Oracle AI World 2025: A New Era of AI Integration and Multicloud Strategy

Oracle AI World 2025: The First Look Under New Command

Oracle hosted its premier annual conference in Las Vegas, renamed in 2025, as "Oracle AI World". This marked the first annual conference with new CEOs at the helm: Clay Magouyrk and Mike Sicilia. The big questions top of mind for attendees are the priorities of these new leaders , how they will meet their stated growth goals for OCI, the details behind Oracle's AI strategy, and real examples showing where and how this vision comes together. The impression from Oracle AI World? Clay and Mike: two hands on the wheel, one innovating with infrastructure, the other tuning AI within the enterprise stack. All leaders focused on how Oracle solutions made tangible business gains for its mainstage customers. Notable client stories at the event included ADT, Avis, Exelon, Falabella, GM Financial, In-N-Out Burger, Lloyds Banking Group, Marriott, McDonald's Corporation, Michaels, National Police Board of Finland, OpenAI, Talcott Financial Group, Texas Children's Hospital, TikTok (ByteDance), Uber, UBQ Materials, etc. Interestingly, they included 32 sessions outlining some details of their "Client Zero" story i.e., ways they've transformed their business with AI. Here's our big takeaways from the event: * Clay Magouryk reinforces OCI messaging and doubles down on networks. Magouyrk restated his mission for OCI -- highest performance, lowest cost, and most secure infrastructure. He emphasized key principles like "everything everywhere", performance SLAs, simple global pricing, multicloud interconnect with no cross-cloud egress fees, 100G and 400G direct connect, etc. What was new? Magouryk doubled down on its networking story, announcing its scalable dedicated network Acceleron and calling out its unique architectural decisions. Other announcements included Dedicated Region25 and Multicloud Universal Credits. * Mike Sicilia told Oracle's story through the client voice. Sicilia opened with a classic "AI powering higher performing individuals" statement noting that AI at its best augments individuals for their core responsibilities while also opening their purview to more strategic insight; Sicilia quickly pivoted to changing businesses with client stories from Exelon, Avis, Marriott, and Biofy. These conversations brought out themes on AI revolutionizing supply chain, edge operations, customer experience and loyalty, and real-time decisioning, etc. * Larry Ellison's vision at AI World 2025. Delivered via a sweeping 2-hour keynote, Ellison outlined Oracle's ambition to reshape entire industries using its AI stack. He spotlighted Oracle's role in hosting leading LLMs and detailed use cases across sectors -- from fraud detection in financial services and remote patient management in healthcare, to genomic testing in biopharma, AI-driven greenhouses in agriculture, CO₂ removal via biomineralization, and commercial drone logistics. While bold and expansive, the vision is long-term, with transformative results expected over decades, not years. If Clay and Mike are at the helm, Larry is at the bow focusing on what's ahead. * Oracle's full-stack AI strategy. In the more near-term and pragmatic side, Oracle positioned itself as the "central nervous system" of the agentic enterprise, Oracle's integrated platform promises rapid value for existing customers by simplifying integration and data governance. Oracle offers 'free' agents, powered by the newly available Oracle AI Data Platform, which combines OCI, the Autonomous AI Database, and the Generative AI service. * Most notably: inclusive pricing makes budget certainty the new standard. Oracle is embedding over 600+ new AI agents directly into its Fusion Applications at 'no additional license fee'. Oracle's decision to bundle AI features at no extra cost rewrites the AI business case, replacing metered experimentation with predictable TCO - at least for now. As Executive Vice President Steve Miranda states, 'all the new AI features and agents are provided at no extra cost.' This makes AI a core subscription entitlement, not a premium upsell. Finance leaders must re-baseline multi-year models and use this as leverage to renegotiate all vendor terms that still treat AI as a premium SKU. * The Fusion AI agent marketplace: free yet unincentivized for contributors. A new hub for certified, partner-built AI agents, with ~40 launch partners (e.g., Deloitte, Stripe) each contributing at least five agents via Oracle's Agent Studio. It marks a shift from monolithic apps to composable business capabilities, enabling a network effect where shared workflows accelerate automation and raise the innovation baseline. Currently non-transactional (partners provide IP for free), legal and commercial constraints may limit participation. All agents meet Oracle's enterprise-grade security standards and benefit from full support services. It is unclear whether clients are responsible for the leverage underlying infrastructure. * Oracle Fusion Cloud HCM gets tons of feature updates + Career Coach launch: It's HCM panel started with some impressive stats: 1550+ features were delivered in 2025 alone, 60% cloud HCM customers deploying genAI and Agentic AI, 100+ embedded AI capabilities delivered in Oracle Fusion Cloud HCM through 25D, 13x YoY growth in AI usage in production, and overall AI usage surged 45x. Oracle also launched Career Coach, a new agentic AI feature in Fusion HCM that analyzes the backgrounds, skills, and interests of internal and external candidates to surface better job matches, boost applicant quality, and give hiring teams more comprehensive talent insights. What's the WIM? Oracle's new CEOs reflect its enterprise priorities: Cloud and enabling AI via its powerful enterprise app portfolio. It chose to reiterate core OCI messaging with clarity while delivering a powerful AI stack story that resets client cost expectations. But make no mistake - Oracle is positioning itself to build upon own leverage and increase that leverage with its customers. As is the case for all platforms today offering up "free" in the AI world; it is important to understand the flywheel effect as your platform dependency deepens. This reframes C-suite decisions from tech procurement to strategic alignment with Oracle's operational model -- raising the risk of outsourcing business process innovation and narrowing competitive differentiation to what Oracle's ecosystem allows. Interested in connecting with one of us about the strategy or Oracle's strategy? Clients can reach out to schedule call with our analysts at [email protected].

