In September 2025, Oracle did something it had never done in forty-eight years of existence: it appointed two CEOs who were neither
Larry Ellison nor anyone Larry Ellison had personally recruited from Wall Street. Clay Magouyrk, who had built Oracle's cloud infrastructure from a punchline into a platform hosting OpenAI's training workloads, and Mike Sicilia, who had spent two decades embedding Oracle's application suites into the operational plumbing of hospitals, utilities, and governments — these were engineers, product people, the kind of leaders who speak in API calls and rack-unit density rather than earnings-per-share guidance. The announcement barely registered against the noise of Oracle's fiscal year: a stock that had more than doubled, a market capitalization brushing $920 billion, remaining performance obligations that had exploded past $500 billion, and a $300 billion cloud contract with OpenAI that made the company's previous largest deals look like rounding errors. And yet the structural meaning of that leadership transition — Ellison, at eighty-one, finally handing operational control to people whose expertise was building things rather than buying them or selling them — told you more about what Oracle had become, and what it was betting its next half-century on, than any earnings call could.
The paradox is exquisite. Oracle, the company that Wall Street spent fifteen years writing off as a legacy database vendor destined for irrelevance, the enterprise software dinosaur that missed mobile and was supposedly too late to cloud, had become — by the accident of AI's ravenous demand for infrastructure — the most consequential capital-expenditure story in American technology. In fiscal year 2025, Oracle's cloud infrastructure revenue reached $10.3 billion. Management projected $144 billion by fiscal year 2030. A 14x increase in five years. The number was either the most audacious growth target in enterprise technology history or the kind of projection that precedes a spectacular capital-destruction event. Possibly both.
By the Numbers
The Oracle Machine
$53.0BFY2024 total revenue
$10.3BFY2025 cloud infrastructure revenue
$144BProjected FY2030 cloud infrastructure revenue
~$520BRemaining performance obligations (Q2 FY2026)
~41%Larry Ellison's ownership stake
$920B+Peak market capitalization (Sept 2025)
48 yearsTime from founding to AI inflection
$18.7BFY2024 operating cash flow
The Database and the Ego
To understand how Oracle arrived at this improbable position, you have to understand the man who built it, and the single technical insight that gave him leverage over an entire industry for four decades.
Lawrence Joseph Ellison was born on August 17, 1944, on Manhattan's Lower East Side to an unwed mother who, after he contracted pneumonia at nine months, gave him to her aunt and uncle in Chicago for adoption. He would not meet her again until he was forty-eight. His adoptive father, Louis Ellison — who had taken the surname from Ellis Island, his point of entry to America — had made and lost a small fortune in real estate during the Depression and spent the rest of his life telling Larry he would never amount to anything. It is the kind of origin story that, in retrospect, explains everything about the company that followed: the relentless competitiveness, the pathological need to win, the inability to accept any position other than first.
Ellison dropped out of the University of Illinois after his adoptive mother died, then dropped out of the University of Chicago after a single term. He drifted to California in the mid-1960s, taught himself programming, and bounced between jobs at Amdahl and Ampex. At Ampex he worked on a project for the CIA codenamed — with the kind of unsubtle grandiosity that would become his personal brand — "Oracle."
The intellectual foundation of everything that followed was a twelve-page paper. In June 1970, Edgar F. Codd, a British mathematician working at IBM's San Jose Research Laboratory, published "A Relational Model of Data for Large Shared Data Banks" in the Communications of the ACM. The paper proposed that data could be organized into tables of rows and columns, related by common fields, and queried through a declarative language rather than navigated through the labyrinthine pointer structures that defined existing database systems. IBM, as was its institutional habit, recognized the theoretical elegance and then did almost nothing with it commercially. IBM researchers developed a query language called SEQUEL — Structured English Query Language — and built a prototype called System R. They published the research. They gave the world a map and then declined to walk the territory.
Ellison read Codd's paper. He read the SEQUEL specifications. And he saw, with the clarity of someone who had nothing to lose and no institutional inertia to overcome, that if he could build a commercially viable implementation of a relational database before IBM did, he would own the foundational layer of enterprise computing. In 1977, with $2,000 of his own money — some accounts say $1,200 — he co-founded Software Development Laboratories with Bob Miner and Ed Oates. Miner was the quiet, meticulous engineer who actually wrote most of the early code. Oates contributed the initial architecture. Ellison contributed the vision, the sales ability, and the absolute conviction that they were building the most important piece of software in the world.
