ORCL (Trans): OCI +121%, cash flow still -$5bn

Dolphin Research's takeaways from $Oracle(ORCL.US) FY27 Q1 earnings call

I. Key takeaways

1. Capital structure: Completed the previously disclosed $20bn ATM equity offering in Q1. No update on dividends or buybacks this quarter.

2. Guidance raised

a) FY27 full year: total revenue raised to at least $90bn (+34% YoY). Non-GAAP EPS raised to $8.10.

b) Q2 FY27: total revenue +30%–34% YoY (USD basis), with cloud revenue +65%–71%. Non-GAAP EPS $1.85–$1.93 (+21%–25% YoY), excluding last year's Q2 Ampere gain.

3. RPO and conversion cadence: RPO increased by $26bn vs. Q4, with the vast majority of new contracts prepaid or BYO-hardware, requiring no incremental capital from Oracle. These additions will affect CapEx and revenue in FY28 or later.Management now expects roughly half of RPO to convert into revenue over the next 36 months.

4. Key financials this quarter

a) Topline: revenue of $19.3bn (+30% YoY), a record high, and the first-ever positive QoQ for a Q1 following a record Q4, which historically was followed by a softer Q1. Non-GAAP OP was $8.2bn (+31% YoY).Non-GAAP EPS was $1.92 (+30% YoY).

b) Margins: GPM declined as expected on data center ramp and faster infra growth, offset by lower opex and efficiency gains, keeping non-GAAP OPM around 42% and essentially flat. TTM OP growth accelerated to 21% from 16% in Q4.TTM revenue growth was 5ppts higher vs. Q4.

c) Cash flow and CapEx: operating cash flow hit a record $23bn, including customer prepayment collections. Q1 CapEx was $28bn; FCF was -$5bn.Excluding prepayments, net cash CapEx was $18bn. Full-year CapEx remains $90bn–$95bn, with net cash CapEx capped at $70bn, and spend will be non-linear.

II. Call details

2.1 Management commentary

1. OCI capacity delivery

a) Delivered 850MW of AI compute since Q4-end, including over 300k GPUs. This is nearly 3x Q4's delivery and equals 73% of last fiscal year's full-year delivery.

b) At the Abilene campus, 131k GPUs were delivered in Q1, 1.9x Q4. Six of eight buildings are online, totaling 618MW or 75% of capacity, with customer acceptance compressed to 24 hours.OpenAI trained GPT-6 Astra at this site.

c) The next GW-scale campus, Shackleford, is progressing well. NVIDIA's Vera Rubin system is exceeding expectations on hardware quality, manufacturing yield, and performance, with initial deliveries in Q2.

d) Signed over $30bn of new AI contracts in Q1. GPU utilization was 97.9%.Expiring GPU capacity is being renewed or resold at ~20% premiums to original contracts, with most assets already in service for four years or longer.

2. Database and multicloud

a) Multicloud database revenue rose 353% YoY, with multicloud customers up 180% YoY. Azure and AWS regional expansion progressed as planned, reaching 70 multicloud database regions and 119 AZs.

b) Oracle Interconnect for AWS is now GA, making OCI privately interconnected to all hyperscalers, with zero data transfer fees.

3. Models and developer platform

a) Via Oracle Marketplace, offering OpenAI API, ChatGPT for Work, and Codex (incl. GPT-6 Astra). Bringing Gemini into enterprise apps, adding Grok reasoning, multimodal and voice models, and expanding the open-source catalog to NVIDIA, Qwen, Google, DeepSeek, and more.

b) Entered a multi-year collaboration with Quantinuum; its Helios quantum computer will be deployed at U.S.-based OCI AI data centers to support hybrid quantum + AI workloads.

c) APEX 26.1 launched, now running over 2mn active apps. Introduced APEX Lang, which expresses apps as versionable, governable structured definitions to be generated and modified by AI coding agents.

d) AI Data Platform now integrates with Codex and Claude Code, and adds MLOps capabilities.

