History

History — The Cloud Promise That Finally Arrived (and What It Cost)

For more than a decade Oracle was the database colossus that "missed the cloud" — a cash machine whose cloud promises chronically arrived a year late and a growth-point short. Across fiscal 2022–2024 management told a patient, repetitive "fastest-growing OCI" story while the contracted backlog crept along. Then, in fiscal 2025–2026, the bookings genuinely showed up: remaining performance obligations (RPO) exploded from roughly $98 billion to $638 billion on AI-training demand, vindicating a fifteen-year promise [7] [21]. But the same chapter swung free cash flow from positive $11.8 billion to negative $23.7 billion and ended Oracle's two-decade buyback machine [24]. Credibility on near-term guidance has clearly improved — the team has under-promised and beaten repeatedly; the open question is whether the largest forward commitments in company history actually convert.

The team you are judging — and what they inherited

Oracle is unusual: the person who has authored its strategy for the entire period is not the CEO. Larry Ellison — co-founder (1977), Chairman and CTO — is the constant, and he drives the narrative on every call. The CEO seat changed hands twice in this window. Safra Catz served as CEO from 2014, and on September 22, 2025 she was elevated to Executive Vice Chair, with Clay Magouyrk and Mike Sicilia promoted to co-CEOs [1]. Tellingly, the new CEOs are the two operators who ran the AI buildout — Magouyrk built OCI Gen2; Sicilia rebuilt the industry applications and Oracle Health — so the succession ratified the strategy rather than changing it.

This matters for every other tab: the current leadership did not build a struggling business and fix it — they inherited a dominant one and bet it. When this chapter began, Oracle was already a high-quality, cash-gushing franchise: a near-monopoly database with sticky license-support annuities, generating enough cash to retire one-third of its shares. In fiscal 2021 the board expanded the buyback by another $20 billion and Oracle repurchased $21.0 billion of stock in a single year [2]. The strategic chapter that defines Oracle today — the AI / cloud-infrastructure pivot — began in earnest in fiscal 2024 and inflected hard in fiscal 2025. Judge this team on what they did with an inherited crown jewel, not on whether they created it.

Source: leadership transition per the Sept 22, 2025 press release [1]; chapter dating and score derived from the cited record below.

How the story drifted: from "database" to "gigawatts"

No human re-reads twenty earnings calls to watch the vocabulary move. The corpus does. The center of gravity migrated steadily — from database and license, through "Gen2 OCI / fastest-growing cloud," to AI training, and finally to the language of an electric utility: power, gigawatts, and Stargate. The heatmap below scores how heavily each theme was emphasized by fiscal year.

Source: emphasis characterized from FY2022–FY2026 earnings-call transcripts and 10-Ks; representative anchors include "big four global hyperscalers" [3], "more demand than we can supply" [6], and the gigawatt-scale buildout [22].

What vanished is as revealing as what grew. "Cerner / Oracle Health" — the headline of the 2022 chapter — faded to a footnote within two years. And the on-premise database, written off by the market as a melting ice cube, quietly became the second engine: by fiscal 2026 management was touting "the Oracle database is booming" inside Azure, GCP, and AWS via MultiCloud, with multi-cloud database consumption growing several hundred percent.

The promise era (FY2022–FY2024): patient backlog, a $28 billion detour

Early in the window Ellison was already framing Oracle as one of "the big four global hyperscalers," alongside Amazon, Microsoft, and Google [3] — a claim that, at ~$2 billion of quarterly OCI, was more aspiration than fact. The defining capital-allocation move of this era was the $28.2 billion acquisition of Cerner, closed June 2022 — Oracle's largest deal ever, financed almost entirely with debt to plant a flag in healthcare IT [4]. Cerner did jolt the backlog — RPO jumped to $61.2 billion, up 68%, in the second quarter of fiscal 2023 [5] — but the healthcare thesis was quickly overtaken by a bigger one.

The genuine inflection signal first appears in fiscal 2024, in plain language: "we have far more demand than we can supply… our biggest challenge is building data centers as quickly as possible" [6]. By the fourth quarter of fiscal 2024, Catz reported RPO of $98 billion (up 44%), crediting "record sales in OCI… large language model training" [7]. And here is the single best evidence of management's honest framing: in the third quarter of fiscal 2024, rather than spinning, Catz told investors her own multi-year targets "might prove to be too conservative given our momentum" [8]. Two years later she was proven right — guidance that under-promises and beats is the most credible kind.

The tell (FY2025): capital expenditure breaks away from cash flow

This is the chapter break a snapshot would miss. For two decades Oracle's cash flow comfortably funded both growth and a torrent of buybacks. In fiscal 2025 that ended: capital expenditure roughly tripled, free cash flow turned negative for the first time in memory, and share repurchases collapsed to a token amount. Management was not hiding it — they were re-pointing the entire cash engine at data centers. The 10-K language hardened in step: the first prominent "if we are unable to secure data center capacity… our profitability may decline" risk [9].

Source: FY2021–FY2025 from reported financials; FY2025–FY2026 trailing-four-quarter cash-flow detail and FY2026 free cash flow of negative $23.7 billion per the FY2026 earnings release [24].

The buyback collapse is the cleanest proxy for the bet. Oracle repurchased roughly $21 billion of stock in fiscal 2021 [2]; by fiscal 2025 buybacks had dwindled to a few hundred million as every spare dollar — and then borrowed dollars — went into GPUs. In the second quarter of fiscal 2025 Ellison disclosed GPU consumption "up 336%" [10], and by the third quarter Catz called it "our strongest booking quarter ever by a huge margin as we added $48 billion to our backlog," lifting RPO to $130 billion [11]. RPO closed fiscal 2025 at $138 billion [12] — and Ellison dropped the hint that would detonate one quarter later: "If Stargate turns out to be everything is advertised, then we've understated our RPO growth" [13].