Oracle goes all-in on AI with an 'open' approach

It's all agents and LLMs in Vegas, and even legacy users can partake As Oracle pounds the market with AI announcements across cloud infrastructure, applications, and data analytics, experts have warned that users' path to adoption remains uncertain. Founded in 1977, Oracle built its dominance in enterprise tech from a foundation of databases and then applications through a series of acquisitions. But those attending the company's AI World in Las Vegas this week might note this is no longer the main focus, said Balaji Abbabatulla, Gartner vice president and analyst. "There's a tagline that says AI changes everything, and that definitely holds true for Oracle," he said. "They're definitely going all in for AI. Everything at the conference, everything in the keynotes, is all about AI." Among the broad gamut of announcements - on which Big Red declined to brief The Register - are AI additions to Fusion, Oracle's cloud application portfolio, which includes ERP, finance, HR, and payroll. For example, Oracle is expanding its AI Agent Studio for Fusion, which promises to help developers build, test, and deploy AI agents by extending LLM support to third-party providers including OpenAI, Anthropic, Cohere, Google, Meta, and xAI. It has also announced an agent marketplace for Oracle Fusion Cloud Applications customers to shop for AI agents built by third-party partners including Alithya, Apex IT, Grant Thornton, Huron, IBM Consulting, Infosys, Wipro, Accenture, Deloitte, KPMG, and PwC. Yet any users thinking they can just pick an AI agent off the shelf and deploy need to do some homework. "Oracle would like us to think that there's not much work there, but when you look at data quality and the maturity of data management across customer segments, that essentially defines who can get business value out of AI agents as a technology component," said Abbabatulla. "Ultimately, it is that last mile of inferencing or reasoning which really converts the trained AI to an AI that can deliver meaningful, relevant decisions that you can act on. That's the last mile. It's not the case that every customer is going to be able to do it straight away. Those who have good data management practices, data governance, and have got reasonably good levels of data quality, they should be able to get to the business value journey quicker than others." Kevin Dattolico, Americas regional CEO for Syntax, a provider of Oracle professional and managed services, said customers were also struggling to know where to start. "A lot of customers are slightly overwhelmed," he said. "They've started to go and try to figure out their AI strategy and what they're quickly realizing is that they need a data and data governance strategy in order for it to all make sense. They step back and say, 'Where is our data housed? And what are we going to do?'" To this end, Oracle deserved praise for allowing users to bring its database to be housed with hyperscaler datacenters on its OCI hardware, the so-called "database@" offering available for Google Cloud, Azure, and AWS. Dattolico said over the last few years, Oracle had become the open partner. "They've partnered with all the different LLM providers, and that openness is only for an Oracle solution, but also the rest of the data repositories that are out there. Then also they're partnering and tying into the different hyperscalers as well. Customers who have started to go down a path developing these new applications [in a cloud provider] now they want to be able to bring the data closer to those applications. Oracle have put OCI, in essence, to run within the same confines as AWS, Azure and GCP, which then allows the value of the OCI platform to be realized within those hyperscalers. Typically, customers would have to go through a significant cost migration [to do that]." Dattolico said that openness contributes to reducing cost, and increases agility and flexibility. "In today's world, that's some of the most important things. In the change in dynamic - from political, economical, and everything else - our customers struggle with being agile enough to adapt." Patrick Pugh, PwC's global alliance leader, said customers were looking for orchestration to bring technology from a range of vendors across the technology stack. "Most clients are going multi-tech across their platform. The key is, how do all of these tech companies set up an environment and a structure where agents and people can work across the platform in seamless manners? It's highly competitive, but the tech players, including Oracle, have realized that we live in a multi-tech environment, so you can see the architectures, the intentions and their collaboration doing some doing work that's best for the customers." One way Oracle has diverged from the market for application vendors is to bring AI agents to legacy systems. While rivals Workday and Salesforce were born in the cloud, SAP has said that innovations such as its Joule agent system will only be available on its latest software - S/4HANA - in the cloud, with a large chunk of customers still using its legacy platform, ECC. Oracle has said users of its legacy E-Business Suite (EBS) can take advantage of AI agents. Gartner's Abbabatulla said EBS and PeopleSoft - the widely used HR and financial system Oracle acquired in 2005 - could connect with the LLM agent platform via OCI. "You can do all that stuff. There's also the studio which I can use to build agents on top. It just means the users have got additional bits on top of the technology and the tooling layer that's built on OCI. However customers will be limited by the age of the application," he said. "That doesn't change." Oracle has also offered longer support deadlines than its rivals. EBS and PeopleSoft are supported until 2036, whereas SAP ends mainstream support for ECC in 2027, with extended support ending in 2030. "Oracle's approach is really more the carrot approach than the stick, to say that these are good things you can get from AI," he said. "They've got tools that would help customers to migrate in a manner that makes sense for them, by taking certain data and functions in a phased manner, instead of dumping everything onto SaaS in one go." ®

Oracle wants to bring AI to your data no matter where you work - and that could be exactly what we need for AI to succeed

Oracle looks into the future and sees support ending (Image credit: Future) At its annual AI World conference in Las Vegas (formerly Cloud World), Oracle has lifted the wraps off three separate projects in collaboration with key hyperscalers in order to better serve customers. The updates coincide with the general availability of Oracle Autonomous AI Lakehouse - the company's new all-in-one platform which combines the scalability and flexibility of a data lake with the management and performance benefits of a data warehouse. Oracle Autonomous AI Lakehouse brings AI and machine learning directly into the database, which means customers can use AI where their data lives instead of having to duplicate it and move it elsewhere. With the hyperscaler collaboration, it means that customers can now bring Oracle's AI to their data even if it's stored in Google Cloud, Azure or AWS, keeping them aligned with data residency and compliance requirements. Besides offering Oracle Autonomous AI Lakehouse with these partners, Google Cloud has also brought the number of regions up to eight, from five. Over the course of the next 12 months, Google Cloud has committed to adding a further nine regions, Azure another five, and AWS adding a whole 20 new regions for Oracle customers. Speaking with TechRadar Pro at the event, Oracle VP of Product Management Nathan Thomas emphasized the company's commitment to letting customers place Oracle databases in the cloud of their choice, be that on Oracle Cloud Infrastructure or not. Oracle's Database@ products are designed to bring low-latency, compliant multicloud deployments to where customers are already running their applications, reducing technical hurdles. "Our customers want flexibility... we recognize that and we offer a multicloud environment," Thomas noted. Oracle Applications VP Rajan Krishnan also shared thoughts on the company's multicloud strategy with TechRadar Pro at the event, recognizing that customers today pursue not only multicloud, but also multi-AI options where they can bring together the power of different models from different providers. Krishnan credited Anthropic's work on the Model Context Protocol (MCP) for enabling broader interoperability, likening it to HTTP's standardization of the web. In short, the message at AI World 2025 has been clear in that the future of data is open, flexible and collaborative, and Oracle's partnerships with companies like these is just proof that they can co-exist in the same space, maintaining rivalry and their own differentiators while also better serving customers' needs.

