Whether you're looking for an angel investor, a growth advisor, or just want to connect — I'm always open to great ideas.
Get in TouchOracle commits $50B in FY2026 capex for AI infrastructure, more than doubling last year's spend, as its cloud backlog surges to $553B. First to deploy NVIDIA Blackwell at scale and part of the $500B SoftBank-OpenAI mega-deal, Oracle is reinventing itself as an AI infrastructure powerhouse.
AI, startups & growth insights. No spam.
TL;DR: Oracle has committed $50 billion in capital expenditure for FY2026 AI infrastructure — more than double its prior-year spend of $21 billion — as its cloud backlog surged to a staggering $553 billion, up from $138 billion just one year ago. Cloud infrastructure revenue grew 84% year-over-year to $4.9 billion, and Oracle became one of the first cloud providers to deploy NVIDIA's Blackwell GPU architecture at scale. The company is also a founding partner in the $500 billion Stargate joint venture alongside SoftBank and OpenAI, positioning itself as a serious contender in the AI infrastructure race. The main counterweight: a total debt load approaching $100–125 billion and a free cash flow deficit that reflects just how expensive this transformation is.
Oracle just committed more capital to AI infrastructure in a single fiscal year than the entire GDP of Croatia, Uruguay, or Luxembourg. For a company that spent decades selling database licenses to Fortune 500 IT departments, this is not an incremental strategy update — it is a full reinvention.
Oracle's Q3 FY2026 earnings release, published on March 10, 2026, confirmed a capital expenditure commitment of $50 billion for the full fiscal year — a figure that would have seemed implausible from a company historically associated with software licensing and database management. For context, Oracle's entire capex in FY2025 was approximately $21 billion. Doubling that figure in a single year represents a categorical shift in corporate strategy, not a marginal adjustment.
The spending is directed primarily at building out Oracle Cloud Infrastructure (OCI) data center capacity to serve AI training and inference workloads. This means GPU-dense compute clusters, high-bandwidth networking, power infrastructure, and liquid cooling systems capable of handling the thermal load of next-generation accelerators. These are not software investments — they are physical assets that take years to commission and carry long depreciation tails.
Chairman Larry Ellison, at 81, has been the primary architect of this strategy, personally negotiating GPU cluster allocations with NVIDIA and infrastructure deals with AI companies including OpenAI. That level of executive-level engagement at hardware procurement reflects how competitive and supply-constrained the AI infrastructure market remains in 2026. Securing enough GPUs at scale is not a procurement exercise — it is a strategic relationship-building effort that happens at the CEO-to-CEO level.
The $50 billion commitment temporarily pushed Oracle's free cash flow into negative territory — estimated at a deficit of around $10 billion for the period — which initially spooked investors. Management's position is that this is a deliberate "land grab" for AI capacity at a moment when customer demand has validated the spend. The remaining performance obligation (RPO) data, discussed below, is their primary evidence that the bet is already paying off.
What the $50 billion buys, concretely: expanded data center campuses across the US, Europe, and Asia; additional GPU cluster deployments at scales of 100,000+ units per facility; power procurement and grid interconnection agreements; and off-balance-sheet lease commitments for facilities expected to come online through fiscal 2028. Oracle itself disclosed $261 billion in off-balance-sheet data center lease commitments — which shows that the $50 billion capex figure is actually the smaller part of the total infrastructure obligation the company is undertaking.
The number that made markets stop and pay attention was not the revenue figure or the earnings beat. It was the backlog: $553 billion in remaining performance obligations (RPO) at the end of Q3 FY2026, up from approximately $138 billion one year earlier — a 325% increase in twelve months.
RPO represents contracted future revenue that Oracle has not yet recognized — work customers have already committed to pay for. This is distinct from pipeline or expressed interest. These are signed contracts. When Oracle reports a $553 billion backlog, it is reporting that customers have legally committed to purchasing that amount of cloud services over the duration of their contracts, which in infrastructure deals typically run three to five years.
The scale of the increase is almost without precedent for a company of Oracle's size. Most enterprise software companies report RPO growth in the range of 10–30% annually; 325% growth in a single year is the kind of number that reflects a step-change in demand, not organic expansion. What caused it: a wave of large-scale AI infrastructure contract signings by hyperscaler-class customers and AI companies who needed dedicated GPU capacity and were willing to make multi-year commitments to secure it.
The backlog also grew sequentially — up $29 billion from the prior quarter alone — which means the pipeline is still accelerating rather than plateauing. Management's forward guidance reflects this: for fiscal year 2027, Oracle raised total revenue guidance to $90 billion, a target that would have been considered wildly optimistic as recently as 2024. The $553 billion RPO makes that guidance plausible on paper, though converting contracted backlog to recognized revenue at that scale across a single fiscal year requires substantial execution.
