OpenAI lands on Oracle Cloud: GPT and Codex accessible via existing credits, a $300 billion deal

OpenAI arrives on Oracle Cloud: GPT and Codex accessible via existing credits, a $300 billion deal

🔎 Why OpenAI's arrival on Oracle Cloud changes the game

On June 11, 2026, OpenAI announced the availability of its models on Oracle Cloud Infrastructure (OCI). GPT-4o, GPT-5.5 and Codex are now accessible via Oracle's existing Universal Credits, with no provisioning delays or additional contracts.

This announcement comes exactly one month after OpenAI's integration on AWS Bedrock. In two steps, OpenAI has shattered the exclusivity model that tied it to Microsoft Azure since 2019.

Behind this product integration lies a $300 billion compute contract over 5 years, the largest AI infrastructure deal in history. An amount that propels Oracle to the rank of top-tier hyperscaler and accelerates the preparations for OpenAI's IPO scheduled for late 2026.

The key points

• GPT-4o, GPT-5.5 and Codex are available on OCI Generative AI via existing Oracle credits.
• $300 billion contract over 5 years between OpenAI and Oracle for compute, tied to the Stargate project and an expansion to 7 GW.
• End of Azure exclusivity: Microsoft no longer pays a revenue share to OpenAI, freeing the latter to distribute itself on AWS, Google Cloud and Oracle.
• Oracle integrates GPT-5 into its database and SaaS applications, available in three sizes for its enterprise customers.
• Pre-IPO strategy: OpenAI is maximizing its distribution channels to boost its recurring revenue before going public.

Recommended tools

Models available on OCI — What you can use right now

OpenAI is deploying a targeted selection of its flagship models on OCI. Not the entire catalog, but the high-demand enterprise models.

GPT-5.5 is OpenAI's flagship agentic model, with a score of 98.2 on reference benchmarks. It is the most powerful model on the market ahead of Gemini 3 Pro Deep Think (95.4) and Claude Opus 4.7 Adaptive (94.3). On OCI, it is accessible via a dedicated endpoint with the same performance guarantees as on the direct OpenAI API.

GPT-5.4 Pro (score 91.8) and GPT-5.4 (score 87.6) offer lower-cost alternatives for tasks that do not require the reasoning level of GPT-5.5. They are particularly well-suited for batch processing and automated workflows.

Codex (GPT-5.3 Codex, score 80) is OpenAI's code model, designed for code generation, review, and debugging. Its integration on OCI directly targets development teams that already use Oracle infrastructure for their CI/CD pipelines. For those comparing the best LLMs for coding, Codex remains a solid contender even against Claude Opus 4.7.

GPT-4o completes the offering as a versatile and cost-effective model, ideal for chatbot, classification, and entity extraction use cases.

All these models are listed in Oracle's official documentation of pre-trained models available on OCI Generative AI.

How Universal Credits work to access OpenAI

This is the most strategic part of the announcement: companies do not need to sign a new contract or open a separate OpenAI account.

The existing credits mechanism

Oracle Cloud Universal Credits work like a prepaid wallet covering all OCI services. Until now, these credits funded compute, storage, databases, and Oracle's proprietary models (like Cohere).

Now, calls to OpenAI models are deducted from this same balance. In practice, a company with $500,000 in unused OCI credits can consume them to run GPT-5.5 without any additional purchasing steps.

Zero provisioning delay

According to the practical guide published by MindWiredAI on June 11, 2026, access is immediate. No validation by a sales representative, no provisioning delay. OpenAI models appear in the OCI console just like any other service.

This is a considerable advantage compared to the classic OpenAI API purchasing process, which often involves enterprise negotiations for large volumes.

Governance and security

The enterprise analysis by N1N.ai emphasizes that the integration respects OCI's governance mechanisms: role-based access control (IAM), audit logs, encryption in transit and at rest, and compliance with regulatory standards (SOC 2, HIPAA, FedRAMP).

For regulated companies, this means that data sent to OpenAI models via OCI benefits from the same security framework as the rest of the Oracle infrastructure.

The $300 billion deal — Decoding the largest compute contract in history

The product partnership is just the tip of the iceberg. Behind the scenes, Oracle and OpenAI have signed a $300 billion compute contract over 5 years.

Where does this money come from?

OpenAI raised $122 billion in its latest funding round. A significant portion of this raise is committed to this contract with Oracle to secure the computing capacity needed for the training and inference of its future models.

The contract is an extension of the Stargate project, the mega AI infrastructure project backed by the US government. According to Data Center Frontier, the Stargate capacity increases from 4.5 GW to 7 GW as part of this deal, with superclusters reaching up to 131,072 NVIDIA GPUs.