The explosion (FY2026): from $138 billion to $638 billion of backlog

He had not been bluffing. The RPO line is the hero chart of the decade — flat-ish for years, then a near-vertical move once the AI-training contracts landed.

Source: quarterly RPO disclosures — $98B Q4 FY2024 [7], $130B Q3 FY2025 [11], $455B Q1 FY2026 [14], $523.3B Q2 FY2026 [17], $638B Q4 FY2026 [21].

In the first quarter of fiscal 2026 (reported September 9, 2025) RPO leapt to "top $455 billion… up 359% from last year and up $317 billion from Q4" [14]. On the same call Catz translated the backlog into the most aggressive revenue map Oracle has ever published: OCI growing 77% to $18 billion in fiscal 2026, then "$32 billion, $73 billion, $114 billion, and $144 billion over the following four years," adding "we are off to a fantastic start" [15]. The market's reaction was captured by a veteran analyst on the call: "this is a career event happening right now… it's just amazing" [16].

The backlog kept compounding — $523.3 billion (up 433%) in the second quarter [17], $553 billion in the third on more than "10 gigawatts of power and data capacity coming online over the next three years" [19], and a record $638 billion (up 363%) to close fiscal 2026 [21]. Full-year revenue surpassed $67 billion and operating cash flow hit a record $32 billion [21]. The fifteen-year promise was, finally, delivered.

What it cost — and what is now stretched

The vindication came with the largest financial commitment in Oracle's history, and three caveats a buy-side reader must weigh.

Cash burn and a balance sheet now in the market every quarter. Free cash flow was negative $23.7 billion in fiscal 2026 [24], and fiscal 2027 capital expenditure is guided to roughly $70 billion of net cash outlay, funded by "around $40 billion in debt and equity… including our already announced $20 billion at-the-market equity issuance" [23]. The funding mix has itself become novel: by the third quarter of fiscal 2026 Oracle was layering investment-grade bonds with mandatory convertible preferred stock [20] — the company that once retired a third of its shares is now issuing equity-linked paper to feed the build.

Concentration and thin ramp margins. The backlog is contractual, but it is not diversified the way a SaaS book is. In the fourth quarter of fiscal 2026 management noted "four customers contracting for more than $8 billion this quarter" and a 97.5% GPU utilization rate [22]; the market understands OpenAI/Stargate to anchor a large share. And the economics of AI capacity are thinner than Oracle's legacy 80%+ software margins — management guides AI workloads to "the 30% to 40% range over the life of a customer contract," with reported gross margin stepping down ~5 points during the ramp [18].

The risk language confirms the bet. The 10-K is where the new risks surfaced first and grew. By fiscal 2025 Oracle disclosed that "industry supply capacity for AI accelerators, including graphics processing units, is competitive, and we at times have to accept less favorable terms with suppliers" [25], and booked $43.4 billion of additional data-center lease commitments expected to commence through fiscal 2028 [26]. The 10-K still states no single customer reaches 10% of revenue [27] — a reassurance written before the largest AI contracts began converting to revenue, and worth re-checking each year.

The promise / delivery scorecard

Weighing credibility means checking the promises that mattered to valuation against what landed. The pattern is consistent: the near-term and medium-term targets were met or beaten; the long-dated, AI-scale promises are real contracts but remain unproven on conversion and margin.

Source: targets and outcomes per the FY2024–FY2026 transcripts — "too conservative" [8], OCI ramp to $144B [15], FY2027 and FY2030 targets [23]; full-year delivery [21].

Credibility verdict: 7 / 10

Source: derived from the cited guidance/promise record above (FY2022–FY2026 transcripts and 10-Ks).

Why a 7, not higher and not lower. The case for trust is strong and rare: this team gave near-term guidance it consistently met or beat, framed its own targets as conservative rather than promotional [8], and the backlog it pointed to is contractual RPO, not narrative — it actually printed at $638 billion [21]. Ellison is promotional by temperament, but this cycle the numbers backed the rhetoric. The case against a higher score is that the entire thesis now rests on the most stretched commitments in company history: an OCI ramp to $144 billion, AI margins that are guided but not yet demonstrated, free cash flow at negative $23.7 billion, recurring equity and convertible-preferred issuance, and a backlog concentrated in a handful of AI customers of which only ~12% converts to revenue within twelve months [21]. A team that has earned trust on a two-year track record is now asking for it on a five-year one.

What the story is now

The Oracle narrative today is simpler and bigger, but no longer safer. Simpler, because the perennial "next year" hedging is gone — replaced by one clear identity (the AI cloud-infrastructure provider of record) and a backlog that removes most doubt about demand. Bigger, because the numbers have an order of magnitude they never had. But less safe, because the franchise's defining virtue for forty years — gushing, self-funding cash flow that bought back a third of the shares — has been deliberately spent down and replaced by leverage and execution risk on power, chips, construction timing, and customer concentration.

What has been de-risked: demand, and management's near-term credibility. What is still stretched: the conversion of $638 billion of backlog into profitable revenue, the balance sheet that funds it, and the assumption that thin AI-ramp margins fatten on schedule. Believe the backlog, the beats, and the demand. Discount the out-year revenue map and the steady-state margin until the data centers fill and free cash flow turns. Credibility is improving on the strength of what was delivered — but the bar Oracle set for itself has never been higher, and the next two years will decide whether this was the great vindication or the moment a cash machine over-committed at the top of a capital cycle.