Oracle Cloud Infrastructure

Oracle Cloud Infrastructure (OCI) is the public cloud of Oracle Corporation. Oracle made it generally available on October 20, 2016 under the name Oracle Bare Metal Cloud Services, a rebuilt successor to the company's first infrastructure cloud that Oracle later marketed as its "Generation 2" cloud.^[1][2] Since 2024 OCI has become a major supplier of AI training and inference capacity, with contracts from OpenAI (including sites for the Stargate Project), Meta, xAI and others; in February 2026 Oracle named AMD, Meta, NVIDIA, OpenAI, TikTok and xAI among its largest OCI customers.^[3] For the quarter ended August 31, 2026, Oracle reported $7.4 billion of cloud infrastructure revenue, up 121% from a year earlier, said it had delivered 850 megawatts of additional data center capacity in the quarter and more than 300,000 GPUs to AI customers since the end of May, and reported $664 billion of remaining performance obligations.^[4] Free cash flow was negative $23.7 billion in fiscal 2026 as Oracle invested in the cloud infrastructure business, and borrowings stood at about $125 billion at the end of August 2026.^[5][4]

History

First attempt and the Gen 2 rebuild

Oracle's first infrastructure cloud did not last. Speaking to investors at Oracle CloudWorld in October 2022, Larry Ellison said he had "actually canceled the project" after a fight with Oracle's leadership, because he believed the company was "copying what the other guys were doing."^[2] A new engineering team designed the replacement. Clay Magouyrk, who joined Oracle from Amazon Web Services in 2014, was a founding member of that team and later oversaw what Oracle calls Gen2 OCI.^[8]

Oracle previewed the new service at OpenWorld in September 2016 and made it generally available on October 20, 2016 as Oracle Bare Metal Cloud Services. It opened in a single US-Southwest region made up of three fault-independent availability domains, offering bare-metal servers, block and object storage, identity management, VPN connectivity and a software-defined virtual cloud network.^[1] The service was later renamed Oracle Cloud Infrastructure. At OpenWorld 2018 Oracle presented its "Generation 2" architecture, in which customer workloads run on bare-metal servers while the cloud's control software runs on separate computers; Oracle said the split meant it could not see customer data.^[2]

Regions and deployment models

In October 2026 Oracle's region page listed more than 50 public cloud regions in 28 countries, each offering what Oracle describes as a consistent set of more than 200 services. The same page listed separate US government and Department of Defense regions, UK and Australian government regions, and two EU Sovereign regions in Frankfurt and Madrid. Oracle groups its regions into isolated "realms" that share no physical infrastructure, accounts or network connections.^[7] In December 2025 Magouyrk put the wider total at "over 211 live and planned regions worldwide."^[9]

Oracle also installs OCI inside customer and partner facilities. Its fiscal 2026 annual report lists OCI Dedicated Region (a self-contained OCI instance in a customer's data center), OCI Sovereign Cloud, Oracle Alloy (which lets partners sell OCI services under their own commercial terms), Exadata Cloud@Customer, Compute Cloud@Customer and Roving Edge Infrastructure.^[10] Oracle said in March 2025 that a Dedicated Region could start as small as three racks.^[11] In June 2025 Ellison said Oracle had 29 Cloud@Customer dedicated datacenters live and another 30 to be built in fiscal 2026.^[12]

Multicloud database partnerships

Under Oracle Database@Azure, @Google Cloud and @AWS, Oracle places OCI hardware in the data centers of Microsoft Azure, Google Cloud and Amazon Web Services and runs its Exadata and Autonomous Database services there. By September 2024 the Azure service was running in six Azure regions, the Google Cloud service had become generally available in four Google Cloud regions, and the AWS service had been announced at Oracle CloudWorld.^[13] Multicloud database revenue grew 1,529% in the quarter ended August 31, 2025, and Ellison said Oracle would deliver another 37 datacenters to its three hyperscaler partners, for a total of 71.^[14] By October 2026, under the names Oracle AI Database@AWS, @Azure and @Google Cloud, the services were listed as live in 22 AWS regions, 33 Azure regions and 15 Google Cloud regions.^[7]

AI infrastructure

What is an OCI Supercluster?