For investors, the RPO figure resolves a critical concern: whether Oracle's AI infrastructure investment would find paying customers or would result in stranded capacity. A $553 billion backlog is overwhelming evidence of demand. The risk calculus has shifted from "will customers show up" to "can Oracle build fast enough to serve them."
Oracle's transformation from a database and enterprise software company to an AI infrastructure provider is one of the more striking corporate pivots of the current technology cycle. The company built its business over four decades on Oracle Database — the relational database that ran the back-office systems of virtually every large enterprise on earth. That business generated reliable cash flows, high margins, and a loyal (if sometimes resentful) customer base.
The cloud transition was Oracle's first major pivot, and it was widely seen as late and awkward. Oracle's initial cloud offerings competed poorly with AWS, Azure, and Google Cloud on price, breadth, and developer experience. For much of the 2018–2022 period, Oracle Cloud Infrastructure was viewed as a niche option — adequate for Oracle Database migrations, thin on everything else.
What changed the trajectory was the convergence of AI demand with OCI's specific infrastructure advantages. Oracle had invested heavily in high-performance networking — specifically its RDMA-over-Converged-Ethernet (RoCE) cluster networking — that proved exceptionally well-suited to the inter-GPU communication patterns required by large AI training runs. When hyperscaler GPU capacity became constrained in 2023 and 2024, AI companies looking for training capacity discovered that OCI could often deliver GPU clusters faster than the larger providers, at competitive pricing, with network performance that matched or exceeded alternatives.
Oracle's positioning as an "AI infrastructure specialist" rather than a general-purpose cloud provider is also now a differentiation advantage rather than a liability. Enterprises that want a full-service cloud platform with hundreds of managed services tend to prefer AWS or Azure. AI companies that want raw GPU capacity, high-bandwidth networking, and predictable pricing for large training runs increasingly see OCI as purpose-built for their needs. That is a smaller but higher-value market segment, and Oracle has been capturing it rapidly.
The FY2026 transformation is visible in the numbers: cloud infrastructure revenue growing at 84% while the database and license businesses grow in the single digits. The company Larry Ellison built on Oracle 7 and Oracle 8i is now functionally an AI infrastructure company that also sells database software — not the reverse.
One of Oracle's most significant technical milestones in this cycle was being among the first cloud providers to deploy NVIDIA's Blackwell GPU architecture at meaningful scale. NVIDIA's Blackwell platform — the successor to Hopper — represents a generational leap in AI compute performance, and securing early access in volume is a competitive differentiator with direct commercial implications.
Oracle Cloud Infrastructure deployed thousands of NVIDIA Blackwell GPUs in liquid-cooled GB200 NVL72 rack configurations, making them available for agentic AI and reasoning model workloads. The GB200 NVL72 is NVIDIA's highest-density Blackwell configuration — 72 GPUs interconnected by NVLink in a single rack, designed for the massive-scale inference required by next-generation reasoning models like those that underpin systems similar to OpenAI's o-series.
Oracle is also among the first cloud service providers to offer orderability for the next-generation GB300 NVL72 with Blackwell Ultra GPUs — NVIDIA's next architectural step beyond the initial Blackwell release. OCI Superclusters are designed to scale beyond 100,000 NVIDIA Blackwell GPUs, which puts Oracle's maximum cluster size in the range required for frontier model training runs. At that scale, the interconnect fabric and networking become as important as the GPUs themselves — and Oracle's RoCE-based networking has been validated at multi-thousand-GPU scale.
The Blackwell deployment also underscores Oracle's relationship with NVIDIA at the partnership level. Ellison has publicly described his direct engagement with NVIDIA leadership in securing GPU allocation. In a market where GPU availability has been a strategic constraint, being early to Blackwell deployment signals both relationship depth and technical readiness — Oracle had to have its liquid cooling, power delivery, and networking infrastructure ready to absorb these systems before they were widely available.
For customers, Blackwell availability on OCI matters because it directly affects the economics of AI workloads. The Blackwell architecture delivers substantially higher performance per watt than Hopper, which reduces the cost per token for inference and the time-to-completion for training. An AI company that can run its models on Blackwell on OCI before competitors can access the hardware elsewhere has a real competitive advantage — which is why early Blackwell access has become a selling point in Oracle's enterprise sales conversations.
