Industry

Infrastructure Software: The Plumbing of the Enterprise — and the New Arena for AI

Oracle competes in Software—Infrastructure: the foundational software and, increasingly, the physical computing capacity that businesses run their operations on. This is not consumer software you click on — it is the database that stores a bank's transactions, the applications that run a retailer's payroll and supply chain, and the data-center capacity that trains and serves AI models. Oracle organizes itself into three businesses — cloud and license; hardware; and services — and is the corporate successor to operations begun by Larry Ellison in June 1977 [1].

The single most important thing to understand before reading the rest of this report: this industry is in the middle of a once-in-a-generation demand shock. Oracle's contracted-but-not-yet-delivered backlog — its remaining performance obligations (RPO) — finished fiscal 2026 (ended May 31, 2026) at $638 billion, up 363% in a single year [6]. That number is more than nine times the company's annual revenue. The cause is AI: enterprises and AI labs are signing multi-year contracts to reserve cloud computing capacity years in advance. To understand Oracle, you have to understand the industry it sits in, how it makes money, and why its economics are being rewired in real time.

Sources: RPO and Q4 cloud infrastructure growth, FY2026 revenue and net capex — Q4 FY2026 earnings call [6].

---

1. What "Software—Infrastructure" actually is

Think of enterprise technology as a stack, bottom to top. Oracle is unusual because it sells at every layer of it — most rivals own only one or two.

Source: layer definitions and Oracle's position per FY2025 10-K, Item 1 Business [1]; OCI and database offerings described on p.14 [3] and p.16 [4].

The jargon, defined once:

• Database — the software that stores and retrieves an organization's data. Oracle calls its flagship "the world's most popular enterprise database," and it is the historical cash engine of the company; the newest release (Oracle Database 23ai) bakes in AI vector search so customers can run generative-AI queries directly against their own data [4].
• License + support — the legacy model: a customer buys a perpetual software license once, then pays a recurring ~22% annual fee for support and updates. Highly profitable and very sticky, but slow-growing.
• SaaS (Software-as-a-Service) — the vendor hosts the application and the customer pays a subscription. Oracle's are branded Fusion (ERP/finance) and NetSuite.
• IaaS (Infrastructure-as-a-Service) — renting raw compute, storage and networking. Oracle's is OCI (Oracle Cloud Infrastructure). Customers prepay a fee that is "decremented as the OCI services are consumed" — a consumption meter, like electricity [3].
• RPO (remaining performance obligations) — the dollar value of signed contracts not yet recognized as revenue. The industry's clearest forward-demand gauge, because it is contractual, not a forecast.

The economic prize of this industry is recurring, contracted revenue: support renewals and cloud subscriptions that recur for years with minimal incremental cost. The whole industry is migrating from one-time license sales toward this model — and Oracle is now layering a third, capital-heavy layer (renting AI compute) on top.

---

2. The defining trend: a 30-year shift from licenses to cloud, now overtaken by AI

Read Oracle's mix over five years and you see the textbook infrastructure-software transition. Cloud services rose from 32% of total revenue in FY2023 to 43% in FY2025, and management expects the trend to continue [2]. Within the cloud-and-license franchise, infrastructure (OCI) overtook applications, reaching 56% of cloud services and license support revenue in FY2025 versus 53% two years earlier [3].

Source: Applications vs Infrastructure cloud services and license support revenue, FY2025 10-K segment disclosures [3]; mix percentages on p.11 [2].

That gradual shift is the part of the story that existed before 2024. What changed everything is the AI capacity supercycle. In its September 2025 call (Q1 FY2026), Oracle disclosed it had signed cloud contracts with "the who's who of AI, including OpenAI, xAI, Meta, NVIDIA, AMD, and many others," sending RPO past $455 billion — up 359% year-over-year and up $317 billion in a single quarter [7]. On the same call management laid out a multi-year ramp: OCI to grow ~77% to roughly $18 billion in FY2026, then $32B, $73B, $114B and $144B over the following four years [8].

