---
title: "ORCL (Trans): AI biz. GPM at 32% this quarter"
type: "Topics"
locale: "en"
url: "https://longbridge.com/en/topics/39192236.md"
description: "Below is Dolphin Research's FY26Q3 earnings call Trans for $Oracle(ORCL.US). For analysis of the print, see 'All-in to survive — did ORCL's AI bet win?'.I. Core info recap1) Financing plan: In Feb, the company announced a plan to raise up to $50bn via debt and equity, and committed not to issue additional bonds in calendar 2026.Within days of the announcement, it raised $30bn through a mix of IG bonds and mandatory convertible preferreds (MCPS), with the book heavily oversubscribed. The ATM equity component has not yet launched..."
datetime: "2026-03-11T04:09:10.000Z"
locales:
  - [en](https://longbridge.com/en/topics/39192236.md)
  - [zh-CN](https://longbridge.com/zh-CN/topics/39192236.md)
  - [zh-HK](https://longbridge.com/zh-HK/topics/39192236.md)
author: "[Dolphin Research](https://longbridge.com/en/news/dolphin.md)"
---

> Supported Languages: [简体中文](https://longbridge.com/zh-CN/topics/39192236.md) | [繁體中文](https://longbridge.com/zh-HK/topics/39192236.md)


# ORCL (Trans): AI biz. GPM at 32% this quarter

**Below is Dolphin Research's transcript of**$Oracle(ORCL.US) **FY26 Q3 earnings call. For our take on the results, cf. '**[**Betting to Survive: Did Oracle’s ‘AI wager’ pay off?**](https://longbridge.com/zh-CN/topics/39191852)**'**

**I. Key takeaways from the print**

1\. **Financing plan**: In Feb, Oracle announced plans to raise up to $50 bn via debt and equity, and committed to no further bond issuance within calendar 2026. Within days of the announcement, it raised $30 bn through an investment-grade bond plus mandatory convertible preferred combo, with the order book heavily oversubscribed. The ATM equity program has not been launched yet. The company remains committed to maintaining an IG rating.

2\. **Quarterly performance**: FQ3 was the first quarter in 15 years with 20%+ growth in both organic total revenue and organic Non-GAAP EPS in USD terms. Revenue was $17.19 bn (vs. cons. $16.89 bn, +1.7% beat) and Non-GAAP EPS was $1.79 (vs. cons. $1.69, +5.9% beat).

3\. **Guidance**: For Q4 FY26, at constant FX upper bound, Oracle guides total revenue +20% YoY, cloud (IaaS+SaaS) +48%, and Adj. diluted EPS +17%. **FY27 revenue guidance was raised from $85 bn to $90 bn, implying an acceleration to approx. +34% YoY in FY27 total revenue.**

4\. **RPO**: Remaining performance obligations reached $553 bn.

5\. **TikTok equity investment**: In Jan, TikTok U.S. completed the carve-out of its data biz. Oracle holds 15% of the newco and has a BOD seat. This does not impact current services revenue. The equity stake will be accounted for under the equity method, with Oracle recognizing its share of P&L for the period from end-Jan to Mar 31 in FQ4 (with a two-month lag), booked in non-operating income.

**II. Details from the call**

**2.1 Prepared remarks**

1\. **Cloud apps (SaaS)**

a. At cc FX, cloud apps revenue grew 11% YoY, with ARR reaching $16.1 bn. Fusion ERP +14%, Fusion SCM +15%, Fusion HCM +15%, Fusion CX +6%. NetSuite +11%. Industry SaaS (hospitality, construction, retail, banking, restaurants, local gov., and telco) grew 19% combined.

b. At cc FX, cloud apps deferred revenue grew 14%, outpacing the quarter’s 11% revenue growth and supporting further acceleration. c. Over 2,000 customers went live during the quarter, with median time-to-live continuing to shorten.

d. On the 'SaaS doom' narrative, management believes the risk is failing to adopt AI coding tools, not AI itself. Oracle is embracing AI, using AI coding tools to let smaller engineering teams deliver more complete solutions faster. Over 1,000 AI Agents are already deployed in Fusion, with hundreds in the banking suite, all included as in-app features at no extra charge.

e. Leveraging AI and small engineering teams, Oracle built three new CX apps: lead gen and qualification, sales orchestration and automated selling, and a website generator now used for the new oracle.com. These three products are not available at Salesforce.

f. Notable wins include Memorial Hermann (Fusion ERP/SCM/HCM, displacing Workday), the University of New South Wales (displacing Workday), multiple wins vs. SAP, and a major Wall St. bank standardizing on Fusion ERP to replace SAP.