Oracle's backlog reaches new heights

This $300 billion contract propels Oracle's total backlog to $523 billion. A figure that confirms the rise of Oracle as a central player in AI infrastructure, alongside AWS, Azure, and Google Cloud.

Oracle has more than 100 datacenter regions worldwide, an asset for deploying infrastructure on a national scale. Sovereign cloud is also in its sights, with deployments planned in regions with enhanced data sovereignty.

OpenAI's financial sustainability in question

The AI Track raises a legitimate question: Can OpenAI really honor $300 billion in compute spending over 5 years, or $60 billion per year? For comparison, OpenAI's annual revenues are estimated to be around $10-15 billion in 2026.

The answer relies on two levers: the continuation of fundraising and the exponential revenue growth expected with multi-cloud distribution. The IPO is clearly the catalyst that must bridge the gap.

OpenAI's multi-cloud strategy — Azure, AWS, Oracle, the balancing act

The arrival on Oracle is not an isolated event. It is the third piece of a systematic distribution strategy triggered in the first quarter of 2026.

The end of Microsoft exclusivity

According to VentureBeat, Microsoft and OpenAI have deeply overhauled their historic agreement. Microsoft no longer pays a revenue share to OpenAI for access to models via Azure. In exchange, OpenAI regains its distribution freedom.

This is a major reversal. Since 2019, Azure was the exclusive access channel to OpenAI models for enterprises. This exclusivity made Azure the reference cloud for generative AI for two years.

AWS Bedrock in April 2026

First consequence of this liberation: the arrival of OpenAI on AWS Bedrock in April 2026. GPT-5.5, Codex, and OpenAI's Managed Agents have become accessible to the millions of AWS developers.

The irony is strong: AWS already hosts Anthropic (Claude) as a privileged partner. Bedrock now offers both model ecosystems side by side.

Oracle Cloud in June 2026

Second consequence: OCI integration. Oracle brings a different customer segment than AWS. Oracle enterprises are often large corporations in the banking, healthcare, and heavy industry sectors. Sectors where existing OCI credits represent massive budgets, partly sitting idle.

Google Cloud next?

Forbes analyzes this strategy as methodical preparation for an IPO. OpenAI wants to demonstrate that it is not dependent on a single hyperscaler for its revenues. Logic dictates that Google Cloud should be next on the list, even if no official announcement has been made yet.

This multi-cloud distribution positions OpenAI as an independent model provider, just like Anthropic. Moreover, Anthropic and OpenAI each launch their $10 billion enterprise JV to deploy AI in SMBs and large corporations, marking direct competition on all fronts.

GPT-5 integrated into Oracle's SaaS products — AI at the heart of the database

The integration doesn't stop at the API. Oracle announced that GPT-5 would be embedded directly into its database and SaaS cloud applications (Fusion Cloud, NetSuite, etc.).

Three model sizes for different use cases

According to Constellation Research, Oracle offers GPT-5 in three configurations tailored to its products:

• Large size for complex reasoning tasks in ERP and financial planning modules.
• Medium size for document analysis and natural language processing in the CRM.
• Small size for contextual suggestions and autocompletion in user interfaces.

The advantage of native integration

When GPT-5 is integrated into the Oracle database, data does not need to leave the infrastructure to be processed by the model. This eliminates latency issues, reduces data transfer costs, and simplifies regulatory compliance.

For a bank using Oracle Database for its customer data, being able to run natural language queries directly in the database via GPT-5 is an operational game changer.

Competition with Microsoft Copilot

This is also a strong signal sent to Microsoft. Microsoft integrated GPT-4 into its Office 365 products via Copilot. Oracle is doing the same thing with GPT-5 in its SaaS ecosystem, but with a newer and more powerful model.

Oracle customers no longer need to look toward the Microsoft ecosystem to benefit from generative AI in their business tools.

On-premise: Oracle completes the lineup with Dell

The cloud offering is complemented by an on-premise dimension. OpenAI and Dell push Codex on-premise, marking the end of cloud-only for enterprise AI.

For companies that cannot send their data to the cloud (defense, healthcare, industry), Oracle and Dell offer physical appliances with Codex pre-installed. The GPT-5.3 Codex model (score 80) is particularly suited for this deployment because it is optimized for coding tasks that largely remain internal.

This cloud + on-premise coverage positions OpenAI as an "anywhere" AI model provider, capable of meeting all architecture constraints.

Implications for the AI cloud landscape — Who wins, who loses

The big winner: OpenAI

OpenAI emerges from this June 2026 with an unprecedented distribution position. Azure, AWS Bedrock, Oracle Cloud, on-premise via Dell, and soon probably Google Cloud. No other model provider can claim such coverage.

