Day 1 Keynote
Oracle: Where AI Means Business

Oracle used its October 26, 2026 AI World keynote to argue that enterprise AI pays off only when it acts inside the system of record, under that system's permissions, and leaves an audit trail behind. Oracle is one of very few vendors that owns all three layers that argument needs (the database, the cloud and the business applications), which makes the pitch credible. The harder question is whether Oracle's applications business can mirror the momentum of its infrastructure business, growing 121%.

​

The gap between AI users and AI winners is wide and not closing. McKinsey's 2026 survey found nearly 90% of organizations use AI, yet only about 6% get at least 5% of EBIT from it, a share unchanged from last year. The winners differ in one habit: nearly three-quarters have redesigned workflows around AI, against about a quarter of everyone else. Oracle's packaged agentic apps are pre-redesigned workflows. Your edge comes from how fast and how safely you put them to work.

​

Five conclusions for customers, prospects, partners and investors:

​

• The most consequential announcement was the least flashy. The open-source Fusion Setup and Configuration Registry turns implementation know-how into machine-readable skills that agents execute. If it works at scale, it lowers the largest hidden cost of owning ERP: the implementation.

• Oracle conceded the front door and is fighting for the vault. Fusion Companion ships first as a plugin inside ChatGPT. Oracle is betting that the chat interface becomes a commodity while the governed record, its permissions and its receipts do not.

• "AI with receipts" is Oracle's answer to the failure rate. Gartner expects more than 40% of agentic AI projects to be canceled by the end of 2027, citing cost, unclear value and weak risk controls. Oracle's auditable "outcome receipt" targets the third of those directly.

• Infrastructure momentum is real; applications momentum is still a promise. OCI revenue more than doubled last quarter while cloud applications grew 10%. The agentic apps are Oracle's bid to close that gap, and pricing has not been published.

• Balance-sheet risk is an investor issue first and a customer issue second. Oracle carries $125B of debt, a BBB- rating from S&P and negative free cash flow, with about half its $664B backlog tied to OpenAI. Customers should watch it, not panic over it.

​

What did Oracle announce at AI World 2026?

​

CEO Mike Sicilia framed the keynote, and Chris Leone, EVP of Applications Development, delivered the product news. Leone's most pointed claim: "There's no such thing as what's coming. Everything that I'm talking about here is available to you right now in the current release that you're running."

​

​

​

​

 

 

 

 

 

 

 

 

 

 

Quotes worth remembering

​

"AI should allow you to barely think about software at all." - Mike Sicilia, CEO

​

"Why should someone have to navigate six systems to accomplish one thing?" - Mike Sicilia

​

"More AI intelligence does not mean fewer controls." - Mike Sicilia

​

"This is AI with receipts." - Chris Leone, on the auditable "outcome receipt" each Claw job produces

​

"Agents will be the operator, and the registry will be the authority." - Chris Leone, on the future of implementation

​

"Early on, we used governance… as an accelerator versus a blocker." - Spectrum's HR technology leader, on rolling out AI to more than 100,000 employees

​

"If you have the right data, AI is pretty simple. The real challenge is developing the product… which is really useful." - Head of AI at a European restaurant operator running about 1,500 locations

​

"The success of AI should be defined by how much more becomes possible for people." - Mike Sicilia, closing

​

What the demos proved

​

The load-planning demo carried the most evidence. A Claw job raised truck utilization from 61%, consolidated loads for about $58,000 in savings, processed 149 loads automatically and stopped five for human review because they hit governance limits. That stop is the product: autonomy with a brake, plus a receipt showing who acted, what changed and which procedure applied.

​

Spectrum's demo showed the scale case: one employee request about a family emergency triggered leave, manager coverage, scheduling and a leave-eligibility check across 45 states of rules. Its leader noted requests evolved from "when do I get paid?" to commission and deduction analysis, then to an AI-first HR help desk. The emergency-medicine demo moved patient records, telemetry and ultrasound data to the ER before arrival, which Sicilia described as "not just humans in the loop."

