On January 21, 2025, President Donald Trump stood in the White House alongside three men who would collectively commit $500 billion to building artificial intelligence infrastructure: Sam Altman from OpenAI, Masayoshi Son from SoftBank, and Larry Ellison from Oracle.
At 80 years old, Ellison was the oldest person in the room by two decades. He had founded Oracle 48 years earlier, in 1977, when personal computers were still a novelty and the internet did not exist. Now, in his fifth decade leading the company, Ellison was making Oracle's most audacious bet: transforming a legacy database vendor into the infrastructure backbone for the AI revolution.
"This will be the largest AI infrastructure project in history," Trump declared. The Stargate project, as it was named, would build massive datacenters across the United States and globally, deploying hundreds of thousands of NVIDIA GPUs to train the next generation of AI models. Oracle would provide the cloud infrastructure. OpenAI would provide the AI models. SoftBank would provide much of the capital. Together, they aimed to invest up to $500 billion by 2029.
For Ellison, the announcement represented vindication. Oracle had spent decades competing against Amazon Web Services, Microsoft Azure, and Google Cloud—and losing. AWS commanded 30% of the cloud market. Azure was close behind. Google Cloud had 10%. Oracle had barely 2%.
But Ellison saw an opportunity that his competitors might have missed: AI workloads require fundamentally different infrastructure than traditional cloud computing. They demand massive GPU clusters, proximity to power sources, sovereign data controls for government customers, and the ability to scale to unprecedented levels. Oracle had been quietly building for this moment, constructing over 162 datacenter facilities globally and securing power capacity measured in gigawatts rather than megawatts.
Standing in the White House, Ellison told reporters that Oracle would build datacenters "bigger than anything that's ever been built before." He described facilities consuming 1.2 gigawatts of power—enough to power a small city—and spanning over 1,000 acres. "We're going to have AI that can cure cancer," Ellison said, explaining his vision for AI-designed mRNA vaccines that could be created "robotically" in 48 hours.
The question facing Oracle: Could an 80-year-old billionaire, leading a company known more for database software than cutting-edge cloud infrastructure, execute on a $500 billion vision that would require Oracle to outbuild Amazon, Microsoft, and Google combined?
Lawrence Joseph Ellison's path to becoming one of the world's wealthiest individuals was anything but conventional. Born on August 17, 1944, in New York City to an unwed Jewish mother, Ellison contracted pneumonia at nine months old. His mother, Florence Spellman, sent him to Chicago to be raised by her aunt and uncle, Lillian and Louis Ellison, who adopted the baby.
Ellison grew up in a modest two-bedroom apartment on Chicago's South Side. His adoptive father, a Russian immigrant who worked in government housing, frequently told young Larry that he would never amount to anything—a criticism that would fuel Ellison's competitive drive for decades. Despite showing aptitude for mathematics and science, Ellison struggled with traditional education. He attended the University of Illinois from 1962 to 1964 but dropped out after his adoptive mother Lillian died. He briefly enrolled at the University of Chicago in 1966 but again failed to complete a degree.
In the late 1960s, Ellison moved to California and worked as a computer programmer for various companies. This was the era when computers filled entire rooms, programming required punch cards, and the idea of "software" as a distinct product was still emerging. At Ampex Corporation beginning in 1973, Ellison met two colleagues who would change his life: programmer Ed Oates and supervisor Bob Miner.
The pivotal moment came in 1977. Ellison, along with Oates and Miner, founded Software Development Laboratories with an initial investment of $2,000—$1,200 from Ellison and $800 from the others. Their first contract came from an unexpected source: the Central Intelligence Agency, which needed a database management system for a project code-named "Oracle."
Ellison had read a research paper by IBM scientist Edgar F. Codd describing relational databases—a revolutionary concept that organized data in tables rather than hierarchical structures. IBM was developing its own relational database called System R, but the project remained internal. Ellison realized that he could build a commercial version before IBM brought theirs to market.
In 1979, the company released Oracle Version 2 (they skipped Version 1 to make the product seem more mature). It was the first commercial relational database to use Structured Query Language (SQL), a programming language that would become the industry standard. The company had fewer than 10 employees and generated less than $1 million in annual revenue.
The breakthrough came in 1981 when IBM—needing a relational database for its customers but unwilling to commercialize its own internal System R—decided to standardize on Oracle's database. Suddenly, Oracle had the IBM seal of approval. Fortune 500 companies that trusted IBM's judgment began adopting Oracle's software.
The company grew explosively throughout the 1980s. Revenue increased from $282,000 in 1980 to $15 million in 1983 to $55 million in 1986, when Oracle went public. By 1990, Oracle's revenue exceeded $1 billion, making it one of the fastest-growing software companies in history.
But success brought challenges. In 1990, Oracle faced a crisis when aggressive accounting practices led to restated earnings and a Securities and Exchange Commission investigation. The company's stock plummeted 80%. Ellison, who had built a reputation for brash salesmanship and luxurious spending—yachts, mansions, jets—faced questions about whether Oracle could survive.
Ellison responded by bringing in professional management, tightening financial controls, and focusing Oracle on its core database business. The company not only survived but thrived. Through the 1990s and 2000s, Oracle dominated the enterprise database market, eventually commanding over 40% market share. Ellison pursued an aggressive acquisition strategy, spending over $50 billion to acquire PeopleSoft, Siebel Systems, BEA Systems, Sun Microsystems, and dozens of other companies.
By the time Ellison stepped down as CEO in September 2014, Oracle had grown into a $38 billion revenue company with 130,000 employees. Ellison's personal stake in Oracle had made him one of the world's wealthiest individuals, with a net worth exceeding $50 billion.
But as Ellison transitioned from CEO to Executive Chairman and Chief Technology Officer, Oracle faced an existential threat: the cloud.
In 2006, Amazon Web Services launched its Elastic Compute Cloud (EC2), allowing companies to rent computing capacity by the hour rather than building their own datacenters. The cloud computing revolution had begun—and Oracle was caught flat-footed.
Ellison had been dismissive of cloud computing in its early years. In 2008, he told analysts: "The computer industry is the only industry that is more fashion-driven than women's fashion. Maybe I'm an idiot, but I have no idea what anyone is talking about when they talk about cloud computing."