OCI Supercluster is Oracle's name for large clusters of bare-metal GPU servers joined by a remote direct memory access (RDMA) network. Oracle's large GPU shapes are bare-metal instances, which Oracle says deliver performance without hypervisor overhead.^[15] In September 2024 Oracle published maximum Supercluster sizes for each GPU type: 32,768 A100 GPUs, 16,384 H100 GPUs (up to 65 exaFLOPS), 65,536 H200 GPUs (260 exaFLOPS), 131,072 B200 GPUs and more than 100,000 GPUs in GB200 NVL72 systems. It called the Blackwell configuration, rated at up to 2.4 zettaFLOPS of AI compute, "the world's first zettascale AI supercomputer" offered in a cloud.^[16] The cluster network ran RDMA over Converged Ethernet (RoCE v2) on NVIDIA ConnectX-7 network cards, while GB200 NVL72 clusters were designed around NVIDIA Quantum-2 InfiniBand.^[16]

NVIDIA GPU generations

OCI has offered successive NVIDIA data center GPU generations since the A100. The table lists the dates Oracle or NVIDIA gave and the largest cluster sizes Oracle has stated.

NVIDIA said on April 28, 2025 that Oracle had stood up its first wave of liquid-cooled GB200 NVL72 racks, with thousands of Blackwell GPUs being deployed over Quantum-2 InfiniBand and Spectrum-X Ethernet networking.^[19] OpenAI said Oracle began delivering GB200 racks to the Abilene campus in Texas in June 2025.^[21] In March 2025 Oracle said it was among the first clouds taking orders for Blackwell Ultra systems, including GB300 NVL72 racks.^[11]

AMD Instinct

OCI added AMD accelerators on September 26, 2024, when bare-metal MI300X instances (BM.GPU.MI300X.8, eight GPUs with 192 GB each) became generally available, with Supercluster support for up to 16,384 GPUs.^[15] MI355X instances (BM.GPU.MI355X.8, 288 GB per GPU and a 3,200 Gb/s cluster network) followed on October 14, 2025. Oracle said it was the first hyperscaler to offer the MI355X publicly and that its Supercluster scales to 131,072 GPUs.^[22] The same day Oracle and AMD said OCI would be the launch partner for the first publicly available supercluster built on AMD's Instinct MI450 series, starting with 50,000 GPUs in the third calendar quarter of 2026 and expanding in 2027. The planned clusters use AMD's "Helios" rack design with "Venice" EPYC processors and "Vulcano" networking.^[23]

In December 2025 Oracle described a change in its chip policy. After selling its stake in the chip designer Ampere for a $2.7 billion pre-tax gain, Ellison said Oracle was "committed to a policy of chip neutrality" and would keep buying NVIDIA's latest GPUs while being ready "to deploy whatever chips our customers want to buy."^[9]

Acceleron networking and Zettascale10

On October 14, 2025 Oracle announced OCI Zettascale10, which joins GPUs in several data centers within a two-kilometer radius into multi-gigawatt clusters. Oracle gave a peak of 16 zettaFLOPS and said initial deployments would target up to 800,000 NVIDIA GPUs.^[24] Its network, Oracle Acceleron RoCE, uses the switching built into GPU network cards to connect each card to several physically separate network planes, so traffic can move off a failing plane without forcing a training job to restart from a checkpoint.^[24] Oracle said Zettascale10 was the fabric under the Stargate supercluster in Abilene, and that it was taking orders for deployments to be available in the second half of 2026.^[24]

Major AI contracts

OpenAI and Stargate

OpenAI's first announced OCI agreement came in June 2024, when Oracle, Microsoft and OpenAI agreed to extend the Microsoft Azure AI platform onto OCI to give OpenAI more capacity. Oracle's announcement listed Adept, Modal, MosaicML, NVIDIA, Reka, Suno, Together AI, Twelve Labs and xAI among other OCI Supercluster users.^[25] Oracle was one of the companies that announced the $500 billion Stargate commitment at the White House in January 2025.^[26]

