OpenAI's Frontier Models Land on Amazon Bedrock the Day After MSFT Exclusivity Ends
OpenAI dropped GPT-5.5, GPT-5.4, Codex, and Managed Agents on Amazon Bedrock on April 28 — exactly one day after Microsoft's exclusivity expired. First time AWS customers

$50B
OpenAI dropped GPT-5.5, GPT-5.4, Codex, and Managed Agents on Amazon Bedrock on April 28 — exactly one day after Microsoft's exclusivity expired. First time AWS customers get OpenAI frontier models with native IAM, PrivateLink, guardrails, and CloudTrail logging. Limited preview today, broader rollout to follow.
Here's the deal: this isn't just another announcement. The shift took roughly 24 hours to land, and it rewires assumptions that enterprise architects have been building on for the past three years.
The Players
OpenAI's Bedrock launch is the public end of the 'OpenAI = Azure exclusive' era that started in 2019. On April 27, Microsoft and OpenAI restructured their partnership so OpenAI can serve its products on any cloud; one day later, OpenAI cashed in its $100B+ AWS infrastructure contract and dropped models, Codex, and Managed Agents on Bedrock. The limited preview ships with GPT-5.5, GPT-5.4, the Codex coding agent, and AWS-hosted Managed Agents powered by OpenAI. The unlock is enterprise compliance: OpenAI models inherit IAM, AWS PrivateLink, guardrails, encryption, and CloudTrail logging, which lets regulated industries adopt frontier models that previously failed their procurement. For AWS, this fills the last hole in Bedrock's catalog. For Microsoft, it's the loss of the exclusivity card but a continued revenue share. Analysts call this the inflection point in AI infrastructure politics.
Both sides have been moving fast. Cash, model releases, and infrastructure announcements have been compressing into multi-week cycles instead of multi-quarter ones. Today's news is the visible spike of pressure that has been building all spring.
Outside the headline, the rest of the industry is moving in the same direction — agent platforms, multi-cloud routing, sovereign-AI clauses. GitHub trending, HN front pages, and IR decks are all converging on the same vocabulary, and this story sits at the intersection.
출처: openai.com · 회사 OG · 뉴스 fair use
Numbers and Timing
| Item | Value |
|---|---|
| Available Models | GPT-5.5, GPT-5.4, Codex, Managed Agents |
| Status | Limited Preview |
| Enterprise Controls | IAM, PrivateLink, Guardrails, Encryption, CloudTrail |
| AWS Contract | $50B+ multi-year |
Read each row as a procurement decision, not a spec sheet line. The 'Status' and 'Enterprise Controls' rows are the ones compliance teams care about.
Pricing / Access
| Item | Value |
|---|---|
| input_per_1m | TBD per region |
| output_per_1m | TBD per region |
| free_tier | False |
| api_available_from | 2026-04-28 |
Timeline
| Date | Event |
|---|---|
| 2019-07-22 | Microsoft invests $1B in OpenAI; Azure exclusivity begins |
| 2026-04-23 | OpenAI launches GPT-5.5 in ChatGPT |
| 2026-04-24 | GPT-5.5 hits OpenAI API |
| 2026-04-27 | Microsoft and OpenAI restructure partnership; exclusivity ends |
| 2026-04-28 | OpenAI models go live on Amazon Bedrock |
The compression is the story. The same arc used to take 18-24 months. Today it took roughly six weeks.
Who Wins, Who Pays
OpenAI 모델이 Azure 밖에서 정식 서빙되는 첫 사례 — AI 클라우드 균형이 흔들림.
엔터프라이즈에서는 'OpenAI를 쓰려면 Azure 가야 한다'는 7년짜리 기본값이 무너졌어. AWS에 인프라가 있던 회사는 Bedrock 한 줄로 GPT-5.5를 부를 수 있고, IAM·PrivateLink가 적용되니 컴플라이언스 팀의 빨간펜이 사라져. 마이크로소프트는 매출 쉐어로 단기 손해는 적지만 'OpenAI 강제 록인'이라는 가장 큰 카드가 사라졌고, AWS는 Anthropic·Meta·Mistral·OpenAI를 모두 담은 유일한 메가 카탈로그가 됐어.
Three new openings appear: enterprises that swap backends without rewriting workflows; SaaS vendors that reopen pricing negotiations using this announcement as leverage; and competitors that just lost a card in their next sales cycle.