Oracle AI World 25 - AI adoption and project realities - Oracle customers share their field stories

In my just-released Oracle AI World podcast review with Brian Sommer, we discuss how Oracle is ahead of its customers on agentic AI. Perhaps this is due to Oracle's execution across the stack - but the problem remains. Or is it a problem? I would say it's a problem: I posed a version of this question to Steve Miranda at his media/analyst Q/A. I asked Miranda: what is the difference between customers who are aggressively adopting Oracle AI, and those who are stuck, or moving slowly? Is it a matter of leadership, change management, data quality, or something else? And what can Oracle do to help move those customers forward/get them unstuck, e.g. facilitating peer/industry discussions. Miranda responded by saying that "metadata readiness" is a key: You've got some organizations where the process to get through the AI compliance, AI security ethics, the IT group, the security group is difficult and slow. I think eventually that will quickly turn into not data readiness, but let's say metadata readiness. I'll try to give a quick example: if you wanted to run an Oracle P&L for Fusion, you probably couldn't do it in the English language, not because the agent isn't there - the AI agent is there - but because our finance team knows Fusion as cost center 683. No one put down the name, or in a lot of cases, there wasn't precision on the different descriptors of things. Many companies [haven't done that], because there's never been a need to do that. Agentic interactions with users changes data needs. Miranda added: Finance knew what the cost centers were they were doing, or supply chain knew certain terms or codes, etc. You never interacted with these systems through an English language, or whatever language you choose - a natural language question and answer. So right now, there's a little bit of uncertainty. But, Miranda also believes that Oracle gives customers a leg up on adoption, by embedding AI in Fusion Applications, without pricing changes: Now, as far as what we're doing, I tried to make a point during the demos. I try not to just show you a demo; I try to make the point that these [demo scenarios] are embedded. You don't have to set up security; you don't have to set up the UI. You create the agent, or you modify the agent, and you publish it, and it's there via ask if you ask for it - these are very quick to turn on. And I think right now, there's a certain apprehension level that's hard for me to articulate why. There's a lot of confusion, a lot of fear, a lot of different concerns about AI, but I don't think that's a long term problem. With that in mind, let's look at two customers we spoke with in-depth at the show. Their stories are not necessarily reflective of all Oracle customers, but let's compare/contrast. Start with Choctaw Nation, the third largest American Indian population in the United States (Brian Sommer and I spoke with a team from Oklahoma; since each Choctaw Nation makes independent software/operational decisions, this piece is about the Oklahoma-based tribal organization). One thing both customers we spoke to had in common: whatever they do with applications - or AI - it must adhere to complex regulations. Emily Crow, Director, IT Enterprises Services, Choctaw Nation of Oklahoma explained: Our regulatory requirements are fascinating, just because there are so many that we have to meet, mostly because of the diversity of the industries that we work in. Because we run all of our casinos, we have the National Indian Gaming Commission - that has their own set of regulations. But we also run our healthcare system, so we have HIPAA, and all of those regulations that we have to follow. We have federal grant funding for several of our programs. Each federal agency that gives us money has regulations on how we account for the money, how we hire people, all of that. Scale is a must: the Choctaw Nation supports more than 250,000 tribal members and manages a number of casinos, resort hotels, restaurants, 32 medical facilities, and over 60,000 acres of ranching and agriculture. More than 13,000 full-time employees drive almost 150 programs that provide education, healthcare, housing, and other essential services. The Choctaw Nation now runs on Oracle Fusion Applications, including Finance and HR. I asked Crow: how well has Oracle handled these compliance needs? Are there gaps or manual reporting needs? Crow: So far, out of the box Is working for all of it. I don't know that we've loaded anything separate into the system specific to those. Early adoption of generative AI is part of this story: Choctaw Nation has already adopted 40+ generative AI capabilities, and they are looking towards Oracle's AI Agent Studio as well. We asked the Choctaw Nation team: of the gen AI features adopted so far, which have had the biggest impact? Both were in HR. Crow: 'Performance review AI assist' - the usage of that was phenomenal, and in Goal Creation as well - we saw a huge uptick in goal creation. But hold up - were there any concerns over the AI perhaps writing performance reviews that missed a crucial issue, or a subjective area? No one really worried about the data aspect of it. If anything, there were some employees who commented that they felt like a review created by AI would feel too impersonal. And I could definitely see that, which is why one of the things I want to enhance is sort of some sort of guidance for people, to drive the acknowledgement that you do need to refine what's in there. Make it your voice. But I also think that the more data each manager puts in it, the better that generative AI will be, and it might become more of your voice over time. I did force it on my team. I hit the 'AI assist,' and just hit submit: 'You don't like your review. You're in charge of the Oracle product. Go for it. Go fix it.' It was actually pretty spot on. It was pretty good, actually; I was impressed. Reading Oracle's own writeup on the Choctaw Nation project, I was struck by the HR scenarios. After all, if Choctaw Nation can't support meaningful work - and professional growth - for its members, then Fusion Apps is falling short for them. But Crow and her colleague Kamwin Story told us that their moving to Dynamic Skills and Oracle Grow has indeed helped them to support these goals. Crow: One of the big focuses for us is not just employee growth, but retention. We don't have a huge population in southeast Oklahoma to pull from for our employee base. So keeping our employees that are successful, retaining them in the organization, and growing them up into leadership and executive leadership positions is vitally important for us for our sustainability - but we also focus on the tribal members themselves. Our chief is big on 'give you a hand up, not a handout.' We don't just mail checks to our members like a lot of the tribes do. We help them out with assistance, and we want to make them self-sustaining. One of the ways we do that is by focusing on employing them as as our primary employee base. [Oracle] Grow and skills are definitely a piece of that strategy, to help identify where they need to improve what job they want to move into. So many of our employees start in the casino or in our hospitality wing. They might be front desk hotel employees. They might be a waitress, whatever it might be. It's your foot in the door to the Choctaw Nation. But if they want to be an attorney one day; if they want to be a nurse; if they want to study drone technology; if they whatever it might be, we're doing it