2\. **Multi-cloud database**

a. Multi-cloud database revenue grew 531% YoY, with demand exceeding supply. b. Global regional coverage is live across partner clouds: 33 regions on Microsoft, 14 on Google. On AWS, Oracle entered Q3 with 2 regions live and exited with 8, tracking toward 22 by end-Q4.

c. AI is accelerating adoption of database cloud services. Customers need to move their most valuable data to the cloud to use the latest AI capabilities (vector embeddings, MCP server access, advanced security controls, etc.), and to co-locate data with AI Agents.

3\. **AI infrastructure**

a. AI infra revenue grew 243% YoY, with demand for GPUs and CPUs continuing to outstrip supply. b. **Through partners, Oracle has secured over 10 GW of power and DC capacity for the next three years; over 90% is funded, with the remainder expected to close this month.**

c. Operating efficiency keeps improving: **standardized DC build designs, 3x manufacturing sites, and 4x rack output; time from rack delivery to revenue shortened by 60%.** d. New biz model gaining traction: since the last call, **the 'customer-supplied hardware + prepay' model has signed over $29 bn in contracts, enabling expansion without incremental debt or equity.**

e. **In Q3, Oracle delivered over 400 MW of capacity to customers, with 90% on time or early.** f. **Q3 AI delivery GPM was 32%, above the 30% guide.** Adjacencies (general compute, storage, load balancing, security, etc.) account for 10%-20% of AI DC spend with higher margins. **Database services carry 60%-80% GPM.**

g. Oracle continues to offer the latest accelerator options, including NVIDIA, AMD, and emerging designs such as Cerebras and Positron.

4\. **OCI tech wins**

a. Lockheed Martin chose OCI high-performance compute to extend AI. Lucid Motors picked OCI core services to enter Europe. Brazil’s Claro selected OCI Alloy for sovereign AI. Air France-KLM won in multi-cloud with Oracle Database@Azure, achieving 13x performance and significant cost reductions.

**2.2 Q&A**

**Q: Where do you see halo effects from AI infra across Oracle’s other businesses? And any color on FY27 capex?**

A: We do see pronounced halo effects across several areas. First, apps: because we train many models on OCI and the models are close to the apps, we can embed high-quality AI services directly as features. We are the custodian of customers’ mission-critical/data systems, and tightly coupling that data with AI models lets customers realize AI value very quickly. Second, we help customers create budget using OCI infrastructure. We are faster and cheaper than all competitors, and when customers consider large-scale app or infra transformations, **we often free up budget by moving workloads to OCI given better performance, efficiency, and lower cost.**

Third is **sovereign AI**. Our sovereign story is not new nor a reaction to recent events, and the global pipeline continues to grow. We are highly differentiated in form factor, delivering the full OCI stack whether it is three racks or 500 racks, which is a major market differentiator. On capex, **we plan to provide FY27 guidance next quarter after fiscal year-end**. Note the **decoupling of capex from Oracle’s own cash needs — new financing mechanisms may support additional capex without Oracle funding it directly**. We remain committed to an IG rating, and **the current financing plan remains as announced — up to $50 bn within 2026.**

**Q: As Oracle pivots more toward AI inference, what is the data center siting strategy? Do large centralized DCs far from population/fiber hubs conflict with inference needing to be closer to users?**

A: Inference demand is growing rapidly and broadly as model utilization rises and new use cases emerge; anyone who has used Claude or Codex lately knows these tools are changing how we do everything. **On siting, customers care about location for different reasons — cost, availability, or sovereignty.** For latency specifically, latency is relative: if you are doing ultra-low-latency stock trading, a 100 ms cross-coast round trip is unacceptable, but if you are asking a model a business question and the model needs a few seconds to think, an extra 40 ms from NYC to Wyoming will not matter.

