Oracle has written $664 billion to the log. The commit is still years away.
In a database, a change is written to the write-ahead log long before it is committed to disk — and until it commits, it is a promise, not a fact. Oracle has logged a $664B backlog and $161.8B of property against negative $28.7B of trailing free cash flow, funded at BBB−. Q1 FY27 beat on every line and the stock still closed down. Six analyst lenses, three scenarios, four horizons.
Gray line = Oracle’s actual monthly closes from Sep 2024 into today — the +36% single-day repricing of 10 Sep 2025, the $331 high on 11 Sep 2025, the worst week since the 2001 dot-com bust in late June 2026, the $114.50 low on 28 Jul 2026, and the round trip back to $150.28. Colored paths = synthesized scenario midpoints forward, default weighting bear 30% · base 45% · bull 25%. Faint vertical rules mark the four horizon nodes. Wall Street’s 12-month consensus of $239 is the dotted clay line. Scale is linear; marks are mid-year.
Re-weight the scenarios
Those probabilities are a judgment call, and on a stock this binary they are the judgment call. Drag to set how likely the bear and bull cases are (base takes the remainder); the blended target below, the dotted line on the chart, and the prob-weighted row of every scenario card update live.
Six analyst lenses, six answers — from $105 to $240
The same quarter supports a price target that halves the stock and one that doubles it, depending entirely on which framework you trust. Each lens below is a synthesized expert perspective reasoned independently from the same dated facts, with its own 12-month target and the arithmetic behind it.
The Contracted Compounder
RPO of $664B is 9.3x trailing revenue — already signed, with ~50% converting inside 36 months. Demand is physically observable, not modelled: 850MW added in one quarter, 300,000+ GPUs delivered, 97.9% utilisation, renewals repricing 20% higher. The >$30B of Q1 contracts structured as prepayment or bring-your-own-hardware changes the capital model in the bulls’ favour. Gross margin fell on mix; operating margin rose 390bps to 35.3% on a headcount cut from 162k to 141k.
The Cash Counter
A great annuity has been pledged as collateral for a leveraged GPU landlord. The pre-AI Oracle — FY2024’s $53.0B revenue and $11.8B of free cash flow — is worth roughly $42/share standalone, so $108 of today’s $150 is the AI option. The cumulative cash hole is ~$66B through FY27 and widening; the $20B ATM is already spent; the $2.00 dividend (~$6B/yr) is funded by debt, not cash. FCF inflects no earlier than FY2029 — which is literally S&P’s downgrade trigger.
The Depreciation Trade
Reported earnings exist because depreciation schedules, not economics, decide when the compute dies. PP&E tripled to $161.8B with $90–95B more coming, and management did not address useful lives on the call. On a 3.5-year economic life rather than six, incremental depreciation is ~$13.1B pre-tax — about $3.44 a share. Meanwhile ~50% of a backlog Oracle won’t confirm sits with a counterparty that lost ~$14B in 2026. RPO is a backlog, not a receivable.
The Borrowed Moat
One real moat is being used to underwrite a second business that has none. Licence and support — the genuinely defensible part — is $5.5B a quarter and shrinking 3%, while the moatless OCI becomes the valuation. The 20% renewal premium is scarcity rent, not moat rent: a true moat shows up as margin expansion, and gross margin fell 730bps. Database@AWS/Azure/GCP is the one genuine structural asset. Oracle Health losing to Epic a third straight year is what eroding switching costs look like.
The Contested Tape
Price sits +6.8% over the 50-day ($140.68) and −10.4% under the 200-day ($167.65) with RSI at 51 — no trend, no edge. The 11 September outside-reversal on the best news of the year (gap to $166.00, close $150.28) is the classic exhaustion print. Short interest of 1.66% means there is no squeeze fuel: the −54% drawdown was long holders liquidating. And the correlation regime has shifted — ORCL now trades with CoreWeave and Nebius, not Microsoft and SAP, which is itself a multiple event.