That dismissiveness would cost Oracle dearly. By the time Ellison recognized cloud computing as a genuine paradigm shift rather than marketing hype, Amazon had a five-year head start. Microsoft, pivoting aggressively under CEO Satya Nadella, poured billions into Azure. Google invested in Google Cloud Platform. Oracle was years behind.
The consequences showed in Oracle's financial performance. From 2014 to 2019, Oracle's revenue growth stagnated, hovering around $37-39 billion annually while AWS grew from $4.6 billion in 2014 to $35 billion in 2019. Wall Street analysts questioned whether Oracle, built for the era of on-premise software licenses, could transform into a cloud-native company.
Ellison made his move in 2016, launching Oracle Cloud Infrastructure (OCI) as a ground-up rebuild of Oracle's cloud offering. Unlike Oracle's earlier cloud efforts—which essentially took existing on-premise software and ran it in Oracle's datacenters—OCI was designed from scratch for cloud-native workloads. Oracle promised better performance and lower costs than AWS.
The market was skeptical. Oracle's reputation for aggressive sales tactics, expensive licensing, and vendor lock-in made enterprises wary of relying on Oracle for cloud infrastructure. Why trust Oracle when AWS, Azure, and Google Cloud offered vendor-neutral platforms?
Oracle's cloud business did grow, but slowly. By 2020, OCI revenue was less than $5 billion annually—a fraction of AWS's $45 billion. Oracle stock traded at modest multiples, with investors pricing in limited growth prospects. Ellison's net worth, heavily concentrated in Oracle shares, stagnated around $60-70 billion.
Then came the AI revolution—and with it, Oracle's unexpected opportunity for redemption.
In November 2022, OpenAI released ChatGPT to the public. Within five days, the chatbot had attracted one million users. Within two months, it reached 100 million monthly active users—the fastest consumer application growth in history.
Behind ChatGPT's viral success was an infrastructure challenge that most users never saw: training and running large language models requires enormous computing resources. GPT-3, released in 2020, had 175 billion parameters and required tens of thousands of NVIDIA GPUs to train. GPT-4, released in March 2023, was rumored to be significantly larger. Future models would be larger still.
The AI infrastructure requirements differed fundamentally from traditional cloud workloads. Training large AI models required tightly coupled GPU clusters with high-speed networking between tens of thousands of accelerators. Inference—actually running the AI models to respond to user queries—required massive parallelism and low latency. Both demanded proximity to enormous power capacity, as GPU clusters could consume 50-100 megawatts for training and potentially gigawatts at scale.
Amazon, Microsoft, and Google Cloud had built their infrastructure for traditional cloud workloads: web servers, databases, storage. They had datacenters globally, but most were designed for 5-20 megawatt capacity. Retrofitting existing datacenters for AI workloads proved challenging. Power constraints, cooling limitations, and network architecture designed for different workloads all created bottlenecks.
Ellison saw the opportunity clearly. In a September 2023 earnings call, he told investors: "The thing that's really different about this moment in time is the AI workload. The AI training workload is like nothing we've ever seen before in terms of the compute requirements and the network requirements."
Oracle had advantages that its larger competitors lacked. First, Oracle was building new datacenters specifically for AI workloads rather than retrofitting existing facilities. The company could design for GPU clusters from the ground up, with appropriate power, cooling, and networking infrastructure.
Second, Oracle had committed to building datacenters at unprecedented scale. While AWS might build a 20-50 megawatt datacenter, Oracle announced plans for facilities consuming 800 megawatts to 1.2 gigawatts—large enough to deploy hundreds of thousands of GPUs in tightly coupled clusters.
Third, Oracle pursued a different datacenter strategy than its competitors. Rather than concentrating on a few mega-regions like AWS (which had 33 regions globally), Oracle built a distributed network of smaller cloud regions. By late 2024, Oracle operated 101 cloud regions worldwide—more than AWS, Azure, and Google Cloud combined. This distributed approach appealed to customers requiring data sovereignty or low-latency access to cloud services.
Fourth, Oracle had cultivated relationships with governments and regulated industries through its on-premise database business. These customers—defense agencies, intelligence services, healthcare systems, financial regulators—had strict requirements around data sovereignty and security. Oracle's sovereign cloud offerings, including Oracle Government Cloud for US defense and intelligence agencies, provided infrastructure that met classification requirements up to Top Secret.
The AI infrastructure opportunity gave Oracle a chance to leapfrog its competitors in a new market category. If Oracle could become the preferred infrastructure provider for AI workloads—particularly for large-scale model training and sovereign AI deployments—the company could finally break out of its 2% cloud market share trap.
Ellison began positioning Oracle aggressively for this opportunity. In fiscal year 2024, Oracle's capital expenditures surged to $6.9 billion, up from $4 billion the prior year. In fiscal 2025, capital expenditures would more than triple to $21.2 billion. The company announced plans to increase CapEx to $35 billion in fiscal 2026.
By comparison, AWS's parent company Amazon spent about $75 billion on CapEx in 2024 across all divisions (including fulfillment centers and other non-cloud infrastructure). Microsoft spent about $53 billion. Google spent about $32 billion. Oracle, a fraction of their size in revenue, was investing at a rate suggesting an all-in bet on becoming an AI infrastructure giant.
The question was whether Oracle could find customers to fill all that capacity.
The relationship between Oracle and OpenAI began in 2023, when OpenAI was experiencing growing pains with its existing infrastructure provider, Microsoft Azure.
Microsoft had invested $1 billion in OpenAI in 2019 and committed to be OpenAI's exclusive cloud provider. When ChatGPT exploded in popularity in late 2022, OpenAI's compute requirements surged beyond what Microsoft had anticipated. OpenAI needed to train successively larger models—GPT-4, then GPT-4.5, then GPT-5—each requiring more GPUs and more interconnected datacenter capacity than the last.
Azure's infrastructure, while extensive, wasn't optimized for the kind of tightly coupled supercomputer clusters that optimal AI training required. Azure datacenters were distributed globally and designed for cloud workloads that could run on discrete servers. Training a 1-trillion-parameter language model required tens of thousands of GPUs connected via ultra-high-speed networking, ideally in a single physical location.