Oracle's AI infrastructure strategy does not exist in isolation — it is embedded in one of the largest and most politically significant technology partnerships ever announced. The Stargate Project, formally announced in January 2025 with considerable White House involvement, is a joint venture among SoftBank, OpenAI, Oracle, and investment firm MGX targeting $500 billion in AI infrastructure investment in the United States through 2029.
Under the Stargate structure, SoftBank holds financial responsibility — meaning it is the primary capital mobilizer — while OpenAI has operational responsibility for the AI workloads the infrastructure will serve. Oracle's role is infrastructure execution: designing, building, and operating the physical data center capacity that Stargate requires. The first Stargate facilities in Abilene, Texas were already operational as of early 2026, with multiple additional campuses under construction or in planning across the US.
The strategic logic of the alliance for Oracle is straightforward: OpenAI is one of the most compute-hungry AI organizations in the world, and a committed infrastructure relationship with OpenAI is effectively a guaranteed demand signal for data center capacity. OpenAI and Oracle have formalized a 4.5-gigawatt partnership to advance the Stargate buildout — a power commitment that represents a significant fraction of the US data center industry's current total capacity.
The Stargate partnership also gives Oracle a geopolitical dimension that pure commercial cloud contracts do not. The project was announced by the President and framed as a national AI competitiveness initiative. That framing gives Oracle a degree of policy protection and government visibility that is strategically valuable as AI infrastructure becomes a contested issue in US-China competition. Being a named participant in the flagship US AI infrastructure initiative is a position Oracle actively cultivated through its Ellison-level relationship investments.
The partnership has not been without friction. Reports emerged of disagreements between OpenAI, Oracle, and SoftBank over governance, control of the data centers, and how responsibilities would be structured. A joint venture at this scale, spanning this many large organizations with their own agendas, was always going to generate internal tension. What matters for Oracle's near-term business is that the underlying demand — OpenAI's need for compute — is real and growing, regardless of how the governance debates resolve.
Oracle's Q3 FY2026 earnings showed cloud infrastructure revenue of $4.9 billion, up 84% year-over-year — a growth rate that significantly outpaces every other segment of Oracle's business and most peers in the cloud infrastructure market. Total cloud revenue (infrastructure plus SaaS) reached $8.9 billion, up 44% year-over-year, with total quarterly revenues hitting a record $17.2 billion, up 22% in constant currency.
The 84% infrastructure growth rate needs context: Oracle's cloud infrastructure business is still smaller in absolute terms than the Big Three. AWS, Azure, and Google Cloud each generate quarterly cloud infrastructure revenue measured in tens of billions. But growth rates tell a different story about trajectory. OCI's 84% growth against AWS's roughly 17% growth, Azure's approximately 31%, and Google Cloud's 28% means Oracle is taking share from a position of momentum, not defending scale.
What is driving the growth: AI training workload demand. Large language model training requires GPU clusters running continuously for weeks or months — producing steady, high-value cloud revenue per contract that differs structurally from the variable, usage-based revenue of typical enterprise cloud workloads. AI companies that sign a contract for a 10,000-GPU cluster for six months of training are committing to a predictable, large revenue stream. Oracle has been successfully competing for these contracts against larger providers.
A secondary driver is enterprise cloud migration — specifically the migration of Oracle Database workloads to OCI. Oracle has used pricing incentives and operational advantages (such as its Autonomous Database and Exadata Cloud Service) to pull Oracle-heavy enterprises onto OCI as their primary cloud platform. These migrations generate infrastructure revenue that is largely captive: an enterprise that runs its Oracle database workloads on OCI has strong economic incentives to consolidate adjacent workloads on the same platform rather than managing cross-cloud data transfer costs and latency.
The forward guidance is bullish: management's $90 billion total revenue target for FY2027 implies continued strong cloud infrastructure growth. At the current trajectory, OCI cloud infrastructure revenue could approach or exceed $25 billion annually by late FY2027 — which would put it in meaningful competition with Google Cloud for the number-three cloud position by revenue.
No honest analysis of Oracle's AI infrastructure strategy is complete without examining the balance sheet cost. Oracle's long-term debt for the quarter ending February 28, 2026 was approximately $100 billion, reflecting a 24% increase year-over-year. More recently, Oracle issued $43 billion in new senior notes, which multiple sources indicate pushed total long-term debt toward the $124–135 billion range depending on measurement methodology.
The debt-to-equity ratio is deeply elevated — Oracle's equity base is relatively thin compared to its debt load, producing debt-to-equity ratios in the 3.7–4x range. Oracle is funding its AI infrastructure buildout primarily through debt issuance rather than retained earnings, which is a rational choice when interest rates are manageable and expected returns on the infrastructure exceed the cost of capital — but which creates meaningful risk if revenue ramp is slower than projected or if capital markets tighten.