In practice, **the current latency bottleneck is less about where the hardware sits and more about what hardware is deployed**. That is why you see so much innovation in AI accelerators — Groq, Cerebras, Positron and others are focused on cutting inference cost while sharply reducing latency, and NVIDIA’s GTC next week will have interesting news. Industry-wide, lowering latency first requires different inference architectures, while **location plays only a small role, giving us more flexibility to place DCs where power and land are abundant** to optimize resources for growing demand.

**Q: How big is the opportunity in AI databases and AI data platforms? What are you seeing in private data training and private LLMs, and how confident are you in the AI database inflection discussed at the Oct Analyst Day?**

A: Two parts: the extent of private LLM adoption, and how AI is used with private data. **Early on, many expected most customers to train their own LLMs; that has not been the case. The fast-growing pattern is customers using the best models and combining them with their private data in a private manner, and we see significant demand here.**

As Mike noted, embedding AI models in our apps is one use case. But not everything runs on Oracle apps, and there are many custom apps. So we added capabilities to Oracle AI Database to connect easily via MCP server or natural language-to-SQL. We also built our AI Data Platform precisely for this problem: you have lots of data — app data, custom data in a data warehouse, or structured DB data — and the platform unifies it, provides an agent platform to build apps quickly, and offers access to best-of-breed models from multiple vendors.

To use the latest and best AI, customers first need to be in the cloud, yet much data still is not. So we are seeing an acceleration in moving the most important private data to cloud environments to leverage the latest AI.

**Q: After completing major debt financing, how confident are you in value creation for AI DCs when considering build costs and financing costs? How does sovereign cloud benefit from the AI DC business?**

A: On overall profitability of AI DCs, there are two layers. First, the accelerators themselves: **we have guided to 30%-40% GPM**, and that remains intact. As we improve DC operations, lower delivery costs, and optimize network, hardware, and power costs, we are seeing a steady upward trend. Second, AI DCs include more than accelerators: there is substantial general compute, high-performance or large-scale object storage, load balancing, identity, and security. These adjacencies typically account for 10%-20% of total spend and carry higher margins, and our rapidly growing multi-cloud database biz. carries 60%-80% GPM. Taken together, OCI margins continue to strengthen.

The only current drag on profitability is costs from a large number of projects under construction. Delivered capacity is contracted at very attractive pricing, and as delivery speeds up, profitability will continue to improve. On sovereign, a year ago it was only about data sovereignty; now it spans sovereign data, sovereign operations, and sovereign contracts. Our Alloy model fits all three, and unlike competitors who deploy edge-only sovereign regions, we deploy the full OCI stack with all services, and can run the full apps suite and AI Data Platform inside sovereign regions with a better margin structure.

Sovereign regions can be large or small with full flexibility, and we can draw sovereign boundaries to customer needs — not just by country. Multinationals can establish sovereign regions across their own DCs to serve specific verticals.

**Q: Will AI usher in 'SaaS doom'? Are customers discussing this thesis?**

A: **In my customer conversations, I have not met a single client ready to abandon their merchandising system**, core banking, checking accounts, or EHR system and replace them with a patchwork of small AI features. In fact, I hear the opposite. **Customers ask: how much AI is embedded out of the box in your apps, and how fast can we adopt it?** They view this as the best path to AI value. We operate highly complex, mission-critical systems with decades of industry knowledge and regulatory compliance, which customers use to run enterprises, governments, and healthcare providers.

We like our position. We have deployed 1,000+ AI Agents in Fusion, with hundreds just in banking. We believe AI is disruptive, and we are the disruptor — embedding AI directly into apps and delivering it free as part of quarterly updates. Rather than declaring SaaS dead, at least for Oracle, AI strengthens our SaaS position and helps us move faster to market.