OpenAI began exploring multi-cloud options. In June 2024, Bloomberg reported that OpenAI signed a $300 billion contract with Oracle to access Oracle Cloud Infrastructure for model training. The deal shocked the industry—it was one of the largest cloud infrastructure commitments ever made, and it positioned Oracle as a peer to Microsoft in supporting OpenAI's infrastructure needs.
According to people familiar with the negotiations, Oracle offered OpenAI several advantages over Azure. First, Oracle was building datacenters specifically designed for AI training at scales Azure couldn't match. Oracle's largest facilities would connect over 100,000 NVIDIA GPUs via high-bandwidth networking—effectively creating supercomputers larger than any Azure datacenter cluster.
Second, Oracle offered more flexible pricing. OpenAI's compute costs were massive—estimated at $500,000 per day just to run ChatGPT's inference, not including training costs. Every percentage point savings on infrastructure translated to millions in reduced burn rate. Oracle, desperate to win the OpenAI deal, reportedly offered pricing below AWS and Azure rates.
Third, Oracle committed to securing GPU capacity at a time when NVIDIA GPUs were the industry's scarcest resource. NVIDIA's H100 GPUs, essential for training large models, had 18-month lead times. Oracle leveraged its relationship with NVIDIA and its willingness to commit enormous CapEx to secure allocations that OpenAI needed.
The Oracle-OpenAI relationship evolved through 2024. Oracle announced it would deploy "AI supercomputers" with up to 131,072 NVIDIA GPUs—clusters larger than anything AWS, Azure, or Google Cloud had publicly disclosed. OpenAI began using Oracle infrastructure for training workloads while continuing to use Azure for inference and customer-facing services.
Then came Stargate.
The Stargate project, announced on January 21, 2025, expanded the Oracle-OpenAI partnership to an entirely new scale. The joint venture, led by SoftBank's Masayoshi Son as chairman, committed to invest $100 billion immediately and up to $500 billion by 2029 in AI infrastructure.
According to details Ellison disclosed during the announcement, Stargate would build datacenters unlike anything the tech industry had seen. The first facility in Abilene, Texas would eventually run 500,000 NVIDIA GPUs consuming 1.2 gigawatts of power across eight buildings spanning 1,000 acres. For context, that's approximately the power consumption of the entire city of San Francisco, dedicated to a single AI training facility.
Stargate planned to build 10 such datacenter complexes across the United States, with international expansion into the United Kingdom, Norway, Japan, and the United Arab Emirates. By the venture's own projections, Stargate would deploy 4.5 gigawatts of AI datacenter capacity initially, expanding toward 10 gigawatts by 2029.
Oracle would build and operate the infrastructure. OpenAI would use it to train increasingly large models. SoftBank would provide much of the capital (Son committed $100 billion). MGX, a UAE-based investment firm, would help fund international expansion. Other partners included NVIDIA (providing GPUs), Arm (chip architecture), and various energy and real estate developers.
For Ellison, Stargate represented Oracle's path to becoming an AI infrastructure giant. If the project succeeded, Oracle would operate some of the world's largest AI training facilities, cementing relationships with OpenAI, NVIDIA, and governments globally. Oracle's cloud business, stuck at low single-digit market share for traditional workloads, could dominate a new category: sovereign AI infrastructure at national scale.
But Stargate also presented enormous risks.
Building a $500 billion AI infrastructure network by 2029 requires Oracle to solve problems that have no clear precedents in the tech industry.
The first challenge is power. A 1.2-gigawatt datacenter requires as much electricity as a mid-sized city. Securing power commitments of that scale involves negotiating with utilities, potentially building dedicated substations or even power plants, and obtaining regulatory approvals that can take years. Oracle must do this not once but dozens of times across the United States and internationally.
According to energy industry analysts, the US power grid is already strained by AI datacenter growth. The North American Electric Reliability Corporation projects that peak electricity demand could grow by 15-25% over the next five years, with AI datacenters accounting for much of the increase. In some regions, utilities have told hyperscalers that new datacenter capacity cannot be added for 3-5 years due to transmission constraints.
Oracle's solution involves locating datacenters near power generation sources and, in some cases, building dedicated power infrastructure. Ellison has discussed using small modular nuclear reactors (SMRs) to power Stargate facilities, though no SMRs are currently commercially available in the United States. In the near term, Oracle relies on natural gas power plants and renewable energy purchases to secure capacity.
The second challenge is construction speed. Oracle has committed to bringing the first Stargate facilities online by late 2025, with the full 10-facility US network operational by 2027. Building gigawatt-scale datacenters in 18-36 months requires parallelizing construction, securing scarce materials (cooling systems, backup generators, specialized electrical equipment), and managing thousands of contractors simultaneously.
Oracle's construction strategy involves modular datacenter designs that can be replicated across sites and partnerships with large-scale real estate developers. The company announced agreements with Digital Realty and other datacenter operators to accelerate buildout. Still, industry observers question whether Oracle can move fast enough to meet its timelines.
The third challenge is GPU procurement. Stargate plans to deploy millions of NVIDIA GPUs over four years. NVIDIA's production capacity, while expanding, struggles to keep pace with demand. In 2024, NVIDIA's datacenter revenue exceeded $100 billion annually, but customers still faced 6-12 month lead times for H100 and H200 GPUs. The newer Blackwell architecture, announced in 2024, had even longer backlogs.
Oracle's strategy relies on its position as one of NVIDIA's largest customers. According to people familiar with the matter, Oracle committed to purchasing over $50 billion worth of NVIDIA GPUs through 2027—one of NVIDIA's largest customer commitments. This gives Oracle priority allocation, but still depends on NVIDIA's ability to scale manufacturing through partners like TSMC.
The fourth challenge is technical execution. Building datacenters is one thing; operating them reliably at unprecedented scale is another. A 131,072-GPU cluster represents over 10 million individual components (GPUs, CPUs, memory modules, network interfaces, switches, storage devices). At that scale, multiple components fail every hour. Oracle's software must detect failures, route around them, and maintain training jobs that might run continuously for weeks or months.