출처: openai.com · 회사 OG · 뉴스 fair use
Historical Echoes
There are two analogues worth holding in mind. AWS-Snowflake in 2017 broke data warehouse lock-in and triggered a price compression cycle. The Codex-to-Copilot arc in 2020-2022 collapsed model exclusivity into a baseline expectation. Both ended with users gaining leverage and the dominant cloud losing pricing power.
The cautionary tale is IBM-Red Hat in 2018 — a structural reset that never produced the cloud share it promised, because internal incentive lines never aligned. Today's deal will face the same test inside both partner orgs.
Competitor Counter-Play
Google Cloud is the most likely fast counter — expect a Vertex AI Garden refresh that bundles Gemini, Anthropic and Mistral with a sharper price card. Anthropic's risk is direct comparability: Opus 4.7 still wins on reasoning, but the 2-3x price gap to GPT-5.5 is now a sales-room talking point. Salesforce, Oracle and ServiceNow benefit, because they can wrap GPT-5 inside their products under any cloud.
What Changes
Developers: Yesterday the canonical OpenAI stack was the OpenAI SDK on Azure endpoints. Today, boto3 + Bedrock calls the same models, with prompt caching, vision, function calls, and JSON mode parity.
Founders: Infrastructure choice expands. Health, finance, and public sector deployments now have an OpenAI path that doesn't require Azure.
Investors: Watch AWS 1Q26 guidance. Bedrock token throughput will likely become a new headline metric next quarter. Azure's OpenAI revenue line gets a 1-2pp short-term drag.
End users: ChatGPT pricing won't move. But your company's internal AI tools may suddenly get cheaper or faster as procurement reroutes.
Stakes
- Wins: OpenAI — diversified channel, new revenue surface.
- Wins: AWS — Bedrock catalog now matches or exceeds Azure's.
- Loses: Microsoft — exclusivity card gone, near-term Azure growth pressure.
- Loses: Anthropic — direct GPT-5.5 comparability in every Bedrock customer.
- Watching: Google Cloud — must answer with sharper pricing and bundling.
- Watching: Salesforce / Oracle — opportunity to lean cloud-neutral in pitches.
A Skeptic Speaks
Ed Zitron (Where's Your Ed At): "OpenAI's growth is just outsourced cloud cash flow."
Zitron's argument: revenue grows because infrastructure cost externalizes, not because unit economics improve. Bedrock changes the channel mix, not the underlying gross margin math.
Tomorrow Morning
- Developers: enable OpenAI models in AWS Bedrock console, queue for limited preview.
- Developers: write one boto3 migration snippet for an existing Azure OpenAI call and ship a PR.
- PMs / Founders: revisit SaaS vendor terms — which clouds can route to OpenAI now?
- Investors: AWS 1Q26 earnings (May 1) — look for new Bedrock metric.
- End users: note any speed/quality change in your company's AI assistants this week.
Sources
관련 기사

OpenAI's GPT-5.5 'Spud' Pretraining Complete – Launch Expected Within Weeks
With pretraining finished on March 24, OpenAI's next frontier model is poised to redefine what's possible in AI, potentially arriving as early as April or May 2026.

Microsoft-OpenAI Exclusivity Is Over — AWS and Google Can Now Sell GPT
On April 27, Microsoft and OpenAI scrapped cloud exclusivity. The AGI clause is dead, revenue share is capped, and OpenAI is free to sell models on AWS and Google Cloud. Microsoft keeps a ~27% stake worth ~$135B.

OpenAI's Lilli Replaces Internal Knowledge Search with AI Agents
OpenAI's internal search system Lilli launches for enterprise. Can it replace Notion and Confluence?
AI 트렌드를 앞서가세요
매일 아침, 엄선된 AI 뉴스를 받아보세요. 스팸 없음. 언제든 구독 취소.