Source: OCI revenue trajectory provided by management, Q1 FY2026 earnings call [8]. Forward figures are company guidance, not results.

---

3. The cycle: how to read where the industry sits

Infrastructure software is only mildly seasonal in the traditional sense — Oracle notes revenue and margins are "typically highest in our fourth fiscal quarter and lowest in our first," driven by sales-compensation cycles common across IT [5]. The real "cycle" today is the investment cycle in AI capacity. The defining ratio to watch is RPO (the backlog), because it leads revenue by years.

Sources: RPO of $137.8B as of May 31, 2025 — FY2025 10-K, Note 1 [9]; FY2026 RPO of $638B — Q4 FY2026 earnings call [6]; prior-year RPO levels as reported in company segment data.

The slope here is the whole story. RPO compounded steadily from $41B to $138B over five years (FY2021–FY2025), then multiplied roughly 4.6x in one year to $638B. As of May 31, 2025, Oracle expected to recognize only ~33% of its then-$137.8B backlog within twelve months [9] — meaning most of this revenue lands in later years. This gives unusually long visibility, but it also means today's reported revenue badly understates the scale of what has been promised, and the company must spend the capital to deliver it before the revenue arrives.

A second tell is multicloud database — Oracle databases running inside rivals' clouds (AWS, Azure, Google Cloud). That revenue grew 1,529% in Q1 FY2026 and 404% year-over-year in Q4 FY2026 [7] [12]. It signals that the database franchise — the legacy crown jewel — is finding a new growth path even where customers chose a competitor's infrastructure.

---

4. Unit economics: a high-margin software model bolting on a capital-heavy utility

Historically this was one of the best business models in technology: write the software once, sell licenses and support at very high incremental margin. Oracle's operating margin was ~31% in FY2025, revenue grew 8% to $57.4 billion [10], and FY2026 revenue surpassed $67 billion for the first time with non-GAAP operating income of about $29 billion [6].

Source: revenue and operating income, FY2021–FY2025, as reported in Oracle 10-K filings; FY2025 total revenues of $57,399M on p.71 [10].

The catch is what AI infrastructure does to that pristine model. Renting GPUs is a capital-intensive utility business grafted onto a software company. Two structural shifts make this concrete:

Capex has exploded while buybacks were switched off. Oracle's capital spending rose from roughly $2.1 billion in FY2021 to $21.2 billion in FY2025 [11], then to a net cash outlay of $48 billion in FY2026 [6]. Over the same window, share buybacks collapsed from ~$21 billion (FY2021) to $0.6 billion (FY2025) — capital was reallocated wholesale from returning cash to building data centers.

Sources: capex of $21.2B in FY2025 — FY2025 10-K p.80 [11]; FY2026 net capex of $48B — Q4 FY2026 earnings call [6]; earlier-year capex and buyback levels as reported in Oracle cash-flow statements. FY2026 buyback shown as nil for scale.

Gross margin is compressing. For FY2026, management said gross margin "stepped down around 5 points as expected" as the infrastructure build-out ramps, because renting depreciating hardware carries far lower margins than licensing software [6]. Management's defense is twofold: contracts are long-dated and contractual (the $638B RPO), and a growing slice of demand is capital-light — Oracle reported $75 billion of "bring-your-own-hardware or prepaid" contracts with no margin degradation, where customers fund the hardware themselves [12].

The investor's core tension, in one line: a 30%+ operating-margin software business is funding a lower-margin, capital-hungry compute-rental business in exchange for a backlog nine times its revenue. Whether that trade creates value depends entirely on the return on tens of billions of capex — the single most important number to track from here.