Oracle's engineering teams have built custom cluster management software, networking protocols optimized for AI workloads, and monitoring systems designed for exascale computing. But the gap between PowerPoint presentations and production systems is vast. AWS, Azure, and Google Cloud have spent 15+ years developing operational expertise in running planetary-scale infrastructure. Oracle must catch up in a fraction of that time.
The fifth challenge is customers. Stargate is being built primarily for OpenAI, but OpenAI alone cannot justify $500 billion in infrastructure investment. Oracle must win additional customers—other AI labs, governments, enterprises training proprietary models—to fill the capacity.
Oracle's customer pipeline includes several promising prospects. The company has disclosed AI infrastructure contracts with X (Elon Musk's social network), Cohere (a Canadian AI startup), and various government agencies. Oracle's sovereign cloud offerings appeal to countries wanting AI capabilities without dependence on US hyperscalers. But whether these contracts will generate the tens of billions in annual revenue needed to justify Stargate's investment remains unclear.
The sixth challenge is financial sustainability. Oracle's capital expenditure surge—from $6.9 billion in fiscal 2024 to a projected $35+ billion in fiscal 2026—represents an unprecedented bet for a company generating $50+ billion in annual revenue. If Oracle cannot fill the datacenter capacity with paying customers, the company will be saddled with expensive, underutilized infrastructure and massive depreciation charges.
Oracle's cloud infrastructure revenue grew 52% year-over-year in Q2 fiscal 2025, reaching about $10 billion annualized. But even doubling that revenue would only justify a fraction of Oracle's planned CapEx. The company needs to grow cloud revenue to $30-50 billion annually to make the economics work—a target that requires winning customers away from AWS, Azure, and Google Cloud at a pace Oracle has never achieved.
Ellison's answer to these challenges is characteristically bold. "We will build more cloud infrastructure datacenters than all of our competitors combined," he told investors in late 2024. Oracle's strategy is to move faster, build bigger, and offer better economics than the competition. Whether an 80-year-old can out-execute Amazon, Microsoft, and Google—companies with larger revenues, more resources, and deeper cloud expertise—is the $500 billion question.
While Stargate grabbed headlines, Oracle's quieter sovereign AI strategy might ultimately prove more significant to the company's transformation.
Sovereign AI refers to nations' desire to build AI capabilities using domestic infrastructure, data, and governance frameworks rather than relying on foreign cloud providers. As AI becomes critical to national competitiveness and security, governments increasingly view AI infrastructure as strategic assets comparable to telecommunications, energy grids, or transportation networks.
Oracle identified this trend earlier than its competitors and built products specifically for government and sovereign cloud requirements. Oracle Government Cloud, launched in 2015, provides cloud infrastructure that meets US Department of Defense security requirements up to Top Secret classification. Oracle EU Sovereign Cloud, announced in 2024, offers European customers datacenters located in Europe, operated by European entities, with customer data kept within EU jurisdiction.
These offerings address concerns that prevent governments from using standard AWS, Azure, or Google Cloud services. When a country's intelligence service or military runs workloads on AWS, the data physically resides in Amazon datacenters, potentially subject to US legal jurisdiction. When researchers train AI models on Azure, Microsoft has theoretical access to model weights and training data. For sovereign cloud customers, these arrangements are unacceptable.
Oracle's solution is "distributed cloud"—essentially full Oracle Cloud Infrastructure regions deployable in customer datacenters or government facilities. A country can have the full public cloud experience (200+ services, APIs compatible with Oracle's global cloud) while maintaining physical control over hardware and data. Oracle provides the software, professional services, and support, but the customer controls the infrastructure.
This model appeals to various customer segments. In 2025, Oracle announced major sovereign cloud deals including:
The United Arab Emirates government committed $13 billion (47.8 billion AED) to an AI and cloud transformation program using Oracle infrastructure. The program aims to migrate all government services to 100% sovereign cloud by 2027, with AI capabilities built on OCI and NVIDIA accelerators deployed locally.
The UK government selected Oracle to provide sovereign cloud infrastructure for sensitive workloads, with a commitment exceeding $5 billion over five years. Oracle would build dedicated datacenters in the UK, operated by British personnel, meeting stringent data residency and security requirements.
Japan's government signed agreements for Oracle to invest over $8 billion in Japanese datacenter infrastructure to support the country's AI sovereignty ambitions. Oracle would partner with local cloud operators to provide OCI services while keeping data within Japanese jurisdiction.
Malaysia's government designated the Johor region as a major AI hub, with Oracle (supporting ByteDance's training workloads) as an anchor tenant. The buildout positioned Johor as "the world's second-largest AI hub" after certain US regions, demonstrating how Oracle's willingness to build large-scale infrastructure unlocked government support.
Several European Union countries negotiated sovereign cloud agreements as the EU AI Act implementation created demand for AI infrastructure meeting European data protection and governance requirements.
Oracle's sovereign cloud strategy created a differentiated market position. AWS, Azure, and Google Cloud all offered government cloud regions, but their models generally involved the hyperscaler operating dedicated infrastructure rather than customers maintaining physical control. For the most sensitive workloads—intelligence services training AI models on classified data, for example—Oracle's distributed cloud model provided the only acceptable option.
The sovereign AI market could grow larger than the commercial AI market. Gartner projected that by 2028, over 40% of AI model training workloads globally would run on sovereign cloud infrastructure as governments and regulated industries prioritized data control over public cloud convenience. If Oracle captured even 25% of that market, it could represent $50-100 billion in annual revenue—transforming Oracle's cloud business.
Ellison understood the strategic significance. "We're not just building datacenters," he said during Oracle CloudWorld 2024. "We're building sovereign nations' AI capabilities. When the UAE or Japan or the UK wants AI that serves their interests, under their control, that's where Oracle comes in."
The sovereign AI strategy aligned with Oracle's historical strengths: selling to governments and regulated industries, navigating complex procurement processes, and building long-term customer relationships based on trust and lock-in. These were precisely the areas where AWS, optimized for self-service developer adoption, struggled.
But sovereign AI also came with challenges. Government procurement moved slowly, often requiring years from initial discussions to production deployments. Sovereign cloud customers demanded extensive customization, local partnerships, and technology transfers that reduced Oracle's margins. Building infrastructure in dozens of countries multiplied operational complexity.
