The end of Azure exclusivity: why it matters

Since the launch of commercial GPT models, an unwritten rule applied: if you wanted OpenAI in your enterprise cloud, you had to go through Microsoft Azure. This exclusive agreement, which brought Azure billions of dollars in new cloud contracts, officially ended on June 1, 2026. OpenAI now offers its most powerful models on Amazon Bedrock — and according to statements from both companies, this is a full-fledged partnership, not a second-class integration.

Pricing on Bedrock mirrors the official OpenAI price list dollar for dollar, with no regional surcharges. GPT-5.5 costs $5 per million input tokens and $30 per million output tokens. Moreover, all consumption is automatically applied to existing AWS cloud commitments — so companies don't have to manage separate billing or request new budgets.

What exactly is available on Bedrock

Amazon Bedrock now offers three key OpenAI products:

• GPT-5.5 — OpenAI's most powerful model with a million-token context window, optimized for complex reasoning, agent workflows, and scientific research. It scores 82.7% on the Terminal-Bench 2.0 benchmark, surpassing Claude Opus 4.7 (69.4%) and Gemini 3.1 Pro (68.5%).
• GPT-5.4 — the previous generation, still highly performant for common enterprise tasks including data analysis and content generation.
• Codex — an autonomous software agent that writes, fixes, tests, and modernizes code. It is deployed via its own application, CLI, or plugins for Visual Studio Code, JetBrains, and Xcode. It operates purely on a pay-per-token basis — no developer licenses.

GPT-5.5 runs in the US East (Ohio) region, while GPT-5.4 is also available in US West (Oregon). For European customers, it's significant that AWS enables data processing in a specific region while complying with GDPR — and that Bedrock inherits all AWS security mechanisms: IAM permissions, isolation via AWS PrivateLink, and complete audit logging through CloudTrail.

What it means for businesses — and for Microsoft

For enterprises already running on AWS, this integration is a massive simplification. Instead of building a hybrid architecture between two clouds, developers can call GPT-5.5 through the same API as any other AWS service. Data stays in an environment that the company knows and that has passed internal security audits.

For Microsoft, this is a strategic blow. Exclusivity on OpenAI models was one of Azure's main draws for AI workloads. Now that customers can choose between Azure and AWS at the same prices and with the same models, other factors decide — infrastructure quality, existing relationships, and the ecosystem of complementary services.

Czech companies that use AWS — and there are plenty of them, from startups to corporations like Škoda Auto or Avast — can now deploy OpenAI models without complex integration with another cloud provider. Simply enable the OpenAI models in the Bedrock console and get started.

New inference architecture: continuity guarantees for critical workflows

AWS has built a new generation of inference engine for these models, solving one of the biggest problems in production AI: what happens when all customers want a response at the same time. Instead of rejecting requests during peaks, the engine uses isolated queues — requests are queued and processed as capacity becomes available. Additionally, the engine remembers the state of each request at the hardware level, so if a node fails, processing can continue without restart.

Technically, this is a major improvement over the classic serverless model, where an unexpected spike would mean error messages and frustrated developers. For companies building critical workflows on AI models, this guarantee of continuity is essential.

Who is already using Bedrock with OpenAI

Among the early customers are two big names that demonstrate the breadth of possible deployment:

Amgen, the biotechnology giant, uses GPT-5.5 to accelerate therapeutic research. Sean Bruich, the company's CTO, stated: "In an industry where questions are extraordinarily complex and standards of scientific accuracy are extremely high, OpenAI models on AWS give us a new way to scale these capabilities within responsible AI."

Autodesk, a leader in design software, is testing Codex on Bedrock to accelerate development workflows. Ritesh Bansal, vice president of AI and ML platform, explains: "With OpenAI models on Bedrock, we're exploring how frontier AI can accelerate development and support more informed decision-making for our customers."

What the future brings: agents and cybersecurity

OpenAI and AWS also announced further phases of collaboration set to arrive in the coming months:

• Amazon Bedrock Managed Agents — a tool for building production agent solutions where OpenAI models act as the agent's brain and AWS provides persistent memory, tool orchestration, and a security layer.
• Daybreak — OpenAI's cybersecurity framework, which will enable security teams on Bedrock to identify and fix vulnerabilities directly in the cloud. It will include specialized cyber models and Codex Security, which will assist with threat modeling, patch validation, and dependency risk analysis.

Daybreak represents an interesting shift: instead of companies fearing that AI will help attackers, they get a tool that actively defends their infrastructure. Early vulnerability detection and automated remediation thus become part of the daily development cycle.

Context for Czech and European companies

For the Czech market, it's significant that AWS is well-established in the country — from e-shops to banks to public administration. Companies that have already invested in AWS infrastructure and have completed compliance requirements can now add the most powerful AI models without additional regulatory hurdles.

From the perspective of the EU AI Act, the advantage is that models run in a proven cloud environment with transparent logging and governance tools. AWS Bedrock also supports Guardrails — configurable security filters that can be adapted to European regulatory requirements, including filtering inappropriate content or protecting personal data.

The GPT-5.5 and GPT-5.4 models themselves understand Czech at a very good level, so they can be used for Czech language tasks including document summarization, customer support, or analysis of Czech texts directly from the AWS environment.

Comparison: OpenAI on AWS vs. Azure

The differences between the two platforms are now blurring, but a few remain:

• Model pricing: identical (same token rates)
• Applied to cloud commitments: yes on both platforms
• Security certifications: both platforms offer SOC, ISO, HIPAA
• Developer tool integration: Azure has an edge thanks to the GitHub Copilot ecosystem, while AWS counters with Codex integration via Bedrock
• GDPR: both platforms enable processing in European regions; AWS has data centers in Frankfurt, Ireland, Sweden, and other countries

In practice, this means the deciding factor will no longer be "which cloud has OpenAI," but which cloud the company already uses for the rest of its infrastructure.

How to get started

Companies interested in OpenAI on AWS can activate the models directly in the Amazon Bedrock console. For Codex, standard AWS authentication is sufficient — developers don't need to create separate OpenAI accounts or manage additional API keys. All communication with models takes place via standard OpenAI Python SDK endpoints, so existing code built on the OpenAI API works practically without modification. AWS also offers a dedicated page for Bedrock Managed Agents with OpenAI, where production agents can be configured.