in some part of our organization, and so they have a way to use that software to say, 'Okay, if I'm if I'm here and I'm a waitress, what do I need to do to get to that level?' And an Oracle can help put those pieces together in a way that their manager just really isn't capable of doing, because they don't know that part of the business. Though the impact of Choctaw Nation's skills/AI initiatives are still early days, the adoption levels are high. As Kamwin Story, IT Senior Manager, Choctaw Nation of Oklahoma, told us, *** list adoption stats here. Another crucial area of this project? Choctaw language translation - and preservation. As of now, there are only a few hundred living Choctaw tribal members that speak Choctaw as their first language. Preserving that language, and using Oracle AI to translate it, is an essential project. There is hard work ahead; the Choctaw language does not translate word for word into English by any means. But Crow is excited about the progress made. The Children's Hospital of Los Angeles (CHLA) is in a different situation. A longtime PeopleSoft customer, their big move to Oracle Fusion Applications is still on the horizon, slated to begin next January. But Oracle is the modernization option of choice. How did Oracle earn this role? First, by helping the CHLA modernize its existing PeopleSoft platform a few years ago, via a move to Oracle Cloud Infrastructure (OCI). Also, CHLA is a Cerner customer. Over the last few years, they assessed Oracle's commitment to integrating (and enhancing) the Cerner health care solutions - and they liked what they saw. One thing the CLHA does have in common with Choctaw Nation: intense regulatory requirements. In turn, this resulted in a heavily customized PeopleSoft instance - something the CHLA team is eager to transition from. During our meeting with CHLA, they laid it out: We have a lot of complexity we have to deal with, from a practice standpoint, being a California Hospital with all of those regulations and compliance. We've made so many work arounds there that it's highly customized, challenging, and everything takes months - even something simple. Obviously, these compliance requirements will not go away when CHLA moves to Fusion Applications - not to mention agentic AI, which will likely involve Oracle's vector database as well. As CHLA told us, it's the strength of their partnership with Oracle that makes this viable: Well, I think it's going to be a partnership with Oracle to do that. We are talking with them about how we can get into a plan to adopt the new EHR, which is when they introduce that vector and that semantic layer, and when they're going to be providing those automated generative summaries, and all of that. But Oracle doesn't have a pediatric client like us that can truly validate it to the extent needed in our ecosystem, so it's going to be a partnership with them. We're having great discussions with them around that today. The CHLA team knows that before they can put Oracle's agentic AI to the test, they need to make this Fusion Applications move, and push off of their older PeopleSoft release. So that invokes the question: what is their take on AI? Are they wary - or ready to move ahead? From an adoption standpoint, CHLA is ready. They already have staff who have used different AI tools, including outside of work. Some of the physicians and clinicians are using 'visit summary' AI tools. So, with Oracle on deck, what is the AI mood? There will be some change management when it comes to interacting with the tools in that regard, but we've got people kind of chomping at the bit at this. Of course, CHLA will have no tolerance for "hallucinations," so there will be rigorous testing before agent activations. And, as they told us, they will need to have humans-in-the-AI-loop for the foreseeable. In terms of whether Oracle came off as relevant or tone deaf, CHLA was the perfect one to ask. During my Larry Ellison keynote watch party appearance with the CRM Playaz, we reviewed Ellison's one man monologue from an analyst perspective, in real-time. But what did the CHLA team think? It's nice to hear someone of Larry's stature describing the things that we've been experiencing... To have him have a level of understanding - take the payer example, and just what a mess that is, and have him dig in on that, and understand how complex things are... That's the stuff that we live every day. The ecosystem he talked about, and the applications of it - like healthcare today - is broken because of all of these different systems, and the lack of being able to communicate together. What we have to do, to be able to communicate with our payers, is insane, really challenging, and that just adds administrative overhead, which rises costs, which just continues to be challenging. So the vision he has is the right one. CHLA knows they have some big change management work ahead - this is no small move. So far, the leadership is on board, as well as operations leads. As they move users off of PeopleSoft next year, that buy-in will be crucial. They spoke of one project event with an operations leadership team. At one point in the event, a leader who has been known to be a bit cynical or resistant, said he was fully on board as well. As CHLA told us, that was a big moment. On the show floor, I spoke with other customers to get a sense of the five questions I started this piece with. Obviously, most Oracle customers are not at the point of speaking to agentic AI project results. Generative AI at scale? That's starting to come in a bit, as per Choctaw Nation's AI Assist successes. But customers seem to like where Oracle is headed here. Perhaps daunted by the prospect of undertaking agentic AI on their own, operating AI in an Oracle context seems to resonate better. As my back and forth with Miranda indicates, that doesn't mean the issue of AI adoption is solved. There is obviously more that Oracle can do. A big part of it is customers sharing adoption lessons with their peers. Too often, vendors get in the middle of these peer interactions with product announcements, instead of facilitating the conversation customers need to have. I prefer to see vendors put customers together and step back, ready to answer roadmap questions where needed, but with the sales cap off. I'd like to see vendors like Oracle develop interactive AI maturity models by industry, highlighting the gotchas and ROI opportunities of each phase. But if you believe, as I do, that data quality and platform modernization are crucial components of any type of truly successful AI, Oracle is well positioned on that. To hear customers say that Ellison's keynote resonated with them has to be encouraging for Oracle. While that was a provocative keynote, it was not casual viewing; it required real engagement. Ellison wants to fix healthcare; an organization like CHLA wants to deliver a better service to its customers, while avoiding the burnout grind that is fraying healthcare providers from within. If agentic AI from Oracle can help make that happen, customers like CHLA become Oracle's best AI advocates. I had my differences with Ellison's takes, particularly on AI, some of which I've aired out on these pages. At next year's AI World, more agentic proof points will be needed - with customer results at scale, and candid talks about where agentic processes went wrong, and how they were improved/enhanced/brought to heel. Oracle's agentic evaluation tools will help greatly here, but it's still up to Oracle to air out those conversations. For now, the customer buy-in seems to be there.