Most importantly, sovereign AI contracts often came with geopolitical baggage. Oracle's UAE deal, for example, raised questions about providing advanced AI capabilities to authoritarian governments. The Malaysia buildout supporting ByteDance (TikTok's parent company) intersected with US-China technology competition and data security concerns. Oracle's role in Saudi Arabia's NEOM smart city project drew criticism from human rights organizations.
Ellison, characteristically, showed little concern for such criticisms. His priority was establishing Oracle as the infrastructure provider for governments worldwide, regardless of their political systems. In an industry where AWS, Microsoft, and Google increasingly faced pressure to limit sales to certain governments, Oracle's willingness to serve any paying customer became a competitive advantage.
On October 6, 2025, President Trump announced that Oracle and Larry Ellison would play a "big" role in managing TikTok following a complex arrangement that avoided an outright ban of the Chinese-owned social media platform in the United States.
The TikTok deal had been years in the making. In 2020, during Trump's first term, the administration threatened to ban TikTok over national security concerns related to ByteDance's Chinese ownership. Oracle was among the companies that bid to acquire TikTok's US operations. That deal ultimately didn't materialize, but it established Oracle's relationship with both TikTok and the Trump administration.
The 2025 arrangement differed from an outright acquisition. Under the structure announced, Oracle would host TikTok's US user data in Oracle Cloud Infrastructure and provide "transparency" controls allowing US government officials to audit what data TikTok collected and how it was used. Oracle effectively became TikTok's "trusted technology provider"—a role that generated substantial cloud revenue while positioning Oracle at the intersection of US-China technology policy.
For Ellison, the TikTok deal represented several strategic wins. First, it brought a massive cloud customer: TikTok's US operations generated hundreds of petabytes of data daily, translating to potentially billions in annual infrastructure spending. Second, it demonstrated Oracle's value proposition for sensitive, regulated workloads that required transparency and auditability. Third, it strengthened Oracle's relationship with the Trump administration at a time when government cloud contracts were increasingly important to Oracle's strategy.
That relationship had grown significantly closer during Trump's second term. Ellison had been a Republican donor for years, contributing millions to candidates aligned with his libertarian-conservative politics. But his role in the Trump administration went beyond typical business-politician dynamics.
When Trump announced Stargate in January 2025, Ellison stood directly beside the president, suggesting an advisor-level relationship rather than merely a CEO pitching a project. In subsequent months, Oracle secured several administration-adjacent deals: the TikTok infrastructure role, discussions about Oracle providing cloud infrastructure for government AI initiatives, and reportedly consideration for Oracle cloud services to support various federal agency modernization projects.
Ellison's son, David Ellison, separately negotiated to acquire CBS News through his Skydance Media company, reportedly with encouragement from Trump allies who saw an opportunity to influence a major news network's coverage. The elder Ellison's financial backing for David's media ambitions connected to broader discussions about conservative influence in media.
Critics raised concerns about the concentration of power and potential conflicts of interest. Oracle providing infrastructure for TikTok while Larry Ellison had direct access to Trump created questions about whether Oracle's business interests influenced US policy toward Chinese technology companies. Oracle's sovereign cloud deals with foreign governments, some authoritarian, alongside close ties to the US administration, suggested potential for Oracle to be caught between competing national interests.
Ellison's public statements sometimes intensified these concerns. During an Oracle financial analyst meeting in September 2024, Ellison discussed AI-powered surveillance capabilities, saying: "Citizens will be on their best behavior because we are constantly recording and reporting everything that's going on." He described a future where AI systems monitored police body cameras, analyzed behavior patterns, and flagged potential wrongdoing.
The comments sparked immediate backlash from privacy advocates who saw them as endorsing a surveillance state. Ellison seemed genuinely puzzled by the negative reaction, arguing that AI surveillance could prevent police misconduct and reduce crime. The episode highlighted Ellison's techno-optimist worldview: he saw powerful AI and pervasive data collection as tools for social benefit, not threats to liberty.
Oracle's positioning between government contracts, Chinese technology companies, and controversial AI applications reflected the increasingly complex environment in which tech infrastructure providers operated. AWS, Azure, and Google Cloud all faced similar tensions—balancing commercial opportunities with ethical concerns, serving both democracies and authoritarian governments, building surveillance-capable technologies while professing commitment to privacy.
For Ellison, at 80 years old and with a net worth exceeding $200 billion, these concerns seemed secondary to his primary goal: making Oracle relevant in the AI era. If that required partnerships with TikTok, deals with the UAE, or surveillance capabilities that worried civil libertarians, those were acceptable trade-offs for business success.
Oracle's AI infrastructure bet occurs against a backdrop of intensifying competition among cloud providers, each pursuing different strategies to capture AI workload revenue.
Amazon Web Services maintained its overall cloud market leadership with approximately 31% share, but its AI strategy reflected caution. AWS had invested heavily in its own AI chips—Trainium for training and Inferentia for inference—to reduce dependence on NVIDIA and offer customers lower-cost alternatives. The company had also built strong partnerships with AI startups including Anthropic ($4 billion investment), Hugging Face, Stability AI, and others, offering cloud credits and technical support in exchange for AWS loyalty.
But AWS's distributed datacenter architecture, optimized for broad geographic coverage rather than concentrated supercomputing clusters, created challenges for the largest AI training workloads. AWS had responded by building dedicated AI regions with GPU densities comparable to Oracle's facilities, but the buildout moved slower than competitors' given AWS's massive existing infrastructure base that still needed support.
Microsoft Azure held the strongest position in generative AI thanks to its $13 billion OpenAI investment and equity stake. Azure powered OpenAI's ChatGPT and API services, generating billions in revenue and cementing Azure's position as the default cloud for enterprises adopting OpenAI's technology. Microsoft had integrated AI capabilities across its product portfolio—Office 365, Dynamics 365, GitHub Copilot—creating a flywheel where Azure AI usage drove broader Microsoft product adoption.
Azure's challenge was capacity. The OpenAI partnership consumed so much GPU capacity that Azure reportedly struggled to serve other AI customers' training needs. Microsoft's 2025 capital expenditures exceeded $80 billion as the company raced to build datacenter capacity, but demand outpaced supply. Some AI startups complained that Azure GPU availability was insufficient, pushing them to multi-cloud strategies.