Matthew Symonds, in his remarkable biography
Softwar: An Intimate Portrait of Larry Ellison and Oracle, captures the paradox at the company's core: Ellison was "enormously vain, intellectually dominating, and irrepressibly extroverted but he's also shy, has relatively few close friends, and is in constant need of the emotional reassurance that much of his life had been lacking." The company he built would mirror this psychology precisely — simultaneously the most aggressive competitor in enterprise software and the most insecure, always needing to prove, through every acquisition and every product launch and every keynote, that it was not merely good but the best.
In 1979, Relational Software, Inc. — the company had already changed names once and would change again — shipped what it called Oracle V2 to Wright-Patterson Air Force Base. There was no V1; Ellison reasoned that no one would buy a first version of anything. The product was buggy, slow, and, by Ellison's own later admission, somewhat oversold relative to its capabilities. It didn't matter. It was the first commercially available relational database management system. SQL — the language IBM had invented and then failed to commercialize — became the standard interface, and Oracle became the standard platform. By the time IBM finally shipped its own relational database product, DB2, in 1983, Ellison had a four-year head start and a customer base that included the CIA, the Navy, and a growing list of Fortune 500 companies.
The company renamed itself Oracle in 1983, after its flagship product, and went public in 1986.
Selling the Future Before It Arrives
What Ellison understood, and what made Oracle both extraordinarily successful and periodically dangerous to its own shareholders, was that enterprise software was fundamentally a sales problem before it was an engineering problem. The technology had to be good enough. It did not have to be the best. What mattered was convincing the chief information officer of a Fortune 500 company that Oracle's database was the strategic choice — that it was where the industry was going, that compatibility with Oracle was the safe bet, that the alternatives were risky backwaters.
This insight produced a sales culture of legendary aggression. Oracle's sales force in the 1980s and early 1990s operated on a commission structure that rewarded booking revenue with the intensity of a trading floor. Deals were pulled forward. Licenses were sold with payment terms so generous they bordered on vendor financing. The gap between what the database could actually do and what Oracle's sales team promised it could do was, for years, a running joke in enterprise IT — and also the engine of the company's revenue growth.
There's a restlessness about him. Much about Ellison is paradoxical, even contradictory. He is ultra-confrontational in business, but he goes to almost any length to avoid confrontation on a personal level. He either delegates to the point of detachment or is obsessively controlling down to the last detail.
The near-death experience came in 1990. Oracle had been booking revenue on contracts where customers had not yet paid — a practice that, while technically permissible under the accounting rules of the era, created a widening chasm between reported revenue and actual cash. When the economy softened and customers delayed or canceled purchases, Oracle reported its first quarterly loss. The stock fell 80% from its peak. The company laid off 10% of its workforce. Ellison, who had been spending lavishly on homes, yachts, and a lifestyle that made other tech CEOs look monastic, was forced to confront the possibility that he had built a sales machine without a financial foundation.
The recovery was orchestrated by Jeff Henley, a quiet, methodical CFO who imposed financial discipline on a company that had none. Henley, who would later serve as chairman for two decades, was in many ways Ellison's opposite — cautious where Ellison was audacious, process-oriented where Ellison was instinctive, focused on cash flow where Ellison was focused on market share. The partnership worked because each man supplied what the other lacked. Under Henley's stewardship, Oracle shifted from booking revenue aggressively to managing for operating margins and free cash flow. It was the beginning of a financial identity that would persist for three decades: Oracle as a cash-generation machine, a business that prized operating leverage over top-line growth, that bought back stock relentlessly, and that treated its installed base of database customers less as a market to delight than as a base to extract from.
The Acquisitions Arms Dealer
By the early 2000s, Ellison had arrived at a strategic conclusion that would define Oracle's next two decades: organic innovation in enterprise software was overrated. What mattered was owning the stack.
The logic was simple and ruthless. Enterprise customers hated complexity. They ran databases, middleware, applications, and operating systems from different vendors, and the integration costs were staggering. If Oracle could own multiple layers of the enterprise technology stack — not just the database but the applications that ran on it, the middleware that connected them, and eventually the hardware that hosted everything — it could reduce customer switching costs to near zero while capturing a larger share of each customer's IT budget.
