That near-death moment produced the most durable enterprise software franchise in history. I find this genuinely surprising. Yet here it is, thriving — because enterprises don't choose infrastructure based on developer sentiment. They choose based on where their data already lives. The simplest way to understand how Oracle makes money: imagine you're a Fortune 500 bank. Your core ledger — the system that processes every transaction, every balance, every regulatory report — runs on Oracle Database. Twenty-seven years of stored procedures, custom integrations, compliance logic, and institutional knowledge are baked into that system. So you don't migrate. Now layer the rest on top. OCI is the exciting part. You just need to win the workloads that require specific performance characteristics. AI training on NVIDIA GPU superclusters? Oracle offers bare-metal access with lower latency than AWS. Database workloads that are already Oracle-native? OCI eliminates the rewrite. Strip out interest expense and the underlying operating economics are closer to 35-40% margins. Cloud and software combined now represent 88% of total revenue. What Oracle is really selling, if you step back, isn't software or cloud or databases. It's the cost of change. And every year, Oracle makes the migration path to its own cloud slightly easier than the migration path to anyone else's. Cloud and software combined represent 88% of total revenue. It's a tacit admission that Oracle can't win the broad cloud envelope, but it can own the data layer within someone else's infrastructure. Whether that's genius or capitulation depends on whether you think the database layer or the cloud platform captures more long-term value. In general-purpose cloud, this contest ended a decade ago. Oracle lost. But AI infrastructure reset the battlefield entirely. Oracle's bare-metal GPU clusters eliminate that overhead. When xAI and OpenAI need capacity and can't get it from their primary providers, they call Oracle. This isn't loyalty or brand preference — it's physics and availability. Both companies sell ERP, finance, supply chain, and HR software to the world's largest organizations. SAP has stronger European penetration and a more modern cloud-native architecture with S/4HANA. That double-migration cost keeps accounts locked for years. Snowflake and Databricks pull analytics workloads away from Oracle's data warehouse. PostgreSQL quietly becomes the default for every new application written by developers under 35. Salesforce owns CRM so completely that Oracle's CX suite barely registers in competitive conversations. Epic fights Cerner in healthcare with deeper clinical workflow expertise. Collectively, they represent a generational shift: new systems are being built without Oracle in the architecture. The honest competitive assessment is this — Oracle is unassailable where it already sits, genuinely competitive in AI infrastructure for as long as supply constraints hold, and largely invisible for net-new developer-led projects. The installed base generates cash. That's $25+ billion flowing in every year from customers who pay because leaving is more expensive than staying. Cloud Infrastructure alone grew north of 50%. Fusion ERP grew 14%, HCM and SCM both 15%. Larry Ellison, at 81, still drives the largest deals personally. They erode unless new workloads keep flowing in. That gap matters less for existing Oracle customers (who'll migrate to OCI regardless) and more for net-new workloads where Oracle has no historical relationship. The debt situation deserves honest acknowledgment. Oracle carries approximately $80-90 billion in long-term obligations — the accumulated cost of PeopleSoft, Sun, NetSuite, and Cerner. Interest expense eats into what would otherwise be spectacular margins. Cerner is the wildcard I'd watch most closely. Banks, hospitals, telecom operators, and government agencies have done the math. Most conclude it's cheaper to stay. It's strengthening because Oracle has finally built a credible cloud migration path. OCI's AI infrastructure play adds a new dimension entirely. Oracle doesn't need developers to love it. It needs enterprises with massive compute budgets to find its GPU clusters faster and cheaper than AWS's waitlist. OpenAI and xAI choosing OCI for training workloads validates this approach. New applications use cloud-native architectures. The gravitational pull only works on systems already in orbit. Java ownership (60 billion+ devices) and the Fusion/NetSuite application suite provide additional defensive layers, but the database franchise remains the core. If Oracle Database becomes optional for new enterprise systems — truly optional, not just theoretically replaceable — the entire economic model changes. That hasn't happened yet. Every stored procedure, every integration, every reporting tool, every compliance validation is built around Oracle's SQL dialect, PL/SQL, and data dictionary structures. Strip away the noise and Oracle has two bets that actually determine its trajectory, plus one long-shot that could become defining. The first bet is OCI as an AI infrastructure platform. This isn't a loyalty play — it's a capacity arbitrage that works as long as GPU demand exceeds supply. This is less glamorous but arguably more valuable long-term. Autonomous Database automates the maintenance that used to require expensive DBAs. Exadata Cloud Service gives performance-sensitive workloads a migration path that doesn't require compromise. The long-shot is healthcare. Then there's the variable nobody models: Larry Ellison is 81. That's not a succession plan. That's a single point of failure wearing a Hawaiian shirt. Bob Miner was the one who actually built the thing. The insight was genuine — IBM's researchers had published papers describing relational database theory and a query language called SQL, but IBM itself hadn't shipped a commercial product. Miner, a quiet mathematician with real engineering discipline, turned that blueprint into working code. Their first real contract came from a government project with a CIA connection — code-named Oracle. The name stuck. The product they shipped in 1979 was labeled Version 2. There was no Version 1. Ellison figured customers would be nervous buying a first release of essential database software, so he simply skipped the number. The early 1980s were a sprint. Relational databases moved from academic curiosity to enterprise necessity as companies realized they needed flexible data access, not just rigid file storage. Unlike IBM's database (which ran only on IBM hardware), Oracle worked across multiple systems. In an era when enterprises were beginning to diversify their computing environments, that flexibility was worth paying for. The 1986 NASDAQ IPO gave Oracle capital and credibility. Ellison was on magazine covers. Then it nearly died. By 1990, Oracle's aggressive sales culture had metastasized into something dangerous. Salespeople were booking revenue on deals that hadn't actually closed. Customers were being sold products that didn't yet exist. The accounting was, charitably, optimistic. In March 1990, Oracle announced it would miss earnings expectations. The stock dropped 31% in a single day. Ellison fired half the sales organization. Jeff Walker, the CFO, departed. Oracle's auditors forced a restatement. What saved Oracle was the database itself. Ellison rebuilt with discipline he hadn't previously shown. He hired Ray Lane as president in 1992 to professionalize sales operations. And he learned that Oracle's real power wasn't in closing new deals — it was in making existing customers unable to leave. The post-crisis Oracle was a different animal. The database franchise generated cash that funded expansion into enterprise applications, middleware, and eventually cloud infrastructure. Each acquisition followed the same logic: buy the customer relationship, then make it expensive to leave. The through-line from 1977 to today isn't technology. It's the commercial insight that data, once stored in a particular system, becomes extraordinarily difficult to move.