This capacity crunch created Oracle's opening. When OpenAI needed infrastructure beyond what Azure could immediately provide, Oracle offered an alternative. The $300 billion Oracle-OpenAI deal suggested that even Microsoft's closest AI partner saw value in diversifying infrastructure providers.
Google Cloud, meanwhile, pursued a different strategy: TPUs (Tensor Processing Units), Google's custom AI chips, as an alternative to NVIDIA GPUs. Google had developed seven generations of TPUs, optimized specifically for AI workloads, and offered them at prices below comparable NVIDIA GPU instances. Google also invested heavily in AI startups including Anthropic ($2+ billion), Cohere, and others, using cloud credits and technical support to build ecosystem loyalty.
Google Cloud's market share grew to approximately 11% by late 2025, with AI workloads driving much of the growth. The company's AI case study share (18%) exceeded its overall cloud market share (11%), suggesting strong AI positioning. Google's challenge was convincing customers to adopt TPUs, which required rewriting AI training code that developers had standardized on NVIDIA's CUDA platform.
Oracle's competitive advantages in this landscape were specific but potentially decisive. First, Oracle could move faster on large-scale datacenter buildout because it wasn't constrained by an existing infrastructure base requiring ongoing support. AWS, Azure, and Google Cloud all needed to balance investments in new AI-optimized facilities with maintaining and expanding their existing infrastructure. Oracle could focus 100% of its incremental CapEx on AI.
Second, Oracle's willingness to build sovereign cloud infrastructure gave it access to government and regulated industry customers that hyperscalers struggled to serve. This market segment valued data control and sovereignty over raw performance or cost, playing to Oracle's strengths.
Third, Oracle's database business provided customer relationships and enterprise credibility that pure infrastructure plays lacked. When Oracle offered OCI to existing database customers, it came with decades of relationship history, support contracts, and mutual trust. AWS selling to Oracle database customers was a displacement sale; Oracle selling OCI to those same customers was an upsell.
Fourth, Oracle's pricing strategy undercut competitors on specific workloads. Because Oracle was desperate to gain market share, the company offered aggressive discounts to win large contracts. The OpenAI deal, for example, reportedly included pricing that Oracle's CFO later admitted would generate minimal or even negative margins in the early years—but Oracle was willing to accept those economics to establish market position.
However, Oracle's disadvantages were also significant. AWS, Azure, and Google Cloud had comprehensive service portfolios spanning hundreds of offerings—databases, analytics, machine learning tools, IoT, edge computing, and countless other categories. Oracle Cloud Infrastructure offered a narrower set of services, primarily focused on compute, storage, networking, and Oracle's own database and application software.
This meant customers adopting OCI often needed multi-cloud strategies, using Oracle for specific workloads (AI training, Oracle database) while using AWS or Azure for other services. Multi-cloud architectures increased complexity and limited Oracle's ability to capture full customer spend.
Oracle's developer ecosystem also trailed competitors dramatically. AWS, Azure, and Google Cloud had millions of developers worldwide with skills and experience on their platforms. Oracle Cloud Infrastructure had a much smaller developer community, making it harder for customers to find talent capable of managing OCI deployments.
Most fundamentally, Oracle's late start meant the company was competing against entrenched incumbents with massive revenue and customer bases. AWS generated over $100 billion annually; Azure cloud revenue exceeded $120 billion; Google Cloud surpassed $40 billion. Oracle's cloud infrastructure business was still under $15 billion. Catching up required not just matching competitors' growth rates but dramatically exceeding them—a challenge made harder as Oracle's larger competitors also invested heavily in AI infrastructure.
The competitive dynamics suggested Oracle's AI bet was high-risk, high-reward. If AI workloads fragmented across multiple infrastructure providers—with some customers using AWS, others Azure, others Oracle—based on specific requirements like sovereignty, scale, or pricing, Oracle could carve out a sustainable niche generating $30-50 billion in annual revenue. That would transform Oracle from a struggling cloud also-ran into a major player.
But if AI infrastructure exhibited winner-take-most dynamics, with most customers consolidating on one or two providers for simplicity, Oracle risked spending tens of billions on datacenter capacity that remained underutilized. The company would have expensive infrastructure, ongoing operational costs, and insufficient revenue to justify the investments.
Ellison was betting on fragmentation. His view: AI infrastructure was too important, too sensitive, and too diverse in requirements for any single provider to dominate. Countries would demand sovereign AI. Enterprises would require multi-cloud for risk management. Different AI workloads would favor different infrastructure optimizations. This fragmented market would have room for multiple winners—and Oracle, moving aggressively, could be one of them.
Larry Ellison turned 81 in August 2025, making him one of the oldest active executives in the technology industry. His longevity and continued hands-on involvement in Oracle's strategy raise questions about succession, decision-making, and whether an octogenarian can navigate the fast-moving AI landscape.
Ellison shows no signs of stepping back. He remains Executive Chairman and Chief Technology Officer, attending earnings calls, announcing major initiatives like Stargate, and making key strategic decisions. He is Oracle's largest individual shareholder with approximately 41% of the company's stock, giving him effective control regardless of his formal titles.
Those who work with Ellison describe him as intellectually sharp and intensely engaged with technology details despite his age. During earnings calls, Ellison demonstrates deep knowledge of GPU architectures, datacenter power requirements, and AI training methodologies. He personally recruited key executives for Oracle's cloud business and negotiated major partnerships including the OpenAI deal.
But age inevitably creates challenges. Ellison's decision-making style—highly centralized, intuition-driven, willing to override others' concerns—works well when his instincts are correct but creates organizational brittleness if his judgment falters. Oracle has no obvious successor with Ellison's technical depth, business instincts, and willingness to make massive bets.
The company's co-CEOs, Safra Catz and the late Mark Hurd (who died in 2019), were operational executives rather than visionaries. After Hurd's death, Oracle did not appoint a replacement co-CEO, leaving Catz as sole CEO focused primarily on financial management and operations. The Executive Chairman role kept Ellison in charge of strategy and technology direction.