---

5. Competitive structure: an oligopoly at the bottom, a scrum at the top

Oracle describes "intense competition in all aspects of our business," and names a field that reads like a roll-call of the largest companies on earth: Adobe, Alphabet (Google), Amazon, Cisco, Intel, IBM, Microsoft, Salesforce, SAP, plus HPE and Workday. It also flags "low barriers to entry in many of our market segments," where new competitors "frequently emerge" [5]. The right mental model is layer-by-layer, because the structure differs sharply by layer.

Source: competitor set and competitive dynamics per FY2025 10-K, Item 1 Business, Competition [5]; Oracle Database positioning on p.16 [4].

Scale, in context. The cloud-infrastructure layer is dominated by hyperscalers far larger than Oracle. Amazon's AWS alone generated roughly $128.7 billion of revenue and $45.6 billion of operating income in its latest fiscal year [16] — multiples of Oracle's entire OCI business, which management guides to about $18 billion in FY2026 [8]. Oracle is the challenger in IaaS, and its AI-capacity wins are how it is trying to close that gap. The flip side: a challenger growing cloud infrastructure 93% off a smaller base (Q4 FY2026) [6] is reshaping share faster than the leaders.

The broader peer group shows how varied the economics are across this "industry" — hyperscale platforms (Microsoft, Amazon), application suites (SAP, Salesforce, Workday), and the legacy enterprise estate (IBM).

Sources: latest-year revenue and net income per company income statements — Microsoft [17]; Amazon [18]; IBM [21]; Salesforce [19]; SAP, reported in EUR [20]; Workday [22]; Oracle FY2025 per 10-K p.71 [10]. SAP figures are in euros and not converted.

Two takeaways. First, Oracle's profitability is mid-pack-to-strong — well above the pure-SaaS players (Workday, Salesforce) and Amazon's blended margin, but below Microsoft, the industry's profit champion. Second, the competitor "size" varies by two orders of magnitude, which is exactly why the layer-by-layer view matters: Oracle is a giant against Workday in apps, but a challenger against AWS and Azure in infrastructure.

---

6. Regulation and the risks that define the industry's ceiling

Infrastructure software is not a heavily price-regulated industry, but it sits squarely inside three regulatory and physical constraints that any investor must price in. Oracle defers most of its regulatory disclosure to its risk factors rather than a standalone rules section [5], and the substance clusters here:

• Data privacy and sovereignty. Since the EU's GDPR, "the rate of global consideration and adoption of privacy laws has increased," widening the set of jurisdictions where Oracle can face inquiries and audits [14]. Because this business stores other companies' most sensitive data, privacy and data-localization rules are a direct driver of where data centers must be built and how products are designed.
• Export controls and geopolitics. Compliance with sanctions, export controls and other restrictions "could prevent us from serving" certain customers or markets [15] — acutely relevant when the highest-value workloads run on advanced AI chips subject to export rules.
• Power and supply chain — the new physical ceiling. Uniquely for an AI build-out, Oracle flags it "has faced, and may continue to face, rising costs for data center energy demands" and relies on third-party suppliers for the equipment to outfit those centers [13]. Electricity and chip supply, not software, are now the binding constraints on growth across the entire infrastructure-software/AI arena.

---

7. Watchlist: the few signals that would change the industry view

Sources: signals derived from the cited evidence above — gross-margin step-down and net capex [6]; RPO and backlog conversion [9]; energy and supply-chain risk [13]; multicloud growth [12].

The one-paragraph mental model. Software—Infrastructure is the recurring-revenue, high-switching-cost software that runs the enterprise, layered from databases up through applications and, now, raw AI compute. Its economics historically prized software-like margins on contracted, renewable revenue; the AI supercycle is bolting a capital-heavy utility (renting GPUs) onto that model, trading near-term margin and free cash flow for an enormous contracted backlog. Oracle is the rare full-stack player and the most aggressive AI-infrastructure challenger — its $638B backlog [6] is the industry's clearest evidence that demand is real, and its $48B of annual capex is the clearest evidence of what that demand costs. Read the rest of this report with that trade — backlog and growth against capital and concentration — as the lens.