The instrument of this strategy was
Safra Catz. Born in Holon, Israel, in 1961, she had trained as a banker at Donaldson, Lufkin & Jenrette before joining Oracle in 1999. Where Ellison was the visionary and the showman, Catz was the operator and the closer — a deal machine who could structure an acquisition, negotiate the terms, integrate the target, and extract the synergies with a precision that made Oracle's M&A program one of the most efficient in corporate history. She would eventually become co-CEO in 2014, sole CEO after Mark Hurd's death in 2019, and the person most responsible for transforming Oracle from a database company into an enterprise conglomerate.
Oracle's major acquisitions, 2004–2017
2005PeopleSoft acquired for $10.3 billion after a hostile eighteen-month battle.
2006Siebel Systems acquired for $5.85 billion, giving Oracle CRM dominance.
2008BEA Systems acquired for $8.5 billion, consolidating middleware.
2010Sun Microsystems acquired for $7.4 billion — Oracle gets Java and hardware.
2012Taleo (talent management) acquired for $1.9 billion.
2016NetSuite acquired for $9.3 billion — the cloud ERP play.
2022Cerner acquired for $28.3 billion — the largest deal in Oracle's history, entering healthcare IT.
The PeopleSoft deal was the one that defined the template. In June 2003, Oracle launched a hostile takeover bid for PeopleSoft, a rival enterprise applications vendor. PeopleSoft's CEO, Craig Conway, called the bid "atrociously bad behavior" and compared Ellison to a stalker. The Department of Justice sued to block the deal on antitrust grounds. PeopleSoft adopted a poison pill. The fight dragged on for eighteen months, through courtroom battles and raised bids, until Oracle finally closed at $10.3 billion in January 2005. Within a year, Oracle had fired most of PeopleSoft's workforce, migrated profitable customers to Oracle products, and used PeopleSoft's installed base as a recurring revenue annuity. The playbook was set: acquire, consolidate, extract.
The Sun Microsystems acquisition in 2010 was more controversial and more revealing. Sun gave Oracle Java — the world's most widely used programming language and the foundation of Android, Hadoop, and half of enterprise computing — along with the SPARC processor architecture, the Solaris operating system, and MySQL, the open-source database that powered much of the internet. Ellison's rationale was vertical integration: Oracle would become, like IBM, a company that sold the complete stack from silicon to software. The hardware business never thrived under Oracle's ownership. But Java gave Ellison something far more valuable: a platform choke point and, eventually, a $billions-dollar lawsuit against Google for using Java APIs in Android. (Oracle lost that case at the Supreme Court in 2021, but the fight itself — nine years of litigation, twice to the highest court — revealed Ellison's instinct for treating intellectual property as a weapon.)
The Cerner acquisition in 2022, at $28.3 billion, was the most expensive and the most strategic. Electronic health records represented one of the last unconsolidated enterprise software markets, and Cerner — despite being the number-two player behind Epic — gave Oracle a massive installed base of hospitals and health systems. Ellison's thesis, articulated repeatedly in keynotes, was that healthcare data was fragmented and unusable, and that Oracle's database and cloud infrastructure could unify patient records in ways that would transform clinical outcomes. The vision was characteristically grand. The execution remained to be proven.
The Cloud Gap
For all of Ellison's strategic acuity, Oracle missed the most important platform shift in enterprise technology since the personal computer. And the man who built the original cloud was one of Ellison's closest friends.
Amazon Web Services launched in 2006. Microsoft Azure followed in 2010. Google Cloud Platform arrived in 2012. By the time Oracle introduced Oracle Cloud Infrastructure (OCI) as a serious product in the late 2010s, the three hyperscalers had a decade's head start, hundreds of billions in cumulative capital expenditure, and an ecosystem of developers, tools, and integrations that made switching costs nearly insurmountable.
Ellison's initial response to cloud computing was dismissal. In 2008, he mocked the concept at a financial analyst conference: "The interesting thing about cloud computing is that we've redefined cloud computing to include everything that we already do." It was vintage Ellison — combative, funny, and wrong. By the time he reversed course and committed Oracle to building genuine cloud infrastructure, AWS had a $25 billion run rate and Microsoft Azure was growing at 50% annually.