On June 30, 2025, Oracle disclosed in a regulatory filing that it had signed several large cloud agreements, including one "expected to contribute more than $30 billion in annual revenue starting in FY28." It did not name the customer.^[27] On July 22, 2025, OpenAI said it had agreed with Oracle to develop 4.5 gigawatts of additional Stargate capacity in the United States, which together with Abilene would bring Stargate capacity under development to more than 5 GW running more than 2 million chips. OpenAI described Stargate as spanning data center partnerships with Oracle, SoftBank and the neocloud CoreWeave.^[21]

On September 23, 2025, OpenAI named three new Oracle-developed sites in Shackelford County, Texas, Doña Ana County, New Mexico, and the Midwest, which with a potential 600 MW expansion near Abilene could deliver more than 5.5 GW.^[26] Oracle's June 2026 results presentation showed five data center campuses: Abilene and Shackelford in Texas, Doña Ana in New Mexico, Saline in Michigan and Port Washington in Wisconsin.^[28] The 1 GW Saline Township campus broke ground on June 1, 2026, with Oracle, Related Digital and Walbridge as OpenAI's partners.^[29] By September 10, 2026, Magouyrk said six of Abilene's eight buildings, 618 MW and 75% of its capacity, had been delivered.^[30]

How much is the OpenAI contract worth?

On September 9, 2025, Oracle reported that it had signed four multibillion-dollar contracts with three customers in the quarter, lifting RPO to $455 billion.^[14] The next day The Wall Street Journal, citing people familiar with the deal, reported that OpenAI had agreed to buy about $300 billion of computing capacity from Oracle over roughly five years starting in 2027.^[32] Oracle's releases and filings have not stated that figure; they attribute the RPO increase to "certain significant cloud contracts" without naming customers or values.^[10] OpenAI, however, described the partnership's size: its September 23, 2025 announcement said the July 4.5 GW agreement "represents a partnership that exceeds $300 billion between the two companies over the next five years," without confirming the start date or payment schedule the Journal reported.^[26]

Meta, xAI and other customers

In December 2024 Ellison said Oracle had signed an agreement for Meta to use OCI and to work with Oracle on AI agents built on Meta's Llama models.^[18] At Oracle's financial analyst meeting on October 16, 2025, Magouyrk said OCI had booked $65 billion of new commitments within a month, "literally seven deals, four customers, all of them other than OpenAI," and Reuters reported that the total included a $20 billion deal with Meta.^[33] Oracle's December 2025 results credited part of that quarter's $68 billion RPO increase to "new commitments from Meta, NVIDIA, and others."^[9]

xAI was renting NVIDIA chips on OCI by 2024. In July 2024 The Information reported that talks to expand the arrangement into a deal worth about $10 billion had ended, partly over Elon Musk's timeline and Oracle's concern that xAI's preferred site lacked enough power; Musk said xAI would build its own H100 system.^[34] In June 2025 Oracle began offering xAI's Grok models through the OCI Generative AI service and said xAI would use OCI to train and serve its next-generation Grok models.^[35]

The list of AI compute deals compares these contracts with those of other providers.

Financials

Revenue growth

Oracle reports OCI within "cloud infrastructure (IaaS)" revenue, which also includes its cloud database services. In the quarter ended May 31, 2026 that line was $5.8 billion, of which $4.8 billion was CPU and GPU infrastructure (up 119%) and $0.8 billion was database.^[28] Cloud infrastructure revenue rose from $10.2 billion in fiscal 2025 to $18.1 billion in fiscal 2026, an increase of 77% that matched the forecast Safra Catz gave in September 2025.^[5][14] In the first quarter of fiscal 2027 it reached $7.4 billion, about 38% of Oracle's $19.3 billion of total revenue.^[4]

Oracle has also published long-range targets. In September 2025 Catz forecast OCI revenue of $18 billion in fiscal 2026, then $32 billion, $73 billion, $114 billion and $144 billion over the next four years, and said most of that revenue was already booked in RPO.^[14] At the October 2025 analyst meeting Oracle raised its fiscal 2030 cloud infrastructure target to $166 billion and said it expected adjusted gross margins of 30% to 40% on AI infrastructure.^[33]