The Rate-Duration Trade
Oracle is levered AI beta priced for falling discount rates in a tightening regime. The 10-year sits near 5% and markets price a hike, not the cut regime that produced the September 2025 melt-up — which explains most of a −54% drawdown as a discount-rate event, not a demand one. Credit is the transmission channel and Oracle is the marginal borrower: BBB−, CDS at an 18-year wide. Server DRAM roughly doubled and NVIDIA flagged >15% price hikes from early 2027 — same compute, more capital.
Panel median 12-month target $145; mean $154; range $105–$240. Note the shape of that distribution: the two lenses that focus on cash and on accounting land within $5 of each other near $105–$110, while the lens that focuses on contracted demand lands at $240. They are not disagreeing about the facts. They are disagreeing about whether the backlog or the balance sheet is the binding constraint.
Wall Street 12-month price targets
Forty-three firms cover Oracle; the mean target is $239.10, about 59% above today, with a Strong-Buy skew (28 strong buy, 8 buy, 7 hold, 1 strong sell). Bars are a selection sorted low to high and colored by rating; the slate dashed line is today’s $150.28.
Every target except one sits above today’s price, and the two cuts that landed on 11 Sep 2026 (Stifel $220→$200 on margin pressure; Wedbush had already gone $275→$240 in June) still left Buy ratings intact. That is the tell: the sell-side is trimming multiples, not theses. The gap between the $110 low and the $400 Street high — a 3.6x spread on the same company — is the widest dispersion in mega-cap software, and it exists because the analysts are not modelling different revenue. They are modelling different depreciation and financing. Firm targets and ratings as reported on 11–12 Sep 2026; the $110 low is the single strong-sell rating in the distribution, firm not identified.
Commit, or roll back
Synthesized scenario midpoints, dated mid-year, each built as non-GAAP EPS × an exit multiple. Returns are shown against today’s $150.28. These are illustrative frameworks for thinking about the range — not forecasts. Actual outcomes can and do fall outside the cone.
1 Year
Mid-20272 Years
Mid-20283 Years
Mid-20295 Years
Mid-2031▸ Bull case — 25% — show the assumptions & math
▸ Base case — 45% — show the assumptions & math
▸ Bear case — 30% — show the assumptions & math
▸ Why the probabilities are bear-tilted — 30 / 45 / 25
Revenue, capex, free cash flow & debt ($B)
The single most important chart in this report. Until FY2024, Oracle was a cash machine. Three fiscal years later, the capex bar is larger than the revenue bar — and the free-cash-flow bar has gone through the floor. The bull case and the bear case both live in that gap.
Fiscal years end 31 May. Capex (clay) went from $6.9B in FY2024 to $55.7B in FY2026 and a guided $90–95B in FY2027 — a 13-fold increase in three years, and in FY2027 it exceeds revenue. Free cash flow (terracotta below the zero line) went +$11.8B → −$0.4B → −$23.7B, with S&P modelling a free operating cash flow deficit of roughly −$42B for FY2027. Total debt (slate) is shown as reported including leases: $94.4B → $109.0B → $167.4B; at 31 Aug 2026 it was $155.9B on that basis, or $125.3B of reported borrowings — and neither figure includes the roughly $288B of future datacenter lease commitments disclosed in the 10-Q but not yet on the balance sheet. FY2027 debt is not guided, so no bar is drawn. The bull reads this chart as investment; the bear reads it as an obligation.
The EPS ladder underneath every target ($)
None of the price targets in this report are pulled from the air — each is this earnings path multiplied by an exit multiple. Which is exactly why the depreciation debate matters: it moves the ladder, not the multiple.
Non-GAAP is the honest basis here: reported GAAP swings on non-operating items — FY2026’s second quarter booked $2.10 of GAAP EPS against $1.27 the following quarter on investment gains, which is why the trailing 23.6× GAAP multiple flatters the stock. Gray = reported ($5.56 FY24, $6.03 FY25, $7.63 FY26). Clay solid = the FY2027 $8.10 company guidance raised on 10 Sep 2026; clay faded = our base-case FY2028 model of $10.40, which is a model and not a consensus figure. Note the FY27 step: only +6% on a year when revenue grows 34%. That compression — growth in the top line, depreciation and interest in the bottom — is the whole argument in one bar.