4. Applications (SaaS)

a) SaaS grew 10% overall; Fusion +14%; industry apps >20%; Oracle Health continued to accelerate. NetSuite lagged peers as deal cycles lengthened last fiscal year.

b) Embedded AI was invoked over 150mn times in the quarter (+42% QoQ). AI agents ran over 3.5mn production executions, nearly doubling QoQ, with 2,300+ agents in production at customers (+90% QoQ).Fusion alone consumed 900bn tokens.

c) NetSuite Next is GA. NetSuite AI Connector Service securely connects data to external assistants like ChatGPT and Claude, now used by 10k+ customers.Customer Every Man Jack estimates ~$350k and nearly 5,000 hours saved per year.

d) Launched the Magenta healthcare management system this month, positioned beyond EHR by linking clinical research and care delivery. At AI World in Oct., Oracle will release agentic AI deployment accelerators.

e) Customers: Uber, Stanford University, and MUFG went live or expanded on Fusion. Johnson Controls, SNB, Guidewell (serving 45mn+ people), and Petronas added Fusion Agentic apps this quarter.

2.2 Q&A

Q: Peers are still ramping CapEx aggressively. Will Oracle slow after the FY27–28 peak? Could there be a higher peak beyond FY28?

A: No view on peaks beyond FY28; CapEx is now decoupled from growth. Management has been exploring ways to fund growth for several quarters; own balance-sheet CapEx is only one mechanism.Through deep supplier relationships and new commercial models (incl. BYO-hardware), capital can be shared across parties. Separate 'Oracle-paid CapEx' from 'how fast the biz. can grow': the model still needs capital, but not all of it must come from Oracle, reflecting how AI expansion is reshaping the model.

Q: Given the current backlog, when will FCF turn positive again?

A: No timeline given; projects should turn into strong cash generators post-ramp. Management reiterated they are not providing a specific window today.By nature, these assets are strong FCF generators and, once ramped, can convert roughly ~100% of after-tax EBITDA into FCF, making the biz. self-funding at some point. The only swing factor is the pace of growth CapEx deployment, which Oracle still intends to push appropriately to sustain growth.

Q: There are reports of delays at the New Mexico and Wisconsin data centers. What is the real status, and do these affect FY27 guidance?

A: Both sites are on track, with no impact on prior FY27 revenue and profit guidance. Management first framed scale: each site is ~1GW, while none of the 850MW delivered in Q1 came from Shackleford, New Mexico, Wisconsin, or Michigan, highlighting diversification and lack of single-site concentration.Also, GW-scale sites ramp over multiple quarters rather than going live at once, so any single-site delay vs. plan will not create a concentrated quarterly hit.

Specifically, New Mexico construction is progressing well and is in the air-permit process. On-site power will use Bloom fuel cells with materially lower water use and emissions than other on-site generation.Oracle is working with Doña Ana County regulators and the community.

Wisconsin will not use on-site generation and will take grid power. Data center delivery is on track, with ongoing design and delivery adjustments with the Public Service Commission, ATC, and WE Energies.

Q: With RPO still rising, how confident are you in landing the corresponding capacity?

A: Constraints have shifted from GPUs/wafers to power and data centers, but confidence remains high. The bottleneck has migrated—first GPUs and wafers, then power, and now the data centers themselves. Global demand is large, and Oracle is advancing all viable build paths in parallel.Importantly, planning assumes not every milestone will hit 100%, with contingencies built into forecasts and external communication.

Q: With component price inflation, how are contract vs. spot prices evolving? Does your margin framework still hold?

A: Higher costs are being priced through; margin view unchanged. In a demand-constrained market, prices typically rise. Oracle is pricing up across businesses to offset higher costs.On asset life concerns, management pushed back: compute demand and data center needs have only risen over the years, and AI use cases reinforce this. The ability to renew expiring capacity at ~20% premiums is a clear positive signal for demand, growth, and profitability.