The strategy is clear: maximize touchpoints with enterprises to boost recurring revenue before the IPO. Every dollar spent on OCI credits or Bedrock calls turns into a favorable metric for the roadshow.

The second winner: Oracle

Oracle was seen as the fourth hyperscaler, far behind AWS, Azure, and GCP. This $300 billion deal and the integration of OpenAI models change the game. Oracle can now argue to its customers: "Your OCI credits give you access to the best models on the market, including GPT-5.5."

The $523 billion backlog gives Oracle exceptional long-term visibility. The 7 GW Stargate infrastructure positions the company as a player in national AI sovereignty.

The silent loser: Microsoft Azure

Microsoft sacrificed its exclusivity on OpenAI. In exchange, it no longer pays a revenue share, which improves its margins. But by losing exclusivity, Azure loses its main differentiating argument against AWS and GCP.

Enterprises that chose Azure solely for GPT-4/5 no longer have that reason. They can now stay on AWS or Oracle and find the same models there. Microsoft must now compete on Azure's own merits, not on OpenAI exclusivity.

Anthropic under pressure

Anthropic remains AWS's flagship partner, with Claude Opus 4.7 Adaptive (94.3) as an alternative to GPT-5.5 (98.2). But the cohabitation on Bedrock is ambiguous: AWS developers will naturally compare the two. If GPT-5.5 is available in the same place as Claude, with a score nearly 4 points higher, the choice may tip toward OpenAI for critical use cases.

Anthropic compensates by betting on security, output reliability, and Claude Opus 4.7's dynamic workflows. But the competitive pressure is real.

Comparison of OpenAI distribution channels (June 2026)

❌ Common mistakes

Mistake 1: Thinking Oracle credits are free for OpenAI

Universal Credits are already paid for by the customer. But each call to GPT-5.5 consumes them. A company that exhausts its credits on OpenAI API calls will have no budget left for its classic compute. The solution: set quotas per service and monitor consumption via OCI dashboards.

Mistake 2: Choosing OCI solely for OpenAI without evaluating the ecosystem

If your company does not already use Oracle, signing up for OCI credits just to access GPT-5.5 makes no financial sense. The benefit is maximal for organizations that already have dormant credits. Otherwise, the direct API or AWS Bedrock are more relevant.

Mistake 3: Ignoring regionalization constraints

Not all OpenAI models are available in all OCI regions. Oracle is deploying progressively, and some sovereign regions may have reduced catalogs. Check availability in the Oracle documentation before architecting a critical solution.

Mistake 4: Confusing GPT-5 (Oracle SaaS) and GPT-5.5 (OCI API)

GPT-5 integrated into Oracle SaaS applications is not GPT-5.5. These are optimized and sometimes reduced deployments for business use cases. For complex reasoning and agentic tasks, use the OCI API with GPT-5.5.

❓ Frequently Asked Questions

Which OpenAI models are available on Oracle Cloud?

GPT-5.5, GPT-5.4 Pro, GPT-5.4, GPT-5.3 Codex and GPT-4o are listed in the OCI Generative AI documentation. Availability may vary by region.

How do I access OpenAI models on OCI?

Enable the OCI Generative AI service in your console, select the desired OpenAI model, and your calls are deducted from your existing Universal Credits. No separate OpenAI contract is required.

Is the $300 billion deal confirmed?

Yes, analyzed by Intuition Labs, The AI Track and Data Center Frontier. It covers compute over 5 years as part of the Stargate project expanded to 7 GW.

Is OpenAI still a Microsoft partner?

Yes, but exclusivity is over. Microsoft no longer pays a revenue share, and OpenAI can freely distribute on AWS, Oracle and potentially Google Cloud.

Are my data secure on OCI?

Yes, calls to OpenAI models via OCI benefit from Oracle's security framework: IAM, encryption, audit logs, SOC 2, HIPAA and FedRAMP compliance depending on the region.

Can I use Codex on-premise with Oracle?

The on-premise deployment of Codex is done through the Dell partnership, not directly with Oracle. Oracle focuses on the cloud with OCI.

✅ Conclusion

In two months, OpenAI went from a captive Microsoft partner to a multi-cloud distributor covering Azure, AWS Bedrock and Oracle Cloud, with an on-premise pipeline via Dell. The $300 billion deal with Oracle and access via existing credits transform OCI into a major enterprise entry point for GPT-5.5 and Codex. For companies already equipped with Oracle credits, the door is open from today. For others, the choice of OpenAI distribution channel has become a real architecture issue — and it is better to compare the available LLMs before committing.