Remaining performance obligations

RPO is contracted revenue that Oracle has not yet recognized, across all of its businesses. It rose from $138 billion at May 31, 2025 to $638 billion a year later; Oracle expected to recognize about 12% of the May 2026 balance within 12 months, 34% in months 13 to 36 and 34% in months 37 to 60.^[10] Oracle said most of the increase in its third and fourth fiscal 2026 quarters came from large AI contracts in which customers prepaid for GPUs or supplied their own, and that the prepaid and customer-supplied hardware in those contracts totaled $75 billion.^[5] In the quarter ended August 31, 2026 it booked more than $30 billion of additional AI cloud contracts and said their structure would not change its plans to raise capital.^[4]

Capital expenditure and free cash flow

Oracle's capital expenditures, part of the industry's wider AI capital spending, rose roughly eightfold between fiscal 2024 and fiscal 2026, and Oracle attributes its negative free cash flow to investment in the cloud infrastructure business.^[5] Oracle reports both GAAP capital expenditures and a non-GAAP "net cash outlay," which subtracts short-term financing tied to equipment and customer prepayments used for capital spending.^[28]

Oracle had guided to about $50 billion of fiscal 2026 capital spending by March 2026; the GAAP figure for the year came to $55.7 billion.^[36][5] In its June 2026 results presentation Oracle put fiscal 2027 net cash outlay at about $70 billion, and on September 10, 2026 chief financial officer Hilary Maxson said Oracle expected $90 billion to $95 billion of capital expenditures "with not more than $70 billion in net cash CapEx."^[28][30]

Debt and equity financing

Oracle sold $18 billion of bonds in September 2025.^[37] On February 1, 2026 it said it planned to raise $45 billion to $50 billion of gross proceeds during calendar 2026 to build capacity for contracted OCI demand: about half from equity, through mandatory convertible preferred stock and an at-the-market share program of up to $20 billion, and half from a single investment-grade bond sale, with no further bonds that year.^[3] Within days it raised $30 billion through bonds and the convertible preferred.^[36] For fiscal 2026 as a whole Oracle raised $43 billion of debt and $5 billion of equity, and it expected to raise about $40 billion in fiscal 2027, including the $20 billion share program, without issuing more debt in calendar 2026.^[5] It completed the $20 billion share sale in the quarter ended August 31, 2026.^[4]

Notes payable and other borrowings rose from $92.6 billion at May 31, 2025 to $129.5 billion at May 31, 2026, and stood at $125.3 billion on August 31, 2026.^[5][4] Leases add a second layer of commitments. At May 31, 2026 Oracle had $260 billion of lease commitments that had not yet started, substantially all for data centers, expected to begin between fiscal 2027 and fiscal 2029 with terms of 15 to 19 years; they included a lease for which Oracle guaranteed up to $3.3 billion of the landlord's borrowing.^[10]

GPU list prices

Oracle publishes pay-as-you-go (on-demand) list prices per GPU-hour; the prices in its large AI contracts have not been disclosed. When bare-metal MI300X instances became generally available on September 26, 2024, Oracle listed them at $6.00 per GPU-hour.^[15] MI355X instances launched on October 14, 2025 at $8.60, which Oracle's price list states per GPU-hour.^[22][20] The table gives the GPU list prices in Oracle's public price-list data as updated on September 1, 2026. NVIDIA AI Enterprise software is billed separately per GPU-hour, for example $2.50 for H100 and H200 and $4.00 for GB200.^[20] Other providers' rates are compared in cloud GPU pricing comparison.

Leadership and organization

Oracle's board promoted Magouyrk and Mike Sicilia to co-chief executive officers effective September 22, 2025, and Safra Catz, chief executive since 2014, became executive vice chair. Magouyrk, then 39, had been president of OCI since June 2025 and an executive vice president since December 2019; before joining Oracle in 2014 he was a senior engineer at Amazon and AWS. Sicilia, 54, had been president of Oracle Industries. Magouyrk received stock options to buy $250 million of Oracle shares and Sicilia $100 million, with 20% of each grant tied to revenue targets.^[8] Ellison remains chairman and chief technology officer.^[6] He has announced several OCI decisions himself, including the Meta agreement in December 2024 and the chip-neutrality policy a year later.^[18][9]