Nothing here says demand is the problem
Q1 FY2027 (quarter ended 31 Aug 2026), year-over-year. Read these against a stock that is down 54% over the same twelve months.
Olive = the steady core; clay = the frontier lines compounding off smaller bases. Only one line is negative, and it is the one that matters most to the moat: legacy licence and support, −3%. The disconnect the bulls point to is real — total revenue +30%, cloud +62%, OCI +121%, backlog +46%, with the stock halved. But the terracotta bar is the honest counterweight, and it is the longest on the chart: capital expenditure grew 235%, from $8.5B to $28.5B in a single quarter. Every other bar on this chart was bought with that one.
Bull vs. Bear
The entire valuation argument compresses into one disagreement: is the $664B backlog a contracted annuity that converts into cash, or a levered bet on one counterparty’s ability to pay, financed one notch above junk?
▲ THE BULL CASE
- The backlog is verifiable and still accelerating. RPO $664B, up $209B year-on-year and $26B sequentially, against ~$640B expected — 9.3x trailing revenue already signed, with roughly half converting inside 36 months.
- Demand is physical, not modelled. 850MW of capacity added in one quarter, 300,000+ GPUs delivered, 97.9% utilisation, and renewals repricing 20% higher. Full utilisation plus pricing power is a shortage signature, not a bubble signature.
- The capital model changed in Q1. More than $30B of new AI contracts were structured as customer prepayment or bring-your-own-hardware — no incremental Oracle capital — which is why $90–95B of gross capex is framed as net cash capex of “not more than $70B,” and why operating cash flow jumped 184% to $23.1B.
- Operating leverage is already arriving. GAAP operating margin 35.3% vs 31.4%, GAAP EPS +54%, achieved while headcount fell from ~162,000 to ~141,000. The gross-margin decline is mix; the operating line is going the other way.
- Concentration is diluting, not concentrating. Management says the non-OpenAI backlog more than doubled year-on-year, with AMD, Meta, NVIDIA, xAI and TikTok named alongside it.
- The multiple already de-rated. 18.6× guided FY27 earnings with a 0.58 PEG, against Microsoft at ~23.8× growing a third as fast. Forty-three analysts average $239 — and Guggenheim’s Street-high $400 is not a fringe call, it is a database-migration thesis.
▼ THE BEAR CASE
- The cash is going the wrong way and accelerating. Free cash flow: −$0.4B in FY25, −$23.7B in FY26, −$28.7B trailing, with S&P modelling ~−$42B for FY27. Cumulative hole of roughly $66B before FY2028 even starts.
- Depreciation is doing the work earnings should be doing. PP&E tripled to $161.8B in twelve months, Oracle did not address useful lives on the call, and shortening them toward four years is worth roughly $3.44 a share of non-GAAP EPS on the short lens’s math.
- Half the backlog is one unprofitable counterparty. OpenAI is reported at ~50% of RPO — a figure Oracle has never confirmed — while losing ~$14B in 2026 and raising capital at scale. RPO is a backlog, not a receivable, and the Stargate ramp only begins in earnest in 2027.
- The credit market moved first and it moved down. S&P cut Oracle to BBB− on 9 Jul 2026 — one notch above high yield — with explicit triggers at 4.5× leverage or no positive free operating cash flow by FY2029. Five-year CDS reached ~203bp, the widest in nearly eighteen years.
- Costs are inflating into the buildout. Server DRAM roughly doubled in Q1 2026 and NVIDIA has flagged >15% price increases from early-2027 shipments. Gross margin already fell 730bps year-on-year while guidance was raised. Something has to give.
- Equity is now a funding source, not a return. The $20B at-the-market program was fully spent during Q1; diluted share count rose ~144M year-on-year; the $2.00 dividend, roughly $6B a year, is paid out of borrowings.
- The moat is in the shrinking half. Licence and support — the genuinely defensible business — is down 3%, and Oracle Health lost hospital share to Epic for a third consecutive year, with about a third of sampled customers saying Oracle is not in their long-term plans.
Risk map — likelihood × impact
Where each risk sits over a three-to-five-year horizon, not just how large it is. The hot upper-right corner is the one that matters — and unusually for a mega-cap, Oracle has something genuinely sitting in it.