The missed window created Oracle's most serious strategic vulnerability since the 1990 near-death experience. Database customers — Oracle's core franchise, the source of its 80%+ gross margins on license support revenue — began migrating workloads to cloud-native databases offered by AWS (Aurora, Redshift), Google (BigQuery, Spanner), and Microsoft (Azure SQL). Each migration was a potential permanent loss of a high-margin maintenance stream that Oracle had depended on for decades.
Oracle's response was twofold: aggressive lock-in through contractual complexity and audit enforcement (Oracle's notoriously aggressive licensing audit program was, for many CIOs, the company's most recognizable feature), and a belated but serious infrastructure buildout. Ellison personally oversaw the architecture of OCI's second-generation design, which abandoned the first generation's approach and started over with a clean-sheet architecture optimized for security isolation and high-performance computing.
The decision to start over — to essentially throw away the first version and rebuild from scratch — was quintessentially Ellisonian. It cost years and billions. But it produced a cloud infrastructure that, by the early 2020s, was competitive on price and performance for specific workloads, particularly database-intensive applications and high-performance computing. It was not AWS. It was not Azure. But it was good enough for Oracle's existing database customers and, crucially, it was architecturally well-suited for the workload that nobody in 2018 saw coming.
The AI Accident
Oracle's emergence as a major AI infrastructure provider is, in the most generous interpretation, a testament to architectural foresight. In the less generous but arguably more accurate interpretation, it is the most fortunate accident in enterprise technology since IBM's decision to use an open architecture for the PC.
When large language model training began consuming unprecedented quantities of GPU compute in 2022 and 2023, the three hyperscalers — AWS, Azure, and Google Cloud — faced a problem: demand for NVIDIA GPU clusters vastly exceeded supply, and their existing data center footprints, optimized for general-purpose cloud workloads, were not ideally configured for the massive, low-latency GPU clusters that AI training required. Oracle Cloud Infrastructure, by contrast, had been designed from its second-generation architecture onward with a network fabric optimized for bare-metal performance and cluster-scale computing. OCI's RDMA (Remote Direct Memory Access) network architecture, originally built for high-performance database workloads, turned out to be nearly ideal for distributed GPU training.
The timing was exquisite. Just as the world's most capital-intensive companies — OpenAI, xAI, Meta — needed more GPU compute than any single hyperscaler could provide, Oracle had underutilized data center capacity and an architecture that could deliver. The constraints that had made OCI a niche player in general cloud computing — smaller overall footprint, fewer managed services, a less mature developer ecosystem — became irrelevant when the customer only cared about GPU density, network throughput, and time-to-deployment.
Clearly, we had an amazing start to the year because Oracle has become the go-to place for AI workloads. We have signed significant cloud contracts with the who's who of AI, including OpenAI, xAI, Meta, Nvidia, AMD and many others.
The OpenAI relationship became the defining deal of Oracle's AI era. In September 2025, OpenAI signed what was reported to be a $300 billion cloud contract with Oracle for the construction of data centers over five years. The number was so large that it defied conventional analysis. Three hundred billion dollars is more than Oracle's entire market capitalization had been just three years earlier. It implied a construction and capital-expenditure program of unprecedented scale — not just servers and networking equipment but power generation, cooling systems, real estate, and the physical infrastructure of a new computing paradigm.
The deal also crystallized the central risk. Oracle was transforming itself from a high-margin, cash-generative software company into a capital-intensive infrastructure builder, with the bulk of its projected future revenue dependent on a single startup — OpenAI — whose own business model, competitive position, and funding trajectory were far from certain. As one portfolio manager at Argent Capital Management put it: "They are stretching their balance sheet about as far as they can, their free cash flow is negative and their balance sheet is highly levered. Their neck is sticking out."
The Governance Paradox
In September 2014, the Harvard Business Review published an article with a headline that captured the central tension of Oracle's corporate identity: "Oracle: The Worst-Governed, Best-Run Company Around." Larry Ellison had just stepped down as CEO after thirty-seven years, except that he hadn't really stepped down at all. He became executive chairman and chief technology officer. The new co-CEOs, Safra Catz and Mark Hurd, "will continue to do the same things they did as co-presidents, the only difference being that they will now report to the board instead of just to Ellison. But Ellison is of course chairman of that board."