Within OCI, Oracle's executive page in October 2026 listed Mahesh Thiagarajan as "Executive Vice President and Head of Oracle Cloud Infrastructure."^[6] He wrote Oracle's September 2024 zettascale announcement as executive vice president for security and developer platform, and Oracle's October 2025 announcements identified him as executive vice president of OCI.^[16][24] Other OCI executives in Oracle announcements include Karan Batta, senior vice president, who spoke for the multicloud program in 2024, and Henry Jerez, senior vice president for AI infrastructure in 2025.^[13][11] Hilary Maxson is Oracle's chief financial officer.^[6]

Criticism and risks

Dependence on OpenAI

On July 9, 2026 S&P Global Ratings cut Oracle's credit rating to BBB-, one notch above junk, and estimated that roughly half of Oracle's $638 billion of RPO was tied to OpenAI; S&P also said Oracle's free cash flow deficit could widen to $42 billion.^[38] That estimate is S&P's, not a figure Oracle has disclosed. Oracle's annual report warns: "In certain OCI offerings, we are more concentrated among a number of large customers, which could increase these risks." It adds that some customers "may be highly leveraged."^[10] Magouyrk addressed the point in October 2025, saying "we think OpenAI is a great customer, but we have many customers."^[33]

Margins

On October 7, 2025 The Information reported that Oracle's rentals of servers with NVIDIA chips produced $125 million of gross profit, a margin of just under 14%, in the three months ended August 2025, and that Oracle lost nearly $100 million renting Blackwell chips; Oracle shares fell more than 5% after the report.^[39] Oracle's 30% to 40% target for AI infrastructure gross margins came nine days later.^[33] In September 2026 Maxson said she expected a step down in gross margins in fiscal 2027, describing infrastructure as a lower gross margin business.^[30] Oracle's annual report adds that it has at times accepted less favorable terms from GPU and memory suppliers to secure supply, and that the useful lives over which it depreciates this equipment could be shortened, an issue covered in GPU depreciation.^[10]

Financing and debt

A proposed class action filed in New York state court on January 14, 2026 by buyers of the September 2025 bonds alleged that Oracle failed to disclose that it needed to sell significant additional debt. The bondholders cited $38 billion of loans arranged seven weeks after the bond sale to fund two data centers for the OpenAI agreement; Oracle did not immediately respond to Reuters.^[37] The scale of Oracle's borrowing and lease obligations is set out under Financials above.

Power

Oracle's annual report says it has "faced, and may continue to face, challenges with securing reliable and cost-effective power sources" for its data centers, because energy for AI computing is constrained worldwide.^[10] Power has shaped deals before: in 2024 one reported obstacle to the larger xAI contract was Oracle's view that xAI's preferred location had inadequate power.^[34] Oracle said in June 2026 that it was building data centers intended to generate electricity from natural gas fuel cells.^[5]

Construction timeline

On December 12, 2025 Bloomberg News reported that Oracle had pushed back the completion of some data centers for OpenAI from 2027 to 2028 because of labor and material shortages. Oracle spokesperson Michael Egbert said that "there have been no delays to any sites required to meet our contractual commitments, and all milestones remain on track."^[40] Oracle's annual report notes that it relies on third parties to build some capacity and that their failure to obtain financing or finish construction on schedule could delay it.^[10] In September 2026 Bloomberg reported that Oracle had sent a force majeure notice to Stack Infrastructure, the developer of the New Mexico campus; Oracle said the project "remains on our planned schedule" and that such notices "do not, by themselves, establish a project delay."^[31] Earlier that month Magouyrk had told analysts that the New Mexico site was "definitely on track."^[30]

