On April 28, 2026, Amazon Web Services and OpenAI ended one of the most consequential exclusivity arrangements in cloud computing history, making OpenAI’s flagship frontier models – including the just-shipped GPT-5.5 – available on Amazon Bedrock for the first time. The move closes the door on nearly seven years of Microsoft Azure being the only hyperscaler legally permitted to host OpenAI’s proprietary large language models, and opens the door to a multi-cloud era for the world’s most-used AI lab.

The launch follows a $38 billion, seven-year compute commitment that OpenAI signed with AWS in November 2025, which gave OpenAI access to “hundreds of thousands” of NVIDIA GB200 and GB300 GPUs hosted in Amazon EC2 UltraServers, plus the ability to scale to tens of millions of CPUs by the end of 2026. With models now landing inside Bedrock, AWS becomes the exclusive third-party distributor for OpenAI’s enterprise platform, sitting alongside Anthropic, Meta, Mistral, Cohere and Amazon’s own Nova family in a managed catalogue that already serves more than two million developers.

For Microsoft, the loss of exclusivity caps a year of restructuring around OpenAI: Azure remains the “primary” cloud partner under a license that runs through 2032, but the unilateral lock-in that defined the partnership since 2019 is gone. For Andy Jassy, who has spent eighteen months making AWS the default landing pad for every major AI lab, the deal validates a multi-model Bedrock strategy that now spans more than 100 foundation models across six labs.

The April 28 Announcement: What Actually Shipped

The headline of the April 28, 2026 announcement is straightforward: OpenAI’s latest frontier models, including GPT-5.5 and GPT-5.4, are now available on Amazon Bedrock in limited preview, with general availability rolling out “within weeks.” Customers can call the models through the same Bedrock APIs they already use – InvokeModel, Converse, Chat completions and batch inference – and reuse existing IAM policies, guardrails, knowledge bases, and OpenAI SDK integrations pointed at Bedrock endpoints. AWS confirmed that the OpenAI Python and Node SDKs will work unchanged against Bedrock by simply switching the base URL and supplying AWS credentials.

Two open-weight models, gpt-oss-20b and gpt-oss-120b, were also added to Bedrock under the same agreement. The 20B variant is positioned for low-latency inference and edge deployments, while the 120B model targets production reasoning workloads. Both ship with Bedrock’s standard provisioned-throughput pricing model and full access to fine-tuning APIs, marking the first time OpenAI’s open-weight tier has been distributed natively by a hyperscaler other than its own platform.

Beyond the models themselves, AWS and OpenAI confirmed they are co-developing a “Stateful Runtime Environment” – a managed agentic execution layer that retains context across calls, exposes compute and memory primitives, and integrates directly with Bedrock AgentCore. The runtime, scheduled to launch within “the next few months,” is positioned as the technical substrate for OpenAI Frontier, the enterprise agent platform for which AWS now holds exclusive third-party distribution rights.

Microsoft’s Exclusivity Ends After Seven Years

The Microsoft–OpenAI exclusivity that expired on April 27, 2026 dates back to the original 2019 partnership, when Microsoft committed $1 billion to OpenAI in exchange for the right to host its models on Azure. That arrangement was reinforced by Microsoft’s $10 billion investment in 2023 and a deep 2024 expansion that wired Copilot, Bing, and Microsoft 365 into OpenAI’s API. By early 2025, Azure OpenAI Service had become one of the fastest-growing line items in Microsoft’s cloud business, with hundreds of thousands of enterprise customers running production workloads against GPT-4o and the o-series reasoning models.

The amended October 2025 restructuring agreement explicitly cleared a path for OpenAI to run on additional clouds once exclusivity lapsed, while preserving Microsoft’s preferential rights through 2032. The new arrangement gives Microsoft a “non-exclusive license” to host OpenAI’s models, eliminates Microsoft’s right of first refusal on new compute, and resolves the legal overhang that had clouded OpenAI’s $38 billion AWS deal. In a statement on April 28, Microsoft acknowledged the change, saying Azure “remains the primary cloud partner for OpenAI” and that the company would “continue to invest aggressively” in joint capacity.