The governance structure was, by any conventional corporate metric, indefensible. Ellison owned approximately 25% of the company (later growing to roughly 41% as Oracle repurchased shares and his relative stake increased). He controlled the board. He had no meaningful check on his authority. His compensation was occasionally enormous. The company's related-party transactions — including Oracle's $9.3 billion acquisition of NetSuite, a company in which Ellison personally held a major stake — raised perennial questions about conflicts of interest.
And yet. Oracle, under this theoretically dysfunctional governance, generated operating cash flow of $18.7 billion in fiscal year 2024. It maintained non-GAAP operating margins of 44%. It had grown revenue from $10 billion in 2004 to $53 billion in 2024. It had returned tens of billions to shareholders through buybacks while simultaneously funding one of the most aggressive acquisition programs in technology history. The stock price had appreciated more than 450% in five years.
The resolution of the paradox is uncomfortable for governance theorists: alignment of incentives, rather than independence of oversight, was what made Oracle work. Ellison's 41% stake meant that every dollar of value destroyed cost him personally four hundred million dollars per billion. His interests were, by definition, the shareholders' interests. The governance was terrible in theory and functional in practice because the controlling shareholder was also the strategic architect — and happened to be, whatever his personal eccentricities, a genuinely brilliant technologist and competitive operator.
Mark Hurd's death from cancer in October 2019 — he had resigned as co-CEO just weeks before — removed one leg of the leadership tripod that had governed Oracle for five years. Catz became sole CEO. But the real power dynamic never changed. Ellison, from his estates in Lanai and Malibu, continued to set strategic direction, approve major deals, and drive product architecture. The 2025 appointment of Magouyrk and Sicilia as co-CEOs was, in one reading, the first genuine succession plan. In another reading, it was a reconfiguration of the same structure: two operational executives executing a strategy set by the chairman-CTO who owned 41% of the company.
The Personality as Strategy
Mike Wilson's biography of Ellison carries the subtitle God Doesn't Think He's Larry Ellison. The joke, like most good jokes, contains a structural truth. Ellison's personality — the competitiveness, the grandiosity, the refusal to accept any narrative in which Oracle is not the most important company in any market it enters — is not incidental to Oracle's strategy. It is Oracle's strategy.
Consider the pattern. In the 1980s, Ellison declared Oracle's relational database would replace every other data management system on earth. He was substantially right. In the 1990s, he predicted the "network computer" — a thin client that would access applications over the internet — and was mocked for it. He was roughly twenty years early, but the concept he described is essentially what cloud computing delivered. In the 2000s, he declared that enterprise applications from multiple vendors was an absurd inefficiency and proceeded to spend more than $80 billion on acquisitions to consolidate them. In the 2020s, he declared that AI would be "a much bigger deal than the Industrial Revolution, electricity and everything that's come before" and bet Oracle's balance sheet on infrastructure to support it.
The pattern is consistent: Ellison identifies a structural truth about where enterprise computing is going, articulates it in the most provocative terms possible, and then commits Oracle's resources to that vision with a totality that more cautious leaders would find irresponsible. Sometimes the timing is wrong. The network computer was too early. The first version of OCI was too late. But the directional bet is almost always right, and Ellison's willingness to endure years of mockery while executing against a vision that has not yet been validated is, paradoxically, Oracle's most durable competitive advantage.
We don't want to follow others. We always try to do something differently. Being No. 1 is very, very important. It also means you're the best and solving customers' needs.
Symonds captures this quality — the combination of intellectual seriousness and performative egotism — better than any other chronicler. Ellison, he writes, "desperately wants his wealth to do some good in the world, but he recoils at the very idea of altruism. He is ultra-confrontational in business, but he goes to almost any length to avoid confrontation on a personal level." The company embodies the same contradictions. Oracle is simultaneously the enterprise vendor most feared by competitors and most resented by customers. Its database technology is genuinely excellent. Its licensing practices are genuinely punitive. It builds products of extraordinary sophistication and then sells them with a brutality that makes customers feel captive rather than served.
