Oracle’s Cloud Infrastructure (OCI) business is experiencing significant growth, particularly driven by demand for AI computing, and is currently exceeding market expectations. This surge highlights a dramatic transformation in Oracle’s strategy, moving from a software-centric company to a major player in the global cloud infrastructure market. The company’s success hinges on ambitious datacenter investments and strategic partnerships with leading AI customers.
Contents
Key Takeaways:
- Oracle’s cloud growth is fueled by massive AI demand, outpacing earlier forecasts.
- Strategic deals with giants like OpenAI and ByteDance are major revenue drivers.
- Oracle has adopted a hybrid datacenter strategy, blending traditional leasing with bold bets on new “AI-first” developers.
- Underappreciated partnerships, particularly with ByteDance in Asia, are establishing new global AI hubs.
- Oracle possesses key cost advantages, notably in networking, contributing to its competitive edge.
Oracle Cloud’s Evolution to an AI Powerhouse
For many years, Oracle was primarily known as a software company, specializing in databases and enterprise resource planning (ERP). It entered the cloud computing race later than rivals like AWS and Azure. While it successfully transitioned its core software offerings to the cloud and built out its OCI Gen2 Cloud starting in 2016, its infrastructure scale remained smaller than the hyperscalers.
A key early investment that proved crucial for Oracle’s current AI success was in High-Performance Computing (HPC). Leveraging expertise gained partly from its acquisition of Sun Microsystems, Oracle developed high-speed networking capabilities like RDMA over Converged Ethernet (RoCEv2) and became an early adopter of Nvidia GPUs, deploying the A100 in 2020. This foundation in HPC positioned Oracle favorably for the explosion of AI workloads.
Historically, Oracle’s datacenter strategy relied more heavily on leasing capacity from partners like Digital Realty, often with shorter contract terms (2-3 years) compared to the hyperscalers’ longer commitments (8-9 years). This reflected its lower scale and a more cautious approach to infrastructure investment relative to revenue.
Chart showing historical annual capital expenditures for SAP and Oracle.
Oracle’s Transformative AI Bet
The rise of generative AI and the surging demand for high-density GPU clusters dramatically shifted the datacenter market, creating a capacity-constrained environment. Securing the large-scale infrastructure needed for AI required hyperscalers to commit to massive, long-term (10+ years) contracts, often for hundreds of megawatts (MW) of power.
Despite not having the same scale as AWS or Azure, Oracle made a bold strategic pivot starting in late 2023. Recognizing the urgent need for capacity to capture AI opportunities, Larry Ellison committed Oracle to leasing over 2 gigawatts (GW) of datacenter capacity in the US alone within a little over a year. This represented an annual expense exceeding OCI’s revenue in fiscal year 2022, a testament to the scale of the bet. Contract lengths surged to 10+ years, mirroring the demands of the AI era.
This aggressive capacity build-out immediately enabled Oracle to book significant contracts for GPU clusters, as reflected in the rapid growth of its Remaining Performance Obligations (RPO). Larry Ellison’s forecast of booking over $130 billion in contracts in the next twelve months further underscores the expected revenue ramp from these infrastructure investments.
Chart showing Oracle's Reported RPO and OCI Revenue from Q2 FY23 to Q3 FY25.
A Winning “Hybrid” Datacenter Strategy
Oracle’s success in this new environment stems from a unique “hybrid” infrastructure strategy. While traditional hyperscalers often combine self-built facilities (typically for large compute clusters in remote areas) with leased capacity (often in metro areas), Oracle initially lacked significant self-build capabilities.
To compete for the largest AI deals requiring GW-scale capacity, which were increasingly being built in areas with abundant power away from traditional metro hubs, Oracle had to innovate. Tier 1 markets were overcrowded, and suitable large sites were scarce and often pre-leased by competitors.
This led Oracle to engage with a new segment of “AI-first” developers, often companies originally from the cryptomining sector that understood power constraints and had access to sites with ample energy. These developers typically served the Neocloud segment with shorter contract needs, struggling to secure funding for the long-term infrastructure required by hyperscalers.
The Abilene Bet: Powering OpenAI’s Ambitions
Oracle saw an opportunity in this gap. Identifying a potential GW-scale site in Abilene, Texas, Oracle signed a 15-year datacenter deal with Crusoe, an “AI-first” developer. At the time, Crusoe had limited experience building datacenters at this scale.