Oracle's Secret to Building AI Superclusters

Oracle wants to package and put their cloud close to where customers are. "We want to take the cloud to customers, not the other way around," were the words of Pradeep Vincent, senior VP and chief technical architect for Oracle Cloud Infrastructure (OCI), at the Oracle AI World 2025, held from October 13-16 in Las Vegas, US. On the sidelines of the event, Vincent sat down with AIM to discuss the company's evolving cloud strategy, its AI infrastructure ambitions, and India's positioning and probable gains in the next wave of AI adoption. Having witnessed cloud computing evolve over two decades, Vincent described this moment as the most transformative in his career. "This is by far the period of the highest change rate," he said, adding, "it's the most spectacular phase of technology evolution." Oracle's engineering teams continue to ada

Oracle AI World 25 - the AI hot topics and under-the-radar issues you should be watching

Every enterprise AI vendor is flat-out determined to differentiate. Why? Like most enterprise AI vendors worth their billing, Oracle is building specialized agents for better results. How? By utilizing so-called "context engineering" to bring customer data to bear privately and securely (Oracle now uses a range of third party LLMs as the "planning"/orchestration part of those agentic workflows, in an agnostic manner - a notable change, or perhaps evolution is the better word). Oracle has another differentiated aspect - one that featured prominently in this week's AI news announcements. As an infrastructure and apps player, Oracle can provide full AI stack services to customers and partners. Oracle believes it can build the most compelling AI ecosystem around this stack. Time will tell on that - but if you want to dig in, there are plenty of keynote replays that lay out Oracle's AI strategy (also see: my colleague Derek du Preez's keynote review: Ellison's AI vision for Oracle - robot surgeons meet enterprise data reality. On the ground in Las Vegas, I had the chance to probe into this with Oracle leaders and AI experts. I surfaced five differentiating themes that didn't grab the headlines, but I think are important. Yes, the best enterprise AI players all have a strong/secure context engineering strategy, but most don't have the following. The smart AI vendors know that customers want to control their own pace with agentic autonomy, as per their industry regulations, culture, and accuracy tolerance per scenario. Rohit Gupta, CEO of Workday partner Auditoria, calls this an "autonomy toggle." During on Monday analyst session, Steve Miranda, EVP, Oracle Applications Product Development, showed us a different approach via Oracle's AI Agent Studio. As customers (or partners) use the Agent Studio, they can graphically insert human approval steps into autonomous workflows, or remove them when/if it makes sense. Here's how Miranda explained it during a purchase requisition demo: If I get a quote from a supplier, I want to automatically create a purchase order. And you'll see that the agent has many, multiple steps. But in this case, what I'm going to do is I'm going to modify the status to put what's called "human in the middle." I don't want the purchase requisition to be created automatically. I want there to be a human approval. So I just go into the agent steps... I want my app to have a human in the middle, and the preceding steps are also saved there, but [now I have] email verification, because I have a human in the middle - I'm going to go ahead and pick that up. Later, during my AI deep dive with Kaushal Kurapati, Oracle's Head of Agent Platform, he showed me a similar scenario, where the user could add a "human approval node" into the Oracle workflow agent. This is not something in the chat interface; instead, it's built into the underlying workflow. Kurapati explained: ln this case, the code is particular to that agent, but it basically says, 'Do you want to proceed to create the requisition?' It's asking, 'Wait for a human being.' Kurapati confirmed that you can add rules, along the lines of 'only ask for human approval for requisitions above $1,000', to nudge these agentic workflows closer to deterministic. I ask just every enterprise vendor about their agentic evaluation and observability tools. The answers are almost universally unsatisfying. At best, I get 'We make sure to provide the explainability the customer needs.' But explainability and agentic evaluation are not the same. If you want to truly earn customer trust, you need to give them visibility into how agents break down. Without that visibility, how can you iterate and improve AI processes? In Wednesday's AI news, Oracle announced it's evaluation framework for customers. This includes: During our talk, Kurapati added: Explaining what's going on behind the scenes, opening under the hood tells customers and partners, 'Oh, I know what's going on, where the pinpointed errors are, or what's working even, and how many tokens are being consumed, how much time it's taking,' - all of that so that you can optimize your agent, I think that's the key. Enterprise vendors have AI pricing challenges - consumption, value, all-inclusive, add-on? Plus: balancing investor and customer expectations. I get why so many vendors struggle with convoluted AI pricing schemes. LLM pricing has not stabilized, and advanced inference techniques like "reasoning" add to those costs. Oracle, however, sounds about as confident on AI pricing as any vendor I've heard from. During our analyst session, Miranda didn't hold back: There is not an additional charge for our AI agents. These are embedded agents throughout the business process, which are included in the fees. That's something we conferred about... We're not going to charge a customer for a best-in-class general ledger. When we improve that with AI, with a smart GL, that's [not] an extra cost. No - we're not going to do that. We haven't done that. Just like all the technology we've delivered to our customers in an SaaS model consistently over the years. At the same subscription, you get the same solution. You get better and improved AI agents. Given we're now talking about 600 AI agents (400 inside of Fusion apps, 200 in industry solutions), that sounds to me like a pricing policy customers will welcome. However, like almost all enterprise vendors, Oracle does have premium AI features that come at a premium price. During a CX analyst session, Chris Leone, Chris Leone, EVP, Oracle Applications Development, provided more details. Oracle has what they call a "Premium LLM" offering that Oracle optimizes the prompts for, which currently allows 200 million tokens of consumption per month. Or, you can use Oracle's backup agent model, OpenAI's "open source" (ahem) GPT OSS model that Oracle hosts in OCI, for unlimited tokens. Leone: "So you really could use all of our delivered agents... all of that is no cost." Leone says the additional charges come in if you extend one of those models by adding another tool or agent. Now you're using custom AI components, or building your own agents. How does that work? Leone: We have a user based pricing, user times agent, or an employee based pricing, which is employee times agent. We want to be more aligned with our pricing model. We'll see if it works. But we want to have deployment. It's rare to see a vendor emphasize deployment over premium AI pricing. Oracle says it can do this because of how it controls AI infrastructure costs internally (One of my video sparring partners, Constellation's Holger Mueller, believes the ease of Oracle moving data into its vector database accounts for some of this cost control). Some of my analyst colleagues are skeptical about these so-called cost controls, but for whatever the reason, Oracle seems to have a more