**Q: With AI Agent Studio launched in Fusion, how will Oracle’s role evolve as many vendors compete to become the AI interaction layer across enterprise systems?**

A: Data gravity is crucial here, especially for mission-critical data. We launched AI Agent Studio inside Fusion, which is the system of record for customers’ operational and mission-critical data. If you are going to build many AI Agents, where do you start? With the system of record, the center of data gravity, because from inference and RAG perspectives, that data is highly relevant and specific and adds rich context. Agent Studio is not limited to Fusion data: you can build Agents across our industry apps and third-party apps, and third parties can build Agents in it. We offer a best-of-both-worlds solution — a full AI-powered SaaS suite plus the ability to create custom Agents — delivered within standard quarterly platform upgrades and security patching.

Larry added: beyond prebuilt Agents, our AI Data Platform provides a full IDE where customers can build their own Agents using any AI model in Oracle Cloud — for coding, multi-step reasoning, and queries. For example, in the Fusion accounting system, we will launch an autonomous Agent to perform the 'close' — in the future, period close will be done by an AI Agent without human intervention. More importantly, AI lets us extend the boundaries of the SaaS suite to automate entire ecosystems. In healthcare, Epic automates acute-care hospitals, while Oracle automates the entire ecosystem — acute care, clinics, labs, insurers, HCM (nurse training, radiologist scheduling, etc.), hospital finance, FDA approval workflows, and pharma collaboration. The same applies in financial services and retail. AI coding tools enable comprehensive, agent-based software to automate whole ecosystems like healthcare or financial services. That is why we think 'SaaS doom' may apply to others, but not to Oracle.

<End\>

**Risk disclosure and statement:**[**Dolphin Research Disclaimer and General Disclosure**](https://support.longbridge.global/topics/misc/dolphin-disclaimer)

### Related Stocks

- [Jiangsu Azure Corporation (002245.CN)](https://longbridge.com/en/quote/002245.CN.md)
- [Advanced Micro Devices, Inc. (AMD.US)](https://longbridge.com/en/quote/AMD.US.md)
- [ByteDance (BYTED.NA)](https://longbridge.com/en/quote/BYTED.NA.md)
- [SAP SE (SAP.US)](https://longbridge.com/en/quote/SAP.US.md)
- [Alphabet Inc. (GOOGL.US)](https://longbridge.com/en/quote/GOOGL.US.md)
- [Salesforce, Inc. (CRM.US)](https://longbridge.com/en/quote/CRM.US.md)
- [Oracle Corporation (ORCL.US)](https://longbridge.com/en/quote/ORCL.US.md)
- [Alphabet Inc. (GOOG.US)](https://longbridge.com/en/quote/GOOG.US.md)
- [NVIDIA Corporation (NVDA.US)](https://longbridge.com/en/quote/NVDA.US.md)
- [Lucid Group, Inc. (LCID.US)](https://longbridge.com/en/quote/LCID.US.md)
- [Microsoft Corporation (MSFT.US)](https://longbridge.com/en/quote/MSFT.US.md)
- [GraniteShares 2x Long AMD Daily ETF (AMDL.US)](https://longbridge.com/en/quote/AMDL.US.md)
- [Direxion Daily GOOGL Bear 1X ETF (GGLS.US)](https://longbridge.com/en/quote/GGLS.US.md)
- [Direxion Daily GOOGL Bull 2X Shares (GGLL.US)](https://longbridge.com/en/quote/GGLL.US.md)
- [GraniteShares 2x Long NVDA Daily ETF (NVDL.US)](https://longbridge.com/en/quote/NVDL.US.md)
- [XL2CSOPNVDA (07788.HK)](https://longbridge.com/en/quote/07788.HK.md)
- [XI2CSOPNVDA (07388.HK)](https://longbridge.com/en/quote/07388.HK.md)
- [YieldMax NVDA Option Income Strategy ETF (NVDY.US)](https://longbridge.com/en/quote/NVDY.US.md)
- [Direxion Daily NVDA Bear 1X ETF (NVDD.US)](https://longbridge.com/en/quote/NVDD.US.md)
- [T-REX 2X Long NVIDIA Daily Target ETF (NVDX.US)](https://longbridge.com/en/quote/NVDX.US.md)
- [ORACLE CORP DEPOSITARY SH REP 1/2000TH PFD SER D (ORCL-D.US)](https://longbridge.com/en/quote/ORCL-D.US.md)