​

How does Oracle stack up against SAP, Salesforce, Workday and the hyperscalers?

​

Oracle is the only vendor here that owns the database, a hyperscale cloud and a full ERP, HCM and supply chain suite. That vertical position is its edge. Its weakness is that application peers are growing faster and disclosing more about how agents make money.

​

​​

​

​

​

​

​

​

​

​

​

​​Where Oracle is leading

​

Oracle's agents inherit one permission model, one data model and one audit trail, because Fusion, the AI Database and OCI come from the same engineering organization. Rivals assemble that from partners. Spectrum's leader put the buyer's view plainly: "The technology wasn't the challenge." The challenge was making policies, approvals and records across 45 states behave as one experience.

 

The registry is the second advantage, and the less imitable one. SAP and Workday can build agents; few rivals have offered to turn their implementation playbook into open, executable code maintained by product development every release. Industry applications growing more than 20% and multicloud database revenue up 353% show demand where Oracle's data and domain depth meet.

​

Where Oracle is contending

​

Oracle's cloud applications grew roughly in line with SAP's cloud growth of 26% and Workday's subscriptions grew 13.9%. Salesforce reports an agent revenue line; Oracle does not. Until Oracle publishes pricing and adoption numbers for agentic apps, buyers cannot model the cost and investors cannot model the upside.

​

Distribution is the subtler exposure. Shipping Fusion Companion inside ChatGPT, and Leone's aside that tens of millions of Fusion users "don't spend every single day in Codex and Cowork," both concede that the daily interface may belong to someone else. That works only if the system of record stays the place where value is captured.

​

Will Oracle's AI strategy pay off?

​

Oracle's infrastructure business keeps compounding through fiscal 2027 because demand outruns supply, and its applications business re-accelerates modestly, from 10% toward the low teens, over the next four to six quarters. The constraint on the first is capital. The constraint on the second is proof that agentic apps produce revenue, not just demos.

​

OCI is outgrowing every hyperscaler, while Oracle's cloud applications trail every major applications peer. Closing that second gap is the job the agentic apps were built to do.

​

The infrastructure case rests on reported numbers. Oracle booked more than $30B of additional AI cloud contracts in its first quarter, delivered 850MW of new datacenter capacity and guided to at least $90B of fiscal 2027 revenue. The market is pulling, too: Synergy Research measured cloud infrastructure spending up 43% to $143B in Q2 2026, the fastest growth in eight years, and named Oracle among the fastest-growing providers. Gartner forecasts worldwide AI spending of $2.6 trillion in 2026, up 47%, with infrastructure the largest segment.

​

The applications case rests on timing. SAP ECC customers face the end of mainstream maintenance in 2027, and implementation cost has long been the main reason they delay any migration. A registry that cuts configuration effort lowers the cost of choosing Fusion at exactly the moment those customers must choose something. Fusion applications already grow 14% and industry applications more than 20%; agentic apps give sales teams a reason to reopen those conversations.

​

The risk case is financial. S&P cut Oracle to BBB-, one notch above junk, on July 9, 2026, and forecasts a fiscal 2027 free operating cash flow deficit of nearly $42B on capex of $90–95B. About half the $664B backlog is tied to OpenAI. If that customer's funding tightens, Oracle's growth story shifts from "capacity-constrained" to "overbuilt."

​

​

​

​

​

​

​

​

​

​

​

​

Signals to watch

​

• Q2 FY27 earnings, expected in December 2026: whether cloud applications growth moves above 10% and whether Oracle discloses any agentic apps metric.

• Published pricing: for agentic apps, Fusion Companion and Claw compute. Oracle has historically bundled Fusion AI agents into existing subscriptions; Claw's long-running jobs consume real compute, so watch whether that holds.

• A second registry proof point: the 80% figure comes from one early adopter, Lexington Health, describing "certain" design and configuration efforts.