For OpenAI, the unblocking is strategic. The company has been signalling for more than a year that capacity constraints – particularly around inference for ChatGPT, which now serves more than 200 million weekly active users – were limiting growth. Multi-cloud distribution gives OpenAI access to AWS’s NVIDIA GB200/GB300 fleet without renegotiating with Microsoft for every incremental gigawatt, and it allows enterprise customers to satisfy regulatory or procurement requirements that mandate AWS as their primary cloud.

The $38 Billion AWS Compute Deal That Set the Stage

The April 2026 model launch is the consumer-facing manifestation of the $38 billion, seven-year strategic partnership AWS and OpenAI signed on November 3, 2025. Under that agreement, AWS committed to deliver “hundreds of thousands” of NVIDIA GB200 and GB300 GPUs via EC2 UltraServer clusters, plus the ability to scale to “tens of millions of CPUs” for agentic workloads, with the bulk of capacity coming online by the end of 2026 and additional growth slated for 2027 and beyond.

“Scaling frontier AI requires massive, reliable compute. Our partnership with AWS strengthens the broad compute ecosystem that will power this next era and bring advanced AI to everyone,” Sam Altman said when the deal was announced. AWS CEO Matt Garman framed it as a generational endorsement of AWS’s infrastructure: “With this new $38 billion agreement, OpenAI will immediately start using our world-class infrastructure – including Amazon EC2 UltraServers packed with hundreds of thousands of state-of-the-art NVIDIA GPUs and the ability to scale to tens of millions of CPUs.”

The compute deal was the second-largest single AI infrastructure commitment in history at the time of signing, behind only the multi-year arrangements OpenAI has with Microsoft and Oracle. It also created an unusual situation: AWS would be hosting OpenAI training and inference on its own infrastructure for nearly six months before being legally allowed to resell those models to Bedrock customers. The April 28 launch closes that gap.

Models on Bedrock: GPT-5.5, GPT-5.4 and Open Weights

The Bedrock catalogue lists four OpenAI models at launch. GPT-5.5, OpenAI’s flagship released on April 23, 2026, leads the lineup. GPT-5.4 remains available for customers requiring stability or longer context. Two open-weight models, gpt-oss-20b and gpt-oss-120b, round out the offering. Codex, OpenAI’s coding-specialised model, is documented in Bedrock but ships as part of the OpenAI Frontier agent platform rather than the standard model catalogue.

GPT-5.5 represents a meaningful jump over GPT-5.4 across nearly every benchmark OpenAI publishes. On Terminal-Bench 2.0, an agentic coding evaluation, GPT-5.5 scores 82.7% versus GPT-5.4’s 75.1%, a 7.6-point gain. SWE-Bench Pro improved from 57.7% to 58.6%, and ARC-AGI-2 jumped from 73.3% to 85.0%. On long-context retrieval – the MRCR2 needle test at 512K to 1M tokens – GPT-5.5 doubled GPT-5.4’s score, climbing from 36.6% to 74.0%.

Benchmark	GPT-5.5	GPT-5.4	Delta
Terminal-Bench 2.0 (agentic coding)	82.7%	75.1%	+7.6pp
SWE-Bench Pro	58.6%	57.7%	+0.9pp
OSWorld-Verified (computer use)	78.7%	75.0%	+3.7pp
FrontierMath T1–T3	51.7%	47.6%	+4.1pp
FrontierMath T4	35.4%	27.1%	+8.3pp
Humanity’s Last Exam (no tools)	41.4%	39.8%	+1.6pp
ARC-AGI-2	85.0%	73.3%	+11.7pp
MRCR2 Needle (512K–1M ctx)	74.0%	36.6%	+37.4pp
GeneBench (genetic data)	25.0%	19.0%	+6.0pp

Pricing follows OpenAI’s tiered structure. GPT-5.4 is listed at $2.50 per million input tokens and $15 per million output tokens on the OpenAI direct API, while GPT-5.5 Pro carries premium pricing at $30 per million input and $180 per million output – roughly twelve times more expensive than GPT-5.4 for high-end reasoning work. Bedrock pricing matches OpenAI’s published rates at the per-token level, with optional volume discounts available through provisioned throughput contracts and AWS Enterprise Discount Programs.

Both models support a 1 million-token context window via the API, with the GPT-5.5 Codex variant offering a 400K-token window and a 1.5x faster “Fast mode” at 2.5x the standard cost. Output is capped at 128K tokens per call. Both accept multimodal inputs – text, images, and audio – though Bedrock’s initial preview only enables text and image modalities.