welcoming (and easier) pricing model than most of its peers. [Disclaimer: AI pricing is a moving target for all vendors, by the time you read this, it's possible goal posts have moved in some way, or I've missed an exception to the guidelines here. Oracle should be contacted directly for any up to date pricing confirmations for material evaluations]. One of the biggest things I hear from customers: they want to their data - and their organizations - to be "AI ready." This is a monster topic. It obviously starts with quality data, but it doesn't end there. One of my research directions for this year: a precise understanding on what types of structured (and unstructured) data are most LLM friendly. Real-time? Lakehouse? Edge or centralized? Zero copy? That's just the beginning of this geeky discussion. But at Oracle AI World, the focus sharpened: even with quality enterprise data, LLMs need "annotations." During his analyst day briefing, Juan Loaiza, Oracle EVP of Mission Critical Database Technologies, explained what he meant by annotations for LLMs. We have to explain the data to AI. A lot of times, this data can go in tables. It might say 'sales 'or something. But what does that mean? Does it mean that was the list price? Does it mean that was the after-discount price, or after taxes? What currency is that in, all that kind of stuff? It's not obvious by just looking at the table, what the heck is [this table] talking about?... This is one of the big things we're working on with Fusion Apps, which is: you can tell the AI 'Go find stuff,' and it can find things, but it's not quite sure what it means. So we're happy to go back and annotate the data. What are annotations exactly? You could start with metadata, or perhaps master data. The key: the annotations are telling the LLMs something about the data in the relevant document/database etc. During my talk with Kurapati, he elaborated, with talk of API graphs and metadata. We have all these API is the drive building of Fusion Applications, right? And we access those APIs through what are called Business Objects. And now, we want our agents to be able to use the same business objects, to do the functions like tool calling. Now, if there was no metadata on that API, what is that API useful for? That metadata being there, and because it's there in plain language, someone can describe it. Now the agent knows exactly what tool to call for doing particular function. So what we've really learned is that the richer that metadata is around all those API's, [the better]. In fact, there is a knowledge graph aspect there too,; there's an API graph, is how I think about it. This API is connected to that API. If you're calling this API, then subsequently, you probably want to call this other API too, because they're very tightly coupled together. And if you have an API graph and all the metadata associated, that's when an agent becomes much more intelligent and capable. I asked Kurapati: can LLMs help to compile this metadata? To an increasing extent, yes, but the metadata will likely need some level of human review/verification as well. It's late, and I have one more bustling day of Oracle AI World to go, so I'll stop here. These under-reported areas reflect where Oracle has some of its biggest - and most compelling - agentic AI advantages (aside from infrastructure and full stack AI; that's a different discussion). I don't want to give the impression that I don't have criticisms of Oracle's AI strategy. I do, but those criticisms are less about the overall emphasis. Example: while I agree with Ellison about context engineering, I don't agree with his contention that "reasoning" is such a breakthrough. Ellison seems to think LLMs (and massive data centers) are the cornerstones of our AI future. I hope not; I believe we'll see breakthroughs in the years to come that allow for more energy efficient types of AI, with better world models, causality, and so on (after all, the human brain does a heck of a lot, on a lot less wattage). Credit to Ellison for trying to solve truly vexing real world problems with AI, but I'd like to think some of these will be solved by streamlined AI - the type of AI that doesn't force strained energy grid costs onto consumers' utility bills. As for reasoning, there are stacks of studies about this, but I perceive reasoning as a useful approach for some scenarios, rather than a massive advancement (nor is it cheap at inference time). LLM "reasoning" tends to work better on its own internal data than on "out of distribution" data. For enterprises, I believe the big step forward is context engineering in general, as well as smaller/specialized models. What we are learning about "compound" LLM architectures, with non-LLM components providing verification, grounding and tool enhancements, are, in my view, more important for than pure scaling, which is currently slowing in its payoffs (see: the disappointing GPT-5). But that's the kind of thing Oracle will put to the test internally. Whatever ends up working the best, who knows. I may be right, wrong, or somewhere in between. Oracle's massive AI investments will reveal more than what we know right now. I hope that Ellison doubles down on using AI to solve deep/crucial industry problems, healthcare being a prime example. That would drive home what was, without doubt, one of the most audacious - and most intellectually substantial - keynotes of the year to date. And it was just one guy sitting in a chair. If that's not a wake-up call to enterprise marketers, I don't know what is. Onward...

Analysis: How Oracle Is Differentiating For The Agentic AI Era In Quest For $225 Billion

'Today, we are faced with a once-in-a-generation moment where AI changes everything,' Oracle co-CEO Mike Sicilia says. Greater hardware flexibility, deepening multicloud partnerships, allowing enterprises to securely leverage private data for artificial intelligence and leaning into AI agents are some of the ways Oracle, its revamped leadership team and partners are taking on the AI era. The Austin, Texas-based cloud and database products vendor will pursue this strategy as it chases a lofty goal of $225 billion in consolidated fiscal year 2030 revenues, as disclosed at the vendor's AI World conference that recently wrapped in Las Vegas. Reaching that goal would mean a 31 percent compound annual growth rate (CAGR). "Today, we are faced with a once-in-a-generation moment where AI changes everything," one of the nearly 50-year-old company's two new co-CEOs, Mike Sicilia, said in his AI World keynote address. "All of our collective innovations along the way serve as the foundation for our AI platform going forward." [RELATED: Oracle Launches Multi-Cloud Services Reseller Program, Universal Credits] The vendor also said it crossed a remaining performance obligation (RPO) of $500 billion, as promised to analysts on the vendor's September earnings call. Oracle added about $65 billion in total contract value (TCV) in about 30 days for OCI, Morgan Stanley said in a report Friday. Scott Whitley, chief revenue officer of Troy, Mich.