• Partner behavior: whether the large systems integrators build practices on the registry or quietly route clients elsewhere.

​

What does this mean strategically, for customers and prospects?

​

Audit trails become a procurement requirement, not a nice-to-have. Once agents execute transactions, ask every vendor what Oracle now claims to provide: who acted, what changed, which policy applied and where the agent stopped for a human.

​

Questions customers, partners and investors are asking:

​

• What is Fusion Claw? Fusion Claw is Oracle's governed agent runtime inside Fusion Applications. It handles long-running specialist work such as load optimization, reconciliation and workforce planning, stops at policy limits for human review, and produces an auditable "outcome receipt."​

​

• Is any of this available today, or is it roadmap? Oracle says it is live. Leone told the audience everything shown was "available to you right now in the current release." General availability dates, editions and prices for Fusion Companion and Agentic Home were not published in the press materials.

​​

• How much do the agentic applications cost? Oracle has not said. Forkast noted the launch came "with no price attached." Ask your account team whether apps, Companion and Claw compute are included in current subscriptions or metered separately.

​​

• Will the Setup and Configuration Registry make my implementation cheaper? Probably, for configuration and testing; less so for data migration, integration and change management, which are usually the larger risks. The only published figure is Lexington Health's 80% reduction in "certain design and configuration efforts."

​​

• Does Fusion Companion mean my data goes into ChatGPT? Oracle says Companion works through your existing Fusion security, authorization and permissions, and ChatGPT is one supported front end among planned others. Confirm data-residency and retention terms with both Oracle and OpenAI before enabling it.

​​

• Should Oracle's debt worry me as a customer? It is a shareholder risk more than a customer risk today. Oracle's applications and support businesses generate steady cash, and its revenue is still growing 30%. Prudent customers secure renewal price caps and exit-assistance terms anyway.

​​

• Is Oracle ahead of SAP in AI? Oracle is ahead on owning the full stack and on open implementation tooling. SAP is ahead on cloud applications growth, 26% against Oracle's 10%. Salesforce, not either ERP vendor, currently leads on disclosed agent revenue.

​

What's your edge?

​

For CEOs: 

• Answer Sicilia's question with a price tag. List the three "problems you learned to live with" that cost the most, such as freight under-utilization, late closes or HR ticket backlogs. Fund them as outcomes with owners and numbers, not as pilots.

• Make governance the accelerator. Copy Spectrum's model: one standing IT, security, legal and HR review that sorts use cases into repeatable-and-governed (agents act) or sensitive-and-exceptional (humans decide). It shortens approvals instead of adding them.

• Start where the desks aren't. Technicians, store staff and nurses gain the most from voice-first, mobile Agentic Home because they never learned the screens. Measure hours returned to customers and patients.

​

For CFOs:

• Get the price before the pilot. Ask for written pricing on agentic apps, Companion and Claw compute, and model a scenario where seat counts fall. About 20% of McKinsey respondents say AI operating costs, including tokens, already constrain their AI use.

• Use Oracle's timing as leverage. Oracle needs agentic-app references and faster applications growth to balance an infrastructure story that S&P has rated one notch above junk. Early adopters can negotiate price protection, renewal caps and outcome-linked terms now.

• Re-baseline every implementation budget. Ask partners for a registry-based estimate, line by line, and tie fees to milestones. Treat the 80% Lexington Health figure as a hypothesis to test on your configuration work, not as a discount to assume.

​

For CIOs:

• Fix the data before the agent. The restaurant operator's head of AI said it best: with the right data, "AI is pretty simple." Audit master data quality in each workflow you intend to hand to an agent.

• Set the brakes deliberately. In Oracle's demo, five of 154 loads stopped for review. Define the thresholds by value, risk and reversibility in each domain, and start governed full-auto on high-volume, low-risk work.

• Clone the registry this quarter. It is a public Git repository. Have your team test it against your current configuration and use it for the next module rollout or quarterly update regression.