The initial contract in mid-2024 covered an estimated ~220MW, followed by a ~660MW expansion signed in early 2025. This enormous commitment, potentially valued at $15-20 billion over its term, involved paying over a billion dollars per year to a relatively unproven developer. This was a significant risk, not only in terms of the developer’s execution ability but also the credit and duration risk associated with Oracle’s primary customer for this capacity, OpenAI. At the time of the first contract, OpenAI’s annualized revenue was significantly lower than the potential annual cost of this infrastructure.
This bold move was instrumental in making Oracle a key partner in projects often associated with the high-profile OpenAI “Stargate” initiative. While questions remain about the exact structure and financial flow of the Stargate joint venture itself, the physical infrastructure deal in Abilene is very real and is now reflected in Oracle’s financials and RPO growth. This foundational deal positions Oracle for a long-term relationship with a leading AI company, driving future growth potentially through further large-scale deals globally.
Satellite image of a large-scale datacenter construction site.
For more on the proposed Stargate JV, see: https://semianalysis.com/2025/01/23/openai-stargate-joint-venture-demystified/
The Underappreciated ByteDance Partnership and Global Expansion
While the OpenAI relationship garners significant headlines, Oracle’s partnership with ByteDance is arguably the most underappreciated aspect of its current growth story. ByteDance, the parent company of TikTok, is believed to be one of the world’s largest and fastest-growing users of GPUs, potentially rivaling the scale of US hyperscalers, and is currently Oracle’s largest GPU customer.
This partnership extends beyond the US, notably driving the emergence of the Singapore-Johor-Batam hub as a major global AI datacenter center, potentially the world’s second largest. Most of Oracle’s AI capacity in the ASEAN region is thought to be dedicated to ByteDance. Oracle has partnered with developers like GDS International (rebranded as DayOne), known for rapidly building datacenters, to support ByteDance’s aggressive expansion plans in ASEAN, Europe, and Latin America.
ByteDance’s rapid global deployment creates opportunities for GPU cloud providers like Oracle, as the company often contracts capacity with local partners when entering new regions. This international ramp, powered by under-the-radar partnerships with efficient developers, represents a significant ongoing growth engine for Oracle’s cloud business, distinct from and potentially larger than its known commitments in the US.
Oracle’s Cost Advantages in the AI Cloud Race
In the competitive AI cloud market, Oracle competes with established hyperscalers and emerging “Neocloud” providers. Oracle has demonstrated the ability to be both faster and cheaper than hyperscalers in securing large AI contracts, who may prioritize higher-margin opportunities. Compared to Neoclouds, Oracle possesses several key cost advantages.
Firstly, Oracle benefits from networking optimization. Its experience with high-performance networking, including RoCEv2, allows for efficient cluster configurations that contribute to a lower Total Cost of Ownership (TCO) compared to many competitors. This networking capability is a key factor in Oracle’s strong performance in independent GPU cluster ratings.
Chart showing performance results from an independent ClusterMAX rating test.
Secondly, Oracle reportedly works directly with Original Design Manufacturers (ODMs) like Foxconn for its servers, bypassing the additional margin of Original Equipment Manufacturers (OEMs) like Dell or Supermicro, contributing to lower hardware costs.
Finally, Oracle likely benefits from a lower cost of capital for its GPU cloud business. Its investment-grade credit rating and willingness to use debt funding provide a significant advantage over many Neoclouds that face higher debt costs or rely on equity funding for GPU purchases. These combined cost advantages make Oracle’s pricing highly competitive while potentially still generating attractive returns on its large-scale infrastructure investments.
Outlook and Future Considerations
Oracle’s bold pivot and hybrid datacenter strategy have positioned it as a formidable player in the AI cloud market, securing massive contracts with industry leaders. The significant CapEx ramp and RPO growth underscore the scale of its current momentum.
While the focus is heavily on infrastructure build-out and securing large GPU rental contracts, future financial performance will also depend on the ability to upsell higher-margin OCI services to these large customers and the overall sustainability of the GPU business amidst increasing competition. The long-term dynamics of the AI cloud industry, including evolving customer needs and technological advancements, will continue to shape Oracle’s trajectory.
Oracle’s journey from a software giant to an AI infrastructure powerhouse is a compelling case study in strategic adaptation and bold investment in a rapidly changing market. The success of its partnerships and the execution of its global datacenter expansion will be critical to its continued growth.