Bedrock vs Azure OpenAI Service: Feature Comparison

The arrival of OpenAI on Bedrock instantly creates a head-to-head feature contest with Microsoft’s Azure OpenAI Service, which has been the de facto enterprise channel for GPT models since 2023. Both platforms now offer the same flagship models, but the surrounding tooling, governance and integration story differs in ways that matter for procurement decisions.

Capability	Amazon Bedrock (April 2026)	Azure OpenAI Service
Flagship model	GPT-5.5, GPT-5.4	GPT-5.5, GPT-5.4
Open-weight OpenAI models	gpt-oss-20b, gpt-oss-120b	Not available
Other foundation models	Anthropic Claude, Meta Llama, Mistral, Cohere, Amazon Nova (100+ total)	Limited (Microsoft Phi, Mistral, Llama)
Agent runtime	Bedrock AgentCore + Stateful Runtime (preview)	Azure AI Foundry, Copilot Studio
Identity & policy	AWS IAM, KMS, PrivateLink, VPC endpoints	Entra ID, Azure RBAC, Private Endpoints
Guardrails	Bedrock Guardrails (PII, denied topics, contextual grounding)	Azure AI Content Safety
Fine-tuning	Supported on gpt-oss models	Supported on GPT-4o-mini, GPT-5.4
Inference hardware	NVIDIA GB200/GB300, Trainium2	NVIDIA GB200/GB300, Maia 100
Pricing model	On-demand + provisioned throughput	PTU + on-demand standard
Geographic regions	20+ AWS regions at GA	30+ Azure regions

The most consequential difference is the catalogue. Bedrock customers can call GPT-5.5, Claude Opus 4.7, Llama 4 405B and Mistral Large from a single SDK and unified billing surface. Azure OpenAI remains a relatively narrow OpenAI-only product, with Microsoft’s own MAI-1 family available separately through Azure AI Foundry. For enterprises building agent systems that route between models – a pattern that has exploded since the rise of multi-model frameworks like LangGraph and CrewAI – Bedrock’s catalogue is now the easiest single contract to satisfy.

The other major differentiator is hardware diversity. Azure OpenAI runs primarily on NVIDIA GPUs and Microsoft’s Maia 100 silicon, while Bedrock blends NVIDIA capacity with AWS Trainium2 inference accelerators. AWS has been pricing Trainium-served models at a discount of roughly 30–40% relative to comparable NVIDIA workloads, a margin that could become a significant cost lever for high-volume inference customers if OpenAI agrees to make GPT-5.5 available on Trainium in future drops.

Why This Matters: Multi-Cloud AI Becomes the Default

The strategic significance of OpenAI on Bedrock extends well beyond AWS’s catalogue. For three years, the dominant pattern in enterprise AI procurement was simple: if you wanted GPT, you bought Azure. That single fact distorted cloud market share, drove a wave of multi-year Microsoft commitments, and gave Microsoft enormous use in renewal negotiations. Ending the exclusivity flips that dynamic.

“This is the moment multi-cloud AI becomes the default architecture, not the exception,” Forrester analyst Mike Gualtieri wrote in a research note circulated April 28. “Procurement teams have been waiting for OpenAI to be available on AWS. Now that it is, the conversation moves from ‘which cloud has the model I need’ to ‘which cloud has the best price, latency and governance for the model I want.'”

Gartner’s most recent Magic Quadrant for Cloud AI Developer Services, published in March 2026, already showed Bedrock pulling ahead of Azure OpenAI on “completeness of vision” thanks to its multi-model strategy. Adding GPT-5.5 erodes the last remaining advantage Azure had on “ability to execute.” Within weeks of the launch, several large enterprises – including JPMorgan Chase, Salesforce and Pfizer – are expected to either expand their existing Bedrock contracts to include OpenAI workloads or migrate portions of their Azure OpenAI footprint, according to channel sources.

Bedrock’s growth trajectory was already steep before the OpenAI add. AWS reported on its Q4 2025 earnings call that Bedrock revenue had grown more than three times year-over-year, with Anthropic’s Claude family driving a disproportionate share of consumption. Internal AWS forecasts circulated to partners projected Bedrock to cross a $5 billion run rate by the end of 2026 even without OpenAI. With GPT-5.5 in the catalogue, that target now looks conservative.