The Vertical Integrator
The acquisition of Sun Microsystems in 2010 announced a thesis that Ellison had been developing for years: that the future of enterprise computing belonged to vertically integrated stacks, not best-of-breed component vendors. The model he admired was Apple, not the PC ecosystem. Control the hardware, control the operating system, control the database, control the middleware, control the applications. Optimize the whole system. Eliminate the margin and complexity that existed in the gaps between vendors.
Oracle's Engineered Systems — Exadata for databases, Exalogic for middleware, the SPARC SuperCluster — were the physical manifestation of this vision. They were expensive, high-performance machines optimized to run Oracle's software faster than any competitor's hardware could. The strategy worked for a specific segment of the market: the largest enterprises running the most demanding workloads, where performance and reliability justified premium pricing.
But vertical integration is a strategy that compounds on itself. Every layer of the stack that Oracle owned created another switching cost for customers and another reason for competitors to build alternatives. AWS, which had no legacy enterprise software business to protect, offered Aurora (a MySQL and PostgreSQL-compatible database) and Redshift (a data warehouse) specifically to give Oracle database customers an escape route. Microsoft offered Azure SQL Database and the familiar SQL Server as a cloud alternative. Google offered BigQuery and Spanner. The hyperscalers' cloud databases weren't as performant as Oracle's for the most demanding workloads, but they were good enough for 80% of use cases and dramatically simpler to deploy and manage.
The result was a bifurcation. Oracle retained its grip on the most complex, mission-critical workloads — the core banking systems, the large-scale ERP implementations, the government databases where Oracle's reliability and Oracle's support contracts justified Oracle's prices. But the growth market — the new workloads, the startups, the digital-native companies — almost entirely bypassed Oracle for cloud-native alternatives. Oracle's database business remained enormously profitable. It was also, in the language of the industry, a "melting ice cube" — a franchise generating enormous cash flows from an installed base that was gradually, inevitably shrinking.
Nashville and the Geography of Ambition
In December 2020, Oracle announced it was moving its headquarters from Redwood City, California, to Austin, Texas. The move followed Ellison's own relocation to Lanai, Hawaii, during the pandemic and presaged a broader reorientation of Oracle's physical presence that culminated in the announcement of a massive new campus in Nashville, Tennessee.
The Nashville campus — a 2-million-square-foot complex intended as Oracle's "world headquarters" — embodied Ellison's aspirations for the company's next era. It would be a statement of permanence and ambition, the physical manifestation of a company that saw itself as a generational institution rather than a technology startup. It would also, as Fortune reported in January 2026, struggle to attract workers. The challenge of convincing enterprise software engineers to relocate to Nashville — from the Bay Area, from Seattle, from Austin — was a microcosm of the broader tension in Oracle's workforce strategy: a company built on the idea that talent follows opportunity, confronting a labor market where talent increasingly demands that opportunity come to them.
The geography mattered for another reason. Oracle's data center buildout — the infrastructure backbone of its AI ambitions — required locations with abundant power, favorable land costs, and proximity to the transmission grid. The southeastern United States, with its relatively cheap electricity, available industrial land, and business-friendly regulatory environment, was ideal. Nashville was not just a headquarters. It was a bet on the topology of American computing.
The TikTok Sideshow and the Political Machine
In December 2025, TikTok agreed to a U.S. joint venture deal with Oracle, Silver Lake, and MGX, the Abu Dhabi-based technology investment vehicle. The deal, which had been in various stages of negotiation since the Trump administration's first efforts to force a divestiture of TikTok's U.S. operations, gave Oracle custody of TikTok's U.S. user data and positioned the company as the technological guarantor of the platform's compliance with American national security requirements.
The TikTok arrangement was, in strategic terms, a sideshow — it would generate hosting revenue but would not fundamentally alter Oracle's trajectory. Its significance was political. Ellison's relationship with
Donald Trump, cultivated through dinners at Mar-a-Lago and public displays of mutual admiration, had given Oracle a proximity to political power that was unusual for an enterprise software company. "In the case of Larry, Larry Ellison, it's well beyond technology, sort of CEO of everything," Trump said at a press conference the day after his inauguration in 2025. The comment was revealing not for what it said about Ellison but for what it said about Oracle's strategic positioning: in an era of techno-nationalism, where data sovereignty and AI infrastructure were becoming matters of state policy, having the ear of the president was a competitive advantage as tangible as any technical capability.