References

1. ^1 ^2 ^3"Oracle Announces General Availability of Bare Metal Cloud Services." Database Trends and Applications, 2016-10-20. dbta.com/...ty-of-Bare-Metal-Cloud-Services-114301
2. ^1 ^2 ^3"Larry Ellison fought internal battle to kill Oracle's first-generation cloud." The Register, 2022-10-24. theregister.com/...ellison_oracle_cloud
3. ^1 ^2 ^3 ^4"Free Writing Prospectus: Oracle calendar year 2026 funding plan." Oracle Corporation (SEC filing), 2026-02-01. sec.gov/...d36096dfwp
4. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11"Oracle Announces Q1 Results Driven by Triple Digit Growth in Cloud Infrastructure Revenues (Form 8-K, Exhibit 99.1)." Oracle Corporation, 2026-09-10. sec.gov/...orcl-ex99_1
5. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13"Oracle Announces Record Q4 and FY 2026 Results Driven by Cloud Infrastructure & Cloud Applications (Form 8-K, Exhibit 99.1)." Oracle Corporation, 2026-06-10. sec.gov/...orcl-ex99_1
6. ^1 ^2 ^3 ^4"Executive Leadership." Oracle (page as archived by the Internet Archive), 2026-10-07. oracle.com/...executives
7. ^1 ^2 ^3"Public Cloud Regions and Data Centers." Oracle (page as archived by the Internet Archive), 2026-10-14. oracle.com/...public-cloud-regions
8. ^1 ^2"Form 8-K: Oracle Corporation Announces Promotion of Clay Magouyrk and Mike Sicilia to CEOs; Safra Catz Appointed Executive Vice Chair of the Board of Directors." Oracle Corporation, 2025-09-22. sec.gov/...d921500d8k
9. ^1 ^2 ^3 ^4 ^5"Oracle Announces Fiscal Year 2026 Second Quarter Financial Results (Form 8-K, Exhibit 99.1)." Oracle Corporation, 2025-12-10. sec.gov/...orcl-ex99_1
10. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8"Annual Report on Form 10-K for the fiscal year ended May 31, 2026." Oracle Corporation, 2026-06-22. sec.gov/...orcl-20260531
11. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8"Announcing New AI Infrastructure Capabilities with NVIDIA Blackwell for Public, On-Premises, and Service Provider Clouds." Oracle Cloud Infrastructure Blog, 2025-03-26. blogs.oracle.com/...idia-blackwell-dedicated-alloy
12. ^1 ^2 ^3"Oracle Announces Fiscal 2025 Fourth Quarter and Fiscal Full Year Financial Results (Form 8-K, Exhibit 99.1)." Oracle Corporation, 2025-06-11. sec.gov/...orcl-ex99_1
13. ^1 ^2"Oracle Expands Multicloud Capabilities with AWS, Google Cloud, and Microsoft Azure." Oracle, 2024-09-11. oracle.com/...cloud-and-microsoft-azure-2024-09-11
14. ^1 ^2 ^3 ^4 ^5"Oracle Announces Fiscal Year 2026 First Quarter Financial Results (Form 8-K, Exhibit 99.1)." Oracle Corporation, 2025-09-09. sec.gov/...orcl-ex99_1
15. ^1 ^2 ^3"Announcing General Availability of OCI Compute with AMD MI300X GPUs." Oracle Cloud Infrastructure Blog, 2024-09-26. blogs.oracle.com/...ga-oci-compute-amd-mi300x-gpus
16. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9"Announcing World's Largest, First Zettascale AI Supercomputer in the Cloud." Oracle Cloud Infrastructure Blog, 2024-09-11. blogs.oracle.com/...-ai-supercomputer-in-the-cloud
17. ^"Announcing general availability for OCI Compute bare metal instances powered by NVIDIA H100 GPUs." Oracle Cloud Infrastructure Blog, 2023-09-19. blogs.oracle.com/...bility-oci-compute-nvidia-h100
18. ^1 ^2 ^3 ^4 ^5"Oracle Announces Fiscal 2025 Second Quarter Financial Results (Form 8-K, Exhibit 99.1)." Oracle Corporation, 2024-12-09. sec.gov/...orcl-ex99_1