Q: What drives the RPO growth that requires no extra capital from Oracle? Are these AI labs, semis, or sovereign customers?

A: The point is no incremental cash outlay by Oracle, not zero CapEx overall. Three models: supplier financing aligns capacity payments with customer receipts; customers provide hardware and use Oracle's ops expertise, cloud tech, and DCs to form AI clusters; or customers self-finance and prepay, removing the need for Oracle to front the CapEx.Customer types span startups to top investment-grade names. The industry recognizes that 'access to capital' is the binding constraint and is reallocating capital more efficiently.

Q: Update on sovereign cloud/sovereign AI and implications for CapEx and margins?

A: Alloy-based sovereign deployments are expanding in Japan and the Middle East; no quantification provided. Partners include more commercial-leaning sovereign builds and more government-focused sovereign clouds, with broad expansion and strong demand.Many such customers are also procuring GPUs for their sovereign workloads. Management also highlighted general-purpose cloud as a large, fast-growing, and profitable business that consumes capital but far less than giant AI clusters.

Q: GPM is down but OPM is flat. How should we think about GPM vs. OPM going forward?

A: GPM gauges pricing and cost discipline; value creation is best viewed at OPM. GPM matters because it reveals if pricing is right and costs are controlled, and any model miss shows up quickly there, so it's a key internal cross-check.Currently, GPM reflects DC ramp and mix shift between software and infra. Software carries higher GPM but heavier R&D and S&M below GP; infra has lower GPM but lighter R&D/S&M burden at Oracle's scale, which reuses company-wide R&D. Hence, value creation should be assessed at OPM.

Q: From here, does GPM fall further or stabilize for the year?

A: No specific guide; expect GPM to flatten after the next two years of ramp. Management reiterated that GPM would step down this year, as flagged in Q4, and suggested backing into OPM from EPS guidance. More color will come at the Oct. Investor Day.

Q: With AI potentially disrupting SaaS, how do you sustain double-digit growth over the next two years?

A: Three layers of differentiation: end-to-end suites, embedded AI, and a co-planar Agentic Studio. First, a complete suite spanning horizontal and vertical industry apps across healthcare, retail, telecom, and construction, with strong Fusion and industry app performance this quarter.Second, AI is embedded into existing workflows via routine releases, delivering fast ROI without sidecar systems. Third, Fusion Agentic AI Studio shares the same control plane as the apps, benefiting from the same quarterly updates and security patches.

Management provided quant evidence for AI-accelerated go-lives: complex healthcare deployments have compressed from high-double-digit months to single-digit months. Early NetSuite users have moved from double-digit months to single-digit weeks.This accelerates value realization for customers and lets Oracle recognize go-live–linked revenue sooner.

Additionally, SaaS is a natural funnel for IaaS: SaaS customers already consume OCI and are inclined to move non-Oracle workloads to OCI.

Q: What is the biz. model for AI Data Platform? Is it to drive DB/OCI consumption or a standalone software opportunity?

A: All of the above, with no lock-in to Oracle data sources. The platform can run on 100% non-Oracle workloads, pull from any data source, and auto-generate enterprise ontologies. It already supports hundreds of data sources.Whether consumed standalone, with apps, or with OCI, Oracle prioritizes customer/partner choice, making it accretive across lines. As ontology automation uses Oracle Database as a key source among others, it also eases cloud migration of on-prem DBs and boosts multicloud DB growth.

Q: Will you deploy forward-deployed engineers around this platform as a GTM lever?

A: Yes, already deploying, shared with Fusion Agentic Studio. Oracle has forward-deployed engineers at customer sites.Both run on OCI under the same control plane and are treated as a combined platform. These services are essential to speed customer go-lives, which are already measured in weeks for some projects—a timeline unimaginable a year ago in heavily regulated sectors.

<End of transcript>

Risk disclosure and disclaimer:Dolphin Research Disclaimer & General Disclosure