Ellison's political connections extended beyond the United States. His relationship with Israeli politics was well-documented. His investments in healthcare — spurred by a deep personal interest in longevity science — had given him connections to policy circles around aging populations and medical technology. At eighty-one, Ellison was, as Fortune reported, making his "next big bet" on redefining how long and how well humans live. The bet was partly philanthropic, partly commercial, and entirely consistent with a man who had spent five decades refusing to accept limits imposed by others.
The Balance Sheet Bet
By late 2025, Oracle's financial profile had undergone the most dramatic transformation since the 1990 crisis. The company that had spent two decades as a free-cash-flow machine — generating $18.7 billion in operating cash flow in fiscal 2024, funding acquisitions and buybacks from operations — was now projecting negative free cash flow for multiple years as it invested in AI data center infrastructure.
The numbers were stark. Capital expenditures in Oracle's fiscal Q2 2026 were projected at $8.2 billion, double the roughly $4 billion a year earlier.
Free cash flow was estimated at negative $5.9 billion for the quarter, compared to positive $2.7 billion the year before. Oracle had sold tens of billions of dollars in bonds — both in its own name and through project-finance entities backing specific data center developments. The cost of insuring Oracle's debt against default reached its highest level since March 2009.
They are stretching their balance sheet about as far as they can, their free cash flow is negative and their balance sheet is highly levered. Their neck is sticking out.
The debate among investors was not about the magnitude of the opportunity — few disputed that AI infrastructure demand was real and growing — but about the circularity of Oracle's growth narrative. A significant portion of Oracle's remaining performance obligations came from AI companies whose own revenue was heavily dependent on continued investor enthusiasm for AI. OpenAI, Oracle's largest cloud customer, was itself a startup burning cash, dependent on fundraising rounds valued at levels that implied continued exponential growth. If AI spending slowed, or if OpenAI's fundraising trajectory faltered, Oracle's $300 billion contract could prove to be something less than $300 billion.
"It won't matter as much as the overarching story of customer concentration, how are they financing all this?" said Gabelli Funds analyst Ryuta Makino. "They're going to be free cash flow negative for the next couple of years during the data center build out. So there's a lot of question marks surrounding that."
The stock reflected this anxiety. After hitting an all-time high in September 2025 on the euphoria of the AI contract announcements, Oracle shares plunged 33% by December 2025. The Big Short investor Michael Burry publicly added to his short positions. Analysts warned that Oracle's fiscal second-quarter earnings, however strong, would not resolve the fundamental questions about capital structure, customer concentration, and the sustainability of the AI buildout.
This was the Larry Ellison bet in its purest form. The man who had committed Oracle to relational databases before IBM, to enterprise applications before SAP consolidated the market, to cloud infrastructure after the hyperscalers had a decade's head start — that man was now committing Oracle's balance sheet to the most capital-intensive bet in the company's history, on the thesis that AI infrastructure demand would not merely continue but accelerate for the rest of the decade. If he was right, Oracle would become one of the five most important technology companies on earth. If he was wrong, the debt load could become an existential threat.
Co-CEOs and the Architecture of Succession
The October 2025 appointment of Clay Magouyrk and Mike Sicilia as co-CEOs was Oracle's third leadership transition and its most consequential. Magouyrk, who had built OCI's second-generation architecture and overseen its emergence as an AI training platform, represented the infrastructure future. Sicilia, who had spent two decades embedding Oracle's application suites into industry-specific verticals — healthcare, financial services, utilities — represented the application layer that generated Oracle's recurring revenue.
The complementary structure was deliberate. HBR's Michael Watkins, analyzing the appointment, noted that "the rationale for the new dual leadership structure is complementary business expertise and shared commitment to AI." But the subtext was more complex. Oracle under Catz had been a financial engineering masterpiece — disciplined cost management, aggressive share repurchases, margin expansion through operating leverage. Oracle under Magouyrk and Sicilia would need to be something different: a capital-deployment engine capable of building physical infrastructure at a scale the company had never attempted.
Catz, for her part, remained one of the wealthiest self-made women in technology. Her net worth, driven almost entirely by Oracle stock accumulated over twenty-six years, reached approximately $3.4 billion in September 2025 before the stock's subsequent decline. She had climbed from Wall Street investment banker to the CEO suite of one of the world's largest technology companies, and her tenure had overseen Oracle's market capitalization growing from roughly $150 billion to nearly $1 trillion.