- Continued equity dilution
- Memory & GPU cost inflation
- Datacenter / power slippage
- Depreciation-life reset
- Rate / discount-rate shock
- OpenAI concentration
- Downgrade to high yield
- AI-capex digestion
- Key-man & governance
- AI credit event
Depreciation-life reset
What breaks: $161.8B of PP&E on assumed five-to-six-year GPU lives; moving toward four cuts non-GAAP EPS by roughly $3 a share and the targets fall with the ladder, not the multiple.
OpenAI concentration
What breaks: a counterparty reported at ~50% of a $664B backlog restructures or stretches its ramp, and the entire revenue bridge behind the cone resets overnight.
Downgrade to high yield
What breaks: BBB− is one notch up; a cut forces index-mandated selling of the debt, raises the cost of the entire buildout, and validates the equity de-rating.
AI-capex digestion
What breaks: ORCL now correlates to CoreWeave and Nebius, not Microsoft. A sector-wide capex pause takes the multiple regardless of Oracle’s own execution.
Memory & GPU cost inflation
What breaks: DRAM roughly doubled and NVIDIA flagged >15% increases from early 2027 — the same contracted compute now consumes more capital and thinner margin.
Datacenter / power slippage
What breaks: depreciation and interest start on schedule; revenue does not. Oracle has to keep landing 850MW a quarter into interconnection queues it does not control.
AI credit event
What breaks: a funding failure at a major AI lab or neocloud reprices the sector’s credit at once. Low odds — but Oracle is the most levered instance of the trade, with CDS already at an 18-year wide.
Rate / discount-rate shock
What breaks: a long-duration, negative-free-cash-flow asset loses roughly a turn of P/E per 25bp of terminal discount rate — and it refinances into whatever the curve offers.
Key-man & governance
What breaks: Larry Ellison is chairman, CTO, ~40% holder and the dealmaker behind the OpenAI contract; two co-CEOs since Sep 2025 and a CFO since Apr 2026 carry the execution.
Continued equity dilution
What breaks: share count is already up ~144M year-on-year after the $20B ATM; another raise caps per-share upside even in the scenarios where the thesis works.
The jargon, decoded
Hover the dotted terms in the metrics and prose, or scan the desk’s working definitions here.
- RPO / backlog
- Remaining performance obligations — revenue under signed contract that has not yet been delivered or recognised. $664B here. It is a promise to buy, not cash in the bank, and it can be renegotiated.
- Free cash flow
- Operating cash flow minus capital expenditure — what is actually left over. Oracle’s is −$28.7B over the trailing twelve months, because capex ($75.7B) exceeded operating cash flow ($46.9B).
- FCF yield
- Free cash flow divided by market value. Normally the cleanest cheapness test; for Oracle today it is negative, which is why the value lens is bearish on the stock while constructive on the business.
- EV / EBITDA
- Enterprise value (equity plus net debt, $551.7B) over earnings before interest, tax, depreciation and amortisation. 16.1× here. It counts the debt that a P/E ratio ignores.
- Useful life / depreciation
- How many years a GPU or datacenter is written off over. Longer lives spread the cost thinner and flatter near-term earnings; the argument is whether Oracle’s assumed lives match economic reality.
- BBB− / investment grade
- S&P’s lowest investment-grade rating — one notch above high yield. Below it, many funds are mandated to sell the bonds, so the cost of borrowing steps up sharply.
- CDS spread
- Credit default swap spread — the annual cost of insuring against default, in basis points. Oracle’s reached ~203bp, the widest in nearly eighteen years. The credit market’s live opinion.
- ATM offering
- An at-the-market equity program: selling new shares gradually into the open market. Oracle completed $20B of it in one quarter, which funds the buildout but dilutes existing holders.
- Exit multiple
- The P/E assumed at the end of the forecast period. Multiply it by projected EPS to get a target price — every number in the scenario cards is built this way.
- Prob-weighted
- Each scenario’s price multiplied by its probability, summed into one expected value across bear, base and bull. The clay row and the dotted line on the chart.