Ellison's immense wealth—fluctuating between $200-400 billion depending on Oracle's stock price—insulates him from pressure that might force other executives to retire or cede control. He doesn't need Oracle financially. His motivation appears purely competitive: beating rivals, proving skeptics wrong, and cementing his legacy as one of technology's transformative figures.
That competitive drive manifests in Ellison's relationships with other tech leaders. His friendship with Elon Musk, whom Ellison supported in Tesla's early challenges and later in the Twitter acquisition, reflected shared iconoclast tendencies and willingness to make contrarian bets. Ellison served on Tesla's board from 2018 to 2022 and invested $1 billion in the company when many analysts doubted Tesla's viability.
Ellison's rivalry with AWS founder Jeff Bezos, though rarely discussed publicly, shaped Oracle's cloud strategy. AWS's dominance represented a personal affront to Ellison, who had built Oracle into a $200 billion revenue enterprise software company only to be surpassed by Amazon in cloud computing. Oracle's aggressive AI infrastructure buildout can be understood partly as Ellison's determination to beat Amazon in at least one major category.
Ellison's relationship with Jensen Huang, NVIDIA's CEO, proved crucial to Oracle's AI ambitions. According to people familiar with the dynamics, Ellison cultivated Huang through mutual respect for each other's technical accomplishments and Oracle's willingness to commit enormous GPU purchase volumes. When NVIDIA faced allocation decisions about which customers received priority access to limited H100 and Blackwell supply, Oracle's multi-billion-dollar commitments ensured favorable treatment.
At 81, Ellison maintains a lifestyle that few octogenarians could sustain. He is an avid sailor, winning multiple America's Cup championships through his Oracle Team USA. He pilots jets, practices martial arts, and works 70+ hour weeks when major initiatives require his attention. Friends describe him as viewing aging as an engineering problem to be solved through optimization of diet, exercise, and eventually biotechnology.
That technologist's approach to longevity connects to Ellison's public statements about AI's potential for medical breakthroughs. During the Stargate announcement, Ellison described AI-designed mRNA cancer vaccines as a near-term application, suggesting personal interest in life-extension technologies. Oracle's acquisition of health IT company Cerner for $28 billion in 2022—one of Oracle's largest acquisitions ever—reflected Ellison's belief that AI would transform healthcare and Oracle should participate in that transformation.
But Ellison's age and personal wealth concentration also create risks for Oracle. If Ellison's health deteriorates or he dies unexpectedly, Oracle would face simultaneous leadership and ownership transitions. His 41% stake would pass to trusts and foundations, potentially fragmenting ownership. The company's strategy, heavily dependent on Ellison's risk appetite and willingness to invest tens of billions in speculative bets, might shift dramatically under different leadership.
Some investors and analysts view Ellison's continued control as Oracle's greatest risk. They argue the company needs younger leadership, more distributed decision-making, and strategic patience rather than all-in bets. Others counter that Oracle's AI opportunity exists precisely because Ellison is willing to make commitments that a more cautious CEO or management-by-committee would never approve.
Oracle's AI infrastructure bet ultimately must be judged by financial results. Can the company generate returns that justify the tens of billions in capital expenditures and build a sustainably profitable cloud business?
The financial picture for Oracle's cloud business showed strong growth but from a small base. In fiscal Q2 2025 (ending November 30, 2024), Oracle reported cloud infrastructure revenue grew 52% year-over-year to approximately $2.4 billion for the quarter, or about $9.6 billion annualized. Cloud applications and infrastructure combined reached roughly $6 billion for the quarter, or $24 billion annualized.
Oracle's Remaining Performance Obligations (RPO)—contracted revenue not yet recognized—surged to $455 billion in Q1 fiscal 2026, up from $80 billion in Q3 fiscal 2024. This 359% increase reflected major contracts including the OpenAI deal and various sovereign cloud commitments. RPO provided visibility into future revenue but didn't guarantee profitability or cash flow timing.
The company's capital expenditures told a different story. Oracle spent approximately $6.9 billion in fiscal 2024, then tripled that to $21.2 billion in fiscal 2025, with plans for $35 billion in fiscal 2026. Over three years, Oracle would invest $60-70 billion in infrastructure—approaching the company's total revenue for an entire year.
This created a profitability challenge. Cloud infrastructure businesses typically required 3-5 years before datacenters became profitable, as capital costs depreciated over time while revenue grew. Oracle's accelerating CapEx meant the company was continuously building new datacenters that wouldn't contribute positive cash flow for years, even as older datacenters began generating profits.
Oracle's overall operating income remained healthy at approximately $15-16 billion annually, but that profitability came primarily from the legacy database and applications business—the exact segments that were slowly declining or growing single digits. The cloud business, despite rapid revenue growth, consumed capital faster than it generated profits.
Analysts questioned whether Oracle's unit economics on AI infrastructure were sustainable. The OpenAI deal, reportedly priced aggressively to win the contract, might generate gross margins of only 10-20% compared to the 70-80% gross margins on Oracle's database business. If Oracle's new cloud revenue came at dramatically lower margins than its existing business, overall profitability could decline even as revenue grew.
Oracle's defense was that scale would improve economics over time. As datacenters filled with customer workloads and capacity utilization increased, gross margins would expand. Additionally, Oracle aimed to upsell cloud infrastructure customers on higher-margin database and application services, creating bundled revenue streams with blended margins superior to infrastructure alone.
The company's stock performance reflected investor uncertainty. Oracle shares traded around $140-160 in late 2024, giving the company a market capitalization near $450 billion. That valuation implied investors believed Oracle would successfully transform into a cloud and AI infrastructure powerhouse, but the stock's volatility—sometimes gaining or losing 10%+ on earnings announcements—showed ongoing debate about whether the transformation would succeed.
In September 2025, Oracle reported quarterly results that beat analyst expectations, sending shares up 12% in a single day and briefly making Ellison the world's richest person. But days later, skepticism about AI infrastructure economics caused Oracle stock to drop 8%, wiping out $34 billion from Ellison's net worth. These swings illustrated how Oracle's valuation depended on investor confidence in the AI infrastructure thesis.
For context, consider Oracle's peer valuations. Amazon (including AWS) traded at approximately 3x revenue. Microsoft traded at 12x revenue given its software business mix. Google traded at 6x revenue. Oracle at $450 billion market cap and $53 billion revenue traded at roughly 8.5x revenue—a premium to Amazon, suggesting investors valued Oracle's growth potential, but below Microsoft, suggesting concerns about Oracle's ability to sustain high-margin business mix.