-based Oracle solution provider Centroid Systems, told CRN in a recent interview that business is booming with the database giant. The vendor's massive backlog of AI business has served as a tailoring for the solution provider. "I've been working with Oracle for the past decade. And I have just never seen something like this," Whitely said. "I know that's the craze of GPUs (graphics processing units), but even older versions are just getting gobbled up. It's a really fun time to be around it. It's challenging because it changes every day. But I think that's everybody's company right now." The vendor's reveal in September that it had $455 billion in backlog came with criticism that the money comes from contracts secured with OpenAI, Facebook parent Meta and Elon Musk's xAI. The incremental $65 billion TCV comes from seven deals across four customers. Some of that is from Meta. OpenAI contributed no incremental value, which could better assure analysts around Oracle's backlog stability, according to Morgan Stanley. But Oracle co-founder and Chief Technology Officer Larry Ellison dismissed concerns around an AI bubble proving detrimental to the technology in the long run, recalling the early 2000s dot-com crash when naysayers didn't separate the work companies like PayPal did with the work a Pets.com did. "If I can sell pet food in an e-commerce site, that suddenly means I'm an internet company-not really," he said. "So yes, there'll be people spending money on AI because almost every tech company these days calls themselves an AI company. But they're not. A lot of them are not. But AI in terms of its value, this is the highest value technology we have ever seen by far." Ellison called AI training "the largest, fastest growing business in human history-bigger than the railroads, bigger than the Industrial Revolution." "It is a whole new world that is dawning," he said. Here's more on how Oracle sees its place in the agentic AI future and what the channel can take away from the trends the company's executives see in the infrastructure, applications and database markets. Along with the $225 billion consolidated revenue goal Oracle revealed at its conference, the vendor increased its expected Oracle Cloud Infrastructure (OCI) revenues going into the 2030 fiscal year-although its $18 billion for the 2026 fiscal year revenue is unchanged. The $32 billion for 2027 is now $34 billion. The $73 billion for 2028 is now $77 billion. The $114 billion for 2029 is now $129 billion. And the $144 billion for 2030 is now $166 billion. The new numbers reflect a 75 percent CAGR. OCI is on its way to around three-fourths of Oracle revenue compared to 50 percent today, Bank of America said in a report Friday. One of its new co-CEOs, infrastructure-focused Clay Magouyrk, said in his AI World keynote that Oracle's focus on hardware flexibility as now paying off in the AI era when OCI is optimized for a variety of hardware accelerators, investment in disintermediation to drive down network fees for customers, work with Microsoft and Google to charge zero egress fees for multicloud customers and other differentiators. "We're constantly focused on living up to our commitment to be the highest performance, lowest cost and most secure infrastructure possible," he said. "And we get closer to that ideal every single day." Oracle's Abilene, Texas, data center project under development will eventually consume 1.2 billion watts, enough to power 1 million four-bedroom homes in the U.S., and have a cluster with more than 450,000 Nvidia graphics processing units (GPUs) when fully provisioned, Ellison said. "That's a long way from writing code in my bedroom in college," he quipped from the stage. Oracle will make its Zettascale10 AI supercomputer cluster underpinning the Stargate supercluster Oracle is working on with OpenAI available in the second half of the 2026 calendar year. It will scale to 800,000 Nvidia graphics processing units (GPUs) and use Oracle's Acceleron networking architecture. Acceleron should help improve users' infrastructure experiences through host accelerators, fabric architectures and fabric accelerators, according to KeyBanc's report Wednesday. Oracle has also shed some more light on growth and economics of non-AI infrastructure-as-a-service (IaaS) OCI components. Oracle saw 77 percent growth year on year in annual contract revenue (ACR) in the first quarter of fiscal year 2026 in distributed cloud, according to Morgan Stanley's report Friday. This business sees $67 million in average deal sizes and gross margins of up to 60 percent, according to the firm. Oracle saw more than 40 percent growth year on year in the first quarter among cloud native customers, with $97 million average deal size and gross margins of up to 60 percent. In enterprise, Oracle saw 33 percent growth year on year. The vendor's AI IaaS now has about 700 customers on the platform, with revenue more than double year on year, according to Morgan Stanley's report. Oracle did not share details on the customer mix, duration and whether the contracts are for AI training, inference or other purposes. It expects gross margins of up to 40 percent. Oracle has also further broken down the economics for big data center deals, estimating that land, data center and power represent about 35 percent of cost outlay and compute, networking and storage represent the rest, using a 1 gigawatt (GW) $60 billion total contract value (TCV) deal as an example. The vendor expects about $39 billion of costs for a GW of data center capacity, with $10 billion in revenue for six years, but it did not provide its capital expenditure (CapEx) outlook in the years ahead, according to Morgan Stanley. Part of the reason for not sharing a forecast was how fast and how much the company sees changes in the mix and type of GPUs, the amount of storage and general purpose compute and other underlying variables for cost. The capital expenditures for 1 GW of AI infrastructure capacity should be around $25 billion, according to a William Blair report Friday. That GW should generate $10 billion in consumption revenue a year, reaching $100 billion by fiscal year 2030. Oracle might need 10 GW of capacity to meet its revenue target, costing about $250 billion in CapEx. Oracle would get a $49 billion free cash shortfall that debt could finance to the tune of $2.9 billion in additional interest expenses a year with a 6 percent rate. William Blair expects operating margins to fall to 33 percent in fiscal year 2028 and stay steady into the following FY, potentially starting to improve in 2030. Oracle's other co-CEO, Mike Sicilia, who leads the applications side of the business, said that Oracle is helping customers unlock AI use cases with the structured and unstructured data they've long stored in the vendor's database products. He said customers are using AI to decrease hiring time, resolve service tickets, more accurately predict cash flow, flag supply chain risks and more. Employees gain more time for more strategic and creative work with once-manual tasks now automated. "You're getting real results with no extra cost and no waiting," he said. Oracle, Magouyrk said, is allowing customers to use secure controlled access to bring AI models to private data without exposing it to the internet to attain the self-describing attributes of public data. Users can also create a shared index of private data and leave it in its system of origin while controlling user access. He pointed to Oracle's GenAI agent platform as a way to integrate tools into AI workflows for retrieval augmented generation (RAG), coding and other agentic tasks. The vendor expects AI database and AI Data Platform revenue to scale from about $2.4 billion in fiscal year 2025 to about $20 billion by fiscal year 2030 with a 53 percent five-year compound annual growth rate. Multicloud scaling is a driving factor, according to Morgan Stanley. The AI Data Platform, which became generally available last week, brings users automated data ingestion, semantic enrichment and vector indexing with a unified view and governance across all data and AI assets. Global system integrators and consultancies have committed more than $1.5 billion in collective investment in the AI Data Platform, including training more than 8,000 practitioners and developing more than 100 industry-specific use cases, according to Oracle. Among those early GSI partners are Accenture, Cognizant, KPMG and PwC. AI agents will also further differentiate Oracle from the pack, with the vendor now offering more than 400 AI agents in its Fusions application suite and more than 200 in vertical products even though its initial target was 100 agents. More than 2,400 customers use AI in the industry apps, according to Morgan Stanley's Friday report. Agentic use cases range from reducing documentation time per patient at clinics, avoiding calls with humans at support centers and reducing financial crime investigations and upselling customers. In more product news at the application and database layer, the vendor said its new Oracle AI Database 26ai has an autonomous AI data lakehouse and AI capabilities for vector search, database management, data development, app development and analytics. 