The leadership transition also surfaced a question Oracle had never had to answer: What does Oracle look like without Larry Ellison? Not the governance question — Ellison, at eighty-one, showed no signs of reducing his involvement and retained his 41% stake, his chairmanship, and his CTO title. But the strategic question. Oracle's history was, in every meaningful sense, the expression of a single person's competitive psychology, technical intuition, and willingness to bet. Could that be institutionalized? Could Magouyrk and Sicilia replicate the judgment of a founder who had been right about the direction of enterprise computing more often than he had been wrong — even when the timing was off by years or decades?
The Database at the End of the World
Here is the thing about Oracle that even its critics cannot dispute: the database works. After forty-eight years, through every platform shift and competitive assault and technology revolution, Oracle Database remains the system of record for the most demanding, highest-stakes data workloads on earth. The world's largest banks settle trades through Oracle. The world's largest airlines manage reservations through Oracle. Governments run tax systems, healthcare records, and intelligence databases on Oracle. The International Space Station runs Oracle.
The reason is not inertia, though inertia helps. The reason is that Oracle Database, through five decades of continuous investment, has achieved a level of reliability, performance, and feature richness that no competitor has fully replicated. Its optimizer — the engine that determines the most efficient way to execute a query — is arguably the most sophisticated piece of software engineering in the enterprise technology stack. Its RAC (Real Application Clusters) technology provides high availability at a scale that cloud-native databases are only now approaching. Its security features, its partitioning capabilities, its support for multiple data models within a single engine — these are the accumulated advantages of a product that has received more engineering investment over a longer period than almost any other software product in existence.
The Autonomous Database, introduced in 2018, represented Ellison's vision of the database's next evolution: a self-tuning, self-patching, self-securing system that would eliminate the need for database administrators. The name was characteristically immodest. The technology was genuinely impressive — machine learning models that could optimize query performance, apply security patches without downtime, and detect anomalies in real time. Whether it would be enough to retain customers who were increasingly comfortable with cloud-native alternatives was the open question.
In fiscal year 2024, Oracle's cloud services and license support revenues — the recurring revenue stream that included database subscriptions, cloud applications, and infrastructure services — totaled $39.4 billion, up 12% year-over-year. Cloud infrastructure revenue alone was growing at 42% in the fiscal fourth quarter. The database was still the foundation. But the superstructure was changing.
Remaining Performance Obligations
The number that defined Oracle's 2025 was not revenue, not earnings, not even the stock price. It was remaining performance obligations — RPO — the total value of contracted but not yet recognized revenue. In Oracle's fiscal Q2 2026, RPO reached approximately $520 billion, a more than 400% increase from the year prior.
Five hundred and twenty billion dollars. The figure is almost impossible to contextualize. It is larger than the
GDP of Sweden. It exceeds the combined annual revenue of every enterprise software company on earth outside of Microsoft. It represents, if recognized over the expected contract periods, a revenue trajectory that would make Oracle one of the largest companies in the world by any measure.
But RPO is not revenue. It is a promise — a contractual commitment from customers who may or may not actually consume the services they have contracted for, whose own businesses may expand or contract, and whose ability to pay depends on their own fundraising, revenue generation, and strategic priorities. The gap between RPO and recognized revenue is where the bull and bear cases for Oracle diverge most sharply. Bulls see $520 billion as a floor — proof that the demand for AI infrastructure is not merely real but overwhelming, that Oracle has locked in a decade of growth. Bears see $520 billion as a fantasy — a number inflated by contracts with AI companies that are themselves valued on projections of projections, in an industry where yesterday's trillion-dollar TAM estimate becomes tomorrow's writedown.
In Oracle's Redwood City origins and Austin present and Nashville future, in its database roots and cloud infrastructure ambitions, in the personality of an eighty-one-year-old founder who refuses to cede control and the two engineers who have been asked to build the future on his vision — in all of this, the $520 billion number hangs like an oracle's prophecy. Ambiguous. Consequential. Demanding interpretation.
On the wall of Larry Ellison's office, for decades, hung a painting of a samurai warrior. Not a general commanding armies. A single warrior, alone, prepared for combat.