19. ^1 ^2"Oracle Cloud Infrastructure Deploys Thousands of NVIDIA Blackwell GPUs for Agentic AI and Reasoning Models." NVIDIA Blog, 2025-04-28. blogs.nvidia.com/...cloud-infrastructure-blackwell
20. ^1 ^2 ^3"Oracle Cloud price list API, USD products (payload updated 2026-09-01; Internet Archive copy)." Oracle, 2026-09-04. web.archive.org/...products
21. ^1 ^2 ^3 ^4"Stargate advances with 4.5 GW partnership with Oracle." OpenAI, 2025-07-22. openai.com/...dvances-with-partnership-with-oracle
22. ^1 ^2"Announcing General Availability of OCI Compute with AMD Instinct MI355X GPUs." Oracle Cloud Infrastructure Blog, 2025-10-14. blogs.oracle.com/...availability-of-oci-amd-mi355x
23. ^"Oracle and AMD Expand Partnership to Help Customers Achieve Next-Generation AI Scale." Oracle, 2025-10-14. oracle.com/...-next-generation-ai-scale-2025-10-14
24. ^1 ^2 ^3 ^4"Oracle Unveils Next-Generation Oracle Cloud Infrastructure Zettascale10 Cluster for AI." Oracle, 2025-10-14. oracle.com/...ttascale10-cluster-for-ai-2025-10-14
25. ^1 ^2"OpenAI Selects Oracle Cloud Infrastructure to Extend Microsoft Azure AI Platform." Oracle, 2024-06-11. oracle.com/...crosoft-azure-ai-platform-2024-06-11
26. ^1 ^2 ^3 ^4 ^5 ^6 ^7"OpenAI, Oracle, and SoftBank expand Stargate with five new AI data center sites." OpenAI, 2025-09-23. openai.com/...five-new-stargate-sites
27. ^1 ^2"Form 8-K, Item 7.01 (Regulation FD Disclosure)." Oracle Corporation, 2025-06-30. sec.gov/...d136779d8k
28. ^1 ^2 ^3 ^4 ^5 ^6 ^7"Q4 & Fiscal Year 2026 Earnings (presentation slides)." Oracle Corporation, 2026-06-10. s23.q4cdn.com/...Presentation-Slides-Q4-26.pdf
29. ^1 ^2"Building the infrastructure for the Intelligence Age in Michigan." OpenAI, 2026-06-01. openai.com/...stargate-michigan-data-center
30. ^1 ^2 ^3 ^4 ^5 ^6 ^7"Oracle (ORCL) Q1 2027 Earnings Call Transcript." The Motley Fool, 2026-09-11. fool.com/...-orcl-q1-2027-earnings-call-transcript
31. ^1 ^2"Oracle invokes force majeure on Project Jupiter but says data center still on track." Albuquerque Journal, 2026-09-24. abqjournal.com/...3127907
32. ^1 ^2"OpenAI reportedly on the hook for $300B Oracle Cloud bill." The Register, 2025-09-11. theregister.com/...openai_reportedly_on_the_hook
33. ^1 ^2 ^3 ^4 ^5"Oracle expects cloud infrastructure revenue to be $166 billion in FY30." SRN News (Reuters), 2025-10-16. srnnews.com/...e-revenue-to-be-166-billion-in-fy30
34. ^1 ^2"Musk's xAI, Oracle end talks on $10 billion server deal, the Information reports." Reuters (via Yahoo News), 2024-07-09. malaysia.news.yahoo.com/...is-10-billion-171235073
35. ^"xAI's Grok Models are Now on Oracle Cloud Infrastructure." Oracle, 2025-06-17. oracle.com/...acle-cloud-infrastructure-2025-06-17
36. ^1 ^2 ^3"Oracle Announces Fiscal Year 2026 Third Quarter Financial Results (Form 8-K, Exhibit 99.1)." Oracle Corporation, 2026-03-10. sec.gov/...orcl-ex99_1
37. ^1 ^2"Oracle sued by bondholders over losses tied to AI buildout." Reuters (via Yahoo Finance), 2026-01-15. finance.yahoo.com/...holders-over-losses-183335615
38. ^"Oracle is one notch above junk after S&P downgrade as AI data-centre spending burns through cash." The Next Web, 2026-07-16. thenextweb.com/...-downgrade-ai-spending-junk-risk
39. ^"Oracle tumbles after report that it's lost nearly $100 million from renting out access to Nvidia's Blackwell chips." Sherwood News, 2025-10-07. sherwood.news/...early-usd100-million-from-renting
40. ^"Oracle denies report on OpenAI data center delays." BNN Bloomberg (Reuters), 2025-12-12. bnnbloomberg.ca/...rt-on-openai-data-center-delays