100148); GOOGL Q2 2026 8-K, July 22, 2026 (acc 0001652044-26-000066); META Q1 2026 and Q4 2025 releases (acc 0001628280-26-028364, 0001628280-26-003832); MSFT Q3 FY2026 8-K (acc 0001193125-26-191457). Filed figures retrieved via Aether; price data from EvidInvest market data as of July 27, 2026.

Research, not investment advice.