Amazon Web Services quietly ended one of the most significant enterprise AI exclusivities on June 1, 2026. That was the day Amazon Bedrock began offering OpenAI's GPT-5.5, GPT-5.4, and Codex models under its standard managed service, shattering Microsoft Azure's first-party lock on the GPT-5 family. For Windows shops and IT teams that felt forced onto Azure AI Foundry just to access OpenAI's frontier models, the calculation just got rewritten.

The Monopoly Breaks: Bedrock Gets GPT-5.5

The June 1 launch, now in general availability, puts these OpenAI models inside Bedrock's familiar API surface. Billing aligns with OpenAI's own first-party rates, and the models run under AWS's existing identity, networking, and governance controls—no separate Azure subscription needed.

A widely shared comparison published on July 16, 2026, by Tech Insider touted Azure AI Foundry as the exclusive cloud home for GPT-5. But as a follow-up analysis by WindowsForum.com quickly pointed out, that claim was already six weeks out of date. Bedrock's addition of GPT-5.5 and its siblings means organizations can now evaluate the latest OpenAI models alongside Amazon's Titan and Nova, Anthropic's Claude, Meta's Llama, Mistral, and others—all within the same AWS account structure they already use.

What the June 1 Launch Actually Delivered

Here are the concrete changes:

  • Models available: GPT-5.5, GPT-5.4, and Codex. These are the same models previously accessible only through Azure AI Foundry or directly from OpenAI's API.
  • Pricing: Aligned with OpenAI's first-party rates. While exact figures aren't public, Azure's GPT-5 pricing at $1.25/million input tokens and $10/million output tokens serves as a reference.
  • Service tiers: Bedrock now offers more than just serverless. Standard, Priority, Flex, and Reserved options let you trade latency for cost based on workload demands—something Azure's Provisioned Throughput Units (PTUs) can't easily replicate.
  • Governance: All AWS-native guardrails apply: IAM roles, VPC networking, CloudTrail logging, and KMS encryption. For organizations already running on AWS, this eliminates a major compliance headache.

Microsoft still holds deep integration advantages: Azure AI Foundry ties directly into Microsoft 365, Teams, SharePoint, and Copilot. If your productivity stack is entirely Microsoft-centric, that remains a compelling reason to stay. But for everyone else—especially multi-cloud teams or those with existing AWS commitments—Bedrock is now a viable, and often more flexible, alternative for GPT workloads.

Your Next Steps: Re-evaluating the AI Platform Decision

If your team previously committed to Azure AI Foundry primarily because it was the only cloud platform with native GPT-5 access, it's time for a fresh look. Here's how to approach it:

1. Don't get distracted by catalog counts

The "17x gap" between Azure's 1,900+ models and Bedrock's 100+ is largely a catalog count mirage. Azure's number includes community models, embeddings, and long-tail experiments. Bedrock's list covers the models most production teams actually use: Claude, Llama, Mistral, DeepSeek, Titan, Nova, and now GPT-5.5. What matters is which approved models are available in your required region and compatible with your stack—not how many entries sit in a portal.

2. Run a region-specific proof of concept

Spin up a test workload on both platforms. Compare not just inference latency and token cost, but total cost of ownership: egress fees, embedding storage, vector search indexing, and any idle capacity charges. Bedrock's serverless pricing often comes out 15–25% cheaper for bursty or mid-volume workloads (10–50 million tokens/month), according to third-party analyses. For steady high-volume traffic, Azure's PTUs may still win. Bedrock's new Priority and Flex tiers may shift that math further.

3. Audit your existing cloud and license commitments

If you're already deep into AWS with reserved instances and a mature IAM setup, adding GPT-5.5 via Bedrock is operationally trivial. If your data and apps live on Azure, the native Copilot and Microsoft 365 integrations may justify staying. For Google Cloud shops, Vertex AI remains the natural fit with its BigQuery ties and fast AutoML training—but if you need GPT-5.5 specifically, you'll now have to consider adding an AWS footprint or using the direct OpenAI API.

4. Plan agent and fine-tuning rebuilds if you migrate

Fine-tuned models and agent orchestration logic don't travel between clouds. If you move from Azure AI Foundry to Bedrock, budget time to retrain custom models and rebuild agent workflows. The model API swap itself is relatively straightforward; the real cost is in the bespoke agent frameworks and embeddings that each platform wraps around the core LLM.

The Catalog Count Distraction

The Tech Insider comparison that triggered this discussion claimed a 17x advantage for Azure based largely on model-catalog breadth. That's a dangerous oversimplification. A catalog number mixes serverless APIs, community models, specialist models, and models requiring separate deployment—often with different regional and compliance footnotes. For a Windows enterprise, the choice shouldn't hinge on whether a portal exposes 100 or 1,900 entries. It should hinge on which models are approved for the workload, available in the required region, supported by the application stack, and priced predictably at expected volume. Bedrock's addition of GPT-5.5 drives this home: the platform split is no longer clean.

How We Got Here: From Exclusive Deal to Commodity

Microsoft's multi-billion-dollar partnership with OpenAI, originally forged in 2019, gave Azure exclusive rights to host OpenAI's models in the cloud. That exclusivity became a major differentiator when GPT-4 and later GPT-5 models fueled enterprise adoption. For nearly a year, Azure AI Foundry was the only game in town for organizations that wanted a fully managed GPT-5 experience with enterprise controls.

Meanwhile, AWS built Bedrock around a multi-model strategy, hosting Anthropic's Claude, Meta's Llama, Mistral, and its own Titan and Nova models. The glaring hole—no GPT—was a deal-breaker for teams that had standardized on OpenAI's model family. Google Vertex AI, anchored by Gemini, similarly couldn't offer GPT.

AWS's June 1 move changes that. Industry observers had speculated that OpenAI's partnership terms might relax as competition intensified and as Amazon's own Nova models matured. The release suggests OpenAI sees value in making its models accessible across clouds, potentially to capture more enterprise spend and reduce dependence on Microsoft's distribution.

What to Watch in the Second Half of 2026

The AI platform market is now a three-horse race where exclusivities are crumbling. A few developments to track:

  • Google Vertex AI: Will Google seek to host GPT models as well? For now, Vertex doubles down on Gemini and open-weight Gemma, but if Bedrock's GPT addition pressures deal flow, Google may follow.
  • Microsoft's countermove: Expect Azure AI Foundry to lean even harder into Copilot and Microsoft 365 integrations that Bedrock can't replicate, along with tighter coupling to Purview for compliance.
  • Multi-cloud abstraction: With the same frontier models now available on two platforms, the case for building a thin API layer that avoids vendor lock-in becomes stronger. Tools like LangChain and custom routing proxies will see increased investment.
  • Pricing wars: Bedrock's new service tiers (Flex, Reserved) and Azure's PTU model will continue to evolve. Watch for per-request pricing adjustments and new free tiers as competitive pressure mounts.

For Windows and enterprise IT teams, the immediate takeaway is simple: if you settled on Azure AI Foundry solely because you needed GPT-5, your options just doubled. Re-run your evaluations with real workloads, factor in your existing cloud investments, and don't let catalog numbers or outdated exclusivity claims drive a decision that will lock in your AI stack for years to come.