26ai replaces 23ai and is applicable through the October 2025 release update. Oracle has expanded its AI Agent Studio for Fusion Applications platform for building, testing and deploying agents. Part of the expansion is a new AI agent marketplace for Oracle-validated, partner-built AI agents inside Fusion apps. The marketplace features agents built by various Oracle system integrator partners, including 2025 CRN Solution Provider 500 membersAccenture, Wipro, Alithya and Infosys. The AI agents built by system integrators range from one that processes sales orders and automatically adds shipment addresses based on sales and accounts receivables data, one that helps human resources (HR) managers access and update employee data by retrieving data from contracts and one that optimizes negotiations through access to historical purchase order data. The studio gained support for Model Context Protocol (MCP), cards for the Agent2Agent communication standard and controls for allowing agent access to external services without exposing sensitive data through key and authentication token management. The studio has more than 32,000 certified experts trained on its capabilities, according to the vendor. Partners have been positive on Oracle's position in database and applications thanks to its full-stack approach, high performance hardware capability for high-transaction workloads and AI Database's potential to pull more customers into the cloud, according to Morgan Stanley's Friday report. Partners have also seen AI opportunities in human capital management and enterprise performance management. Enterprise resource planning AI products have been seeing a more gradual ramp, on the other hand, according to Morgan Stanley's Friday report. Oracle's recent agentic AI expansion of its ERP and EPM products include payables agents that ingest invoices from emails, portals and documents and apply tax and fraud checks before routing the invoices for approval and payment. A team sync advisor agent submits weekly updates on employee performance progress, challenges and requests. And a quote-to-purchase requisition agent captures suppliers quotes from emails and puts the information in the generated order. Agents added to Oracle Fusion Cloud HCM aim to help recruiters analyze internal and external candidates' backgrounds, skills and interests to surface better job matches and improve insights, according to the vendor. Partners also favorably compared Oracle Fusion with SAP's offers, which could serve Oracle well as back-office software's AI functionality becomes a key decision point for enterprise chief information officers beyond the various merits of company clouds, KeyBanc said in a report Friday. Implementation partners also told the firm that customers are excited for agentic AI but still learning how to use it. These partners have focused on back-office finance and a few other key use cases so far. Although Oracle has been deepening its partnerships with the hyperscalers of Microsoft, Amazon Web Services and Google, Ellison took time in his AI World keynote address to point out that Oracle is more involved in health care applications than its cloud competitors and that Oracle is building scaled applications to automate entire industries, pointing to Oracle's work in health care payment systems as an example. He said that while Oracle and Google are developing AI technology, Microsoft and AWS "may or may not" in a possible reference to Microsoft's close ties to ChatGPT maker OpenAI and AWS' deep ties to Claude maker Anthropic. "Our goals are different than those other clouds," he said. "We're a participant in creating AI technology. And we're also a participant in using that technology to solve problems in different ecosystems." Still, Oracle's increased fortunes from a more hyperscaler friendly approach can't be understated. It saw multicloud consumption revenue grow 16 times year on year. Microsoft Azure drove that multicloud revenue, not a surprise given the longer partnership between both tech titans compared to the newer Google Cloud and AWS partnership, according to Morgan Stanley. Gross margins in the enterprise category can go as high as 80 percent. Solution providers who talked to the investment firm said that they see more customers jumping on multicloud strategies in the quarters ahead. The release of industry product enhancements continued even after AI World's end, with Oracle revealing on Monday that it has enhanced its Oracle Public Safety Suite unified hardware and software platform for law enforcement and first responders with officer-worn cameras and AI-enabled mobile voice controls and insights that aim to improve situational awareness and decision-making, among other advancements. Examples of Oracle going deeper into its partnerships with the hyperscalers include Oracle and Microsoft collaborating on an integration blueprint for organizations to use Oracle Fusion Cloud Supply Chain & Manufacturing (SCM) to improve data-driven decision making and automate processes based on live insights from factory equipment and sensors delivered by Azure IoT Operations and Microsoft Fabric. Oracle also opened a new reseller partner program for solution providers to provide Oracle Database@AWS, Database@Google Cloud and Database@Azure, with participating solution providers including Deloitte, Accenture and Infosys. AI represents an upsell opportunity for Oracle and its partners-customers expanding from one Fusion app to an entire suite can increase their spend by 150 times. About 55 percent of its customer base is in one pillar of the portfolio. Only 2 percent are in an entire suite. Oracle's focus on growing its AI training business will position the vendor for upselling inferencing and platform-as-a-service (PaaS) centered around central processing units (CPUs) and GPUs, according to Bernstein-although training is a lower-margin business for the near term. Bernstein called concerns around Oracle's AI business and profitability "overblown." As for simplifying payment for AI wares, Oracle's new multicloud universal credits licensing model allows for Oracle AI Database and OCI services under one contract across AWS, Microsoft Azure and Google Cloud, opening up more opportunities for partners. Should Oracle make good on the forecast, the company will become the third largest hyperscaler by or before fiscal year 2030, according to a report by Bernstein on Friday. The investment firm contrasted Oracle's strategy of not explicitly charging customers for AI consumption in the software-as-a-service products-instead taking the same credit-based approach as other SaaS vendors-with Oracle rival Salesforce's variety of business models also showcased last week during its Dreamforce 2025 conference. A variety of investment firms pointed out that despite Oracle's revised revenue targets, the vendor still didn't provide discrete gross margin and operating margin targets and the capacity investment needed to meet targets. The vendor appears to face significantly constrained capacity and needs to execute well in the near term to give analysts confidence Oracle can meet its targets, Morgan Stanley said in a report Friday.