The free cash flow picture is the most immediate concern. The $50 billion capex commitment, against Oracle's operating cash generation, has produced a trailing-twelve-month free cash flow deficit of approximately $24 billion. For investors accustomed to Oracle as a cash-generative software business, this is a significant departure. Oracle is spending capital faster than it is generating it — which is sustainable if the backlog converts to revenue on schedule, but creates a funding gap that must be bridged with debt or equity.
Oracle's management has framed the FCF deficit as a temporary consequence of front-loaded infrastructure investment. The argument: data centers built today generate revenue for fifteen or twenty years; the capital is deployed over twelve to twenty-four months but the return period is measured in decades. Additionally, $261 billion in off-balance-sheet data center lease commitments indicates that some of Oracle's infrastructure obligations are structured to remain off the formal balance sheet, which affects how total leverage is measured but not how the underlying obligations affect cash flow.
Analyst opinions are divided. Some, including analysts at Bank of America who argued Oracle had "defused the key risk going into 2026," believe the $553 billion backlog validates the debt-funded expansion and that free cash flow will recover sharply as the backlog converts to recognized revenue. Others point to the $108 billion or higher debt load, the FCF deficit, and the off-balance-sheet lease commitments as a combined leverage position that leaves Oracle with limited margin for error if AI infrastructure demand softens or if construction timelines slip.
Oracle's position in the AI infrastructure market is fundamentally different from the Big Three — and understanding that difference is essential to evaluating whether the $50 billion bet will pay off.
AWS holds approximately 31–32% of the cloud infrastructure market, Azure approximately 23–25%, and Google Cloud approximately 11–13%. Oracle's market share is estimated at roughly 3% — a fraction of the hyperscaler leaders. If the metric is total cloud market share, Oracle is not a hyperscaler competitor. But if the metric is AI infrastructure growth rate and AI workload capture, Oracle's 84% growth against the Big Three's 17–31% growth suggests a very different competitive dynamic.
The structural reason Oracle wins AI infrastructure contracts against larger competitors is a combination of GPU availability, network performance, and pricing. OCI was able to deliver GPU cluster capacity at scale faster than AWS and Azure during periods of Hopper GPU allocation constraint. Oracle's RDMA networking, optimized for the collective communication patterns of distributed AI training, competes effectively on performance per dollar for large training runs. And Oracle has historically been willing to price aggressively to win AI customers who represent high-lifetime-value relationships.
The structural reason Oracle loses general enterprise cloud contracts to AWS and Azure is breadth of services. AWS offers hundreds of managed services — databases, analytics, IoT, serverless, machine learning platforms, CDN, developer tools — that Oracle simply does not match. An enterprise building a cloud-native application from scratch is unlikely to choose OCI unless Oracle Database is a hard dependency or unless GPU availability for AI workloads is the primary use case. Oracle is not competing to be the general-purpose cloud platform for enterprise developers; it is competing to be the preferred GPU infrastructure platform for AI workloads.
That is a large and growing market. AI training and inference spending has become the fastest-growing segment of cloud infrastructure spend, and AI companies are less locked into any single hyperscaler's ecosystem than traditional enterprise customers. An AI startup that trained its first model on AWS will readily migrate to OCI for its second training run if OCI offers better GPU availability or pricing. Customer stickiness in the AI infrastructure market is lower than in the legacy enterprise cloud market — which creates opportunity for Oracle to compete on pure merit for individual workloads rather than fighting against years of existing enterprise relationships.
The question Oracle's management must answer is whether 3% cloud market share can support the leverage required to fund $50 billion in annual capex. The $553 billion backlog suggests the answer is yes — but the proof will be in how efficiently Oracle converts that backlog to revenue over the next eighteen to twenty-four months, and whether it can continue winning new contracts at the pace implied by its guidance.
Alibaba increases T-Head AI chip prices up to 34% and launches the Alibaba Token Hub to dominate the AI token economy. As Fed Chair Powell warns AI data centers are pushing inflation, Alibaba's move signals the global AI cost squeeze is just beginning.
Fed Chair Jerome Powell directly links AI data center buildout to rising inflation. Here's how the $500B AI infrastructure boom is driving up utility costs, affecting interest rates, and hitting consumers.
NVIDIA CEO Jensen Huang projects 7.5M AI agents creating a 100:1 AI-to-human worker ratio. From GTC 2026's IGX Thor launch to the $1.7 GW GPU infrastructure boom, here's what NVIDIA's agent economy vision means.