If Oracle successfully grew cloud infrastructure revenue to $40-50 billion annually by 2028 while maintaining high-margin database business, the company's revenue could reach $80-90 billion with blended margins supporting operating income of $25-30 billion. At historical enterprise software multiples of 15-20x earnings, that could justify a $400-600 billion market capitalization.
But if cloud infrastructure growth stalled at $20-25 billion annually, dragging overall revenue growth to low single digits while CapEx consumed cash, Oracle could face margin compression and stagnant valuation. The company's stock could remain range-bound around $140-180 for years, underperforming tech sector peers.
The financial stakes explained Ellison's urgency. At 81, he likely had 5-10 years to cement his legacy and prove Oracle's transformation. A successful AI infrastructure buildout that established Oracle as a $80-100 billion revenue company would validate his lifetime's work. A failed bet that left Oracle saddled with underutilized datacenters and declining legacy revenue would tarnish his reputation.
Larry Ellison's $500 billion AI infrastructure bet represents the culmination of a career defined by aggressive competition, technical vision, and refusal to accept conventional wisdom about what Oracle could achieve.
The bet's logic is sound. AI infrastructure requirements differ fundamentally from traditional cloud computing, creating an opportunity for Oracle to compete on dimensions—sovereign control, massive scale, specialized performance—where its disadvantages matter less. Governments and enterprises genuinely need AI capabilities that AWS, Azure, and Google Cloud struggle to provide. Oracle's willingness to invest tens of billions in pursuit of that market creates real differentiation.
The execution risks are also real. Building gigawatt-scale datacenters, securing millions of GPUs, filling capacity with profitable customers, and operating at AWS-level reliability all require capabilities Oracle hasn't fully demonstrated. The company's history of over-promising and under-delivering on cloud initiatives creates skepticism about whether this time will be different.
Three scenarios seem plausible:
In the optimistic scenario, Stargate succeeds in becoming the infrastructure backbone for OpenAI and multiple other AI labs. Oracle's sovereign cloud offerings win major government contracts across the US, EU, Middle East, and Asia. By 2028, Oracle's cloud business generates $50+ billion annually, with blended margins supporting strong profitability. Oracle stock reaches $250-300, making Ellison briefly the world's richest person and validating his late-career transformation strategy.
In the pessimistic scenario, AI infrastructure exhibits winner-take-most dynamics favoring AWS and Azure. OpenAI's requirements prove less than anticipated as model scaling hits diminishing returns. Sovereign cloud demand materializes slowly, and governments prove reluctant to commit billions to Oracle infrastructure. By 2028, Oracle's cloud business reaches only $25-30 billion while CapEx consumed $70+ billion, creating questions about return on investment. Oracle stock stagnates, and the company faces pressure to reduce CapEx and refocus on profitable legacy business.
In the mixed scenario, Oracle carves out a sustainable niche in AI infrastructure without becoming a dominant player. Cloud revenue reaches $35-40 billion by 2028, growing faster than legacy business decline but insufficient to transform Oracle's overall growth profile. The company becomes profitable on cloud infrastructure but at margins below legacy software. Oracle stock performs in line with broader tech indices, neither vindicating nor discrediting Ellison's strategy.
Which scenario materializes will depend on factors partially outside Oracle's control: AI model scaling trends, government policy on data sovereignty, NVIDIA's GPU production capacity, competition from AWS/Azure/Google, and macroeconomic conditions affecting infrastructure spending.
But it will also depend on execution—Oracle's ability to build datacenters on time, operate them reliably, win customer trust, and deliver on promises. This is where Ellison's personal involvement matters most. At 81, he cannot personally manage every datacenter deployment or customer negotiation. But his willingness to commit Oracle's resources, override cautious advisors, and bet the company's future on infrastructure transformation creates possibility that wouldn't exist under conventional management.
For Ellison, the AI infrastructure bet is fundamentally personal. He built Oracle from a three-person startup to a global software powerhouse. He survived near-bankruptcy in 1990, the dot-com crash, the cloud transition that Oracle initially missed, and countless competitive battles. Now, in his eighties, he's making one final bet to ensure Oracle remains relevant in the AI era.
Success would cement Ellison's legacy as one of technology's all-time great business builders—someone who not only created a major company but successfully transformed it across multiple technological paradigms. Failure would be a footnote to an already legendary career, but one Ellison clearly wants to avoid.
The question is not whether Ellison can dream big or commit capital. At $500 billion, Stargate proves he can do both. The question is whether Oracle can execute at a scale and speed it has never achieved before, in markets where it lacks incumbency advantages, against competitors with deeper resources and operational expertise.
Larry Ellison has spent a lifetime proving skeptics wrong. His AI infrastructure bet is asking the world to bet on him one more time. By 2029, we'll know whether the 80-year-old's final gamble paid off—or whether even Ellison's legendary competitiveness couldn't overcome Oracle's structural disadvantages in the cloud wars.
One thing is certain: Ellison won't go quietly. If Oracle succeeds, he'll remind everyone who doubted him. If Oracle struggles, he'll double down and find someone to blame. That relentless competitiveness, more than any technology or strategy, defines both Ellison's career and Oracle's culture. For better or worse, Oracle's AI infrastructure future will be decided by an 80-year-old billionaire who still believes he can outwork, out-think, and out-execute anyone in tech.
The AI revolution will determine if he's right.
## Continue reading - [100 Most Influential People in AI: 2025 Power List](https://digidai.github.io/2025/11/07/silicon-valley-ai-100-most-influential-2025/) - [Sam Altman: OpenAI CEO & AGI Race Leader](https://digidai.github.io/2025/11/08/sam-altman-openai-comprehensive-deep-analysis/) - [Satya Nadella: Microsoft](https://digidai.github.io/2025/11/14/satya-nadella-microsoft-ceo-ai-transformation-deep-analysis/) - [Jensen Huang: NVIDIA](https://digidai.github.io/2025/11/15/jensen-huang-nvidia-ai-chip-kingmaker-deep-analysis/)