The Stateful Runtime Environment: AWS’s Agent Bet

The technical centerpiece of the AWS–OpenAI announcement is the co-developed Stateful Runtime Environment, an agentic execution layer that AWS describes as the “operational substrate” for OpenAI Frontier. The runtime exposes managed compute, memory and tool-use primitives so that long-running agents can pause, resume, and migrate across instances without losing context – a notoriously difficult engineering problem that has bedevilled production deployments of LangGraph, CrewAI and similar frameworks.

Under the hood, the Stateful Runtime is built on Bedrock AgentCore, AWS’s managed agent orchestration service launched in late 2025, and integrates with OpenAI’s Responses API. Customers will be able to deploy an OpenAI-built agent inside their AWS account, attach IAM-scoped tool permissions, and bill per-token plus per-second compute through a unified Bedrock invoice. Pricing details have not been disclosed, but AWS has signalled that the runtime will charge a base agent-hour fee on top of underlying model token costs, similar to how AWS Lambda charges per-invocation plus per-millisecond compute.

The runtime is positioned as a direct competitor to Microsoft’s Azure AI Foundry agent platform, which has dominated enterprise agent deployments since its 2025 reposition. By giving OpenAI a deeply integrated agent runtime on AWS, the deal effectively forks the agent-platform market: customers building OpenAI-first agents now have two equally credible managed options, with Bedrock’s broader model catalogue tipping the scales for multi-model orchestration use cases.

Microsoft’s Response and the New Azure Strategy

Microsoft’s public response to the April 28 launch was measured. In a blog post the same day, Azure CTO Mark Russinovich emphasized that Azure remains “the primary cloud partner for OpenAI” through 2032 under the restructured agreement, and pointed to the company’s continued $87 billion fiscal-year 2026 capex commitment, much of it earmarked for AI capacity. Azure OpenAI Service, Russinovich noted, retains “deep, exclusive integrations” with Microsoft 365 Copilot, GitHub Copilot, Dynamics 365 and the Microsoft Security Copilot stack – integrations that Bedrock cannot replicate without parallel investment.

Behind the scenes, Microsoft’s strategy has been pivoting for months. The company’s MAI in-house model program, which surfaced publicly in early 2026, is explicitly designed to reduce dependency on OpenAI for first-party Copilot workloads. Satya Nadella told analysts on Microsoft’s January 2026 earnings call that “diversification of model providers, including our own, is a strategic priority for fiscal year 2026.” The company has also expanded its partnership with Anthropic, integrating Claude into Microsoft 365 Copilot in late 2025, and now deploys Claude across multiple Copilot surfaces alongside MAI-1 and OpenAI’s models.

Wall Street’s read is that the loss of exclusivity is priced in. Microsoft shares dipped 1.4% on April 28 before recovering to close down 0.6%, while Amazon shares climbed 3.2% on the announcement and held the gain through the week. Analysts at Morgan Stanley wrote that the multi-cloud move “modestly compresses Azure’s AI growth premium” but is “more than offset by reduced concentration risk and a clearer path to OpenAI IPO outcomes” for Microsoft as a 27% equity holder in OpenAI’s restructured public-benefit corporation.

Competitive Comparison: Anthropic, Google and the AWS Catalog

The most uncomfortable party in the AWS–OpenAI announcement is arguably Anthropic. AWS has been Anthropic’s primary cloud partner since 2023, anchored by a $13.8 billion total investment commitment and a multi-gigawatt Trainium build-out aimed at making Bedrock the cheapest place on earth to run Claude. With OpenAI now in the same catalogue, Anthropic loses its status as Bedrock’s flagship frontier model – though Claude Opus 4.7 still leads several enterprise benchmarks against GPT-5.5, and Anthropic’s deep AWS integration on Trainium gives it cost-per-token advantages that OpenAI does not currently match.

“Bedrock now hosts every frontier model except Gemini,” IDC research director Ritu Jyoti noted in a client briefing. “The competitive question for Anthropic is no longer ‘how do we win against Claude vs GPT,’ it’s ‘how do we keep our share of Bedrock consumption when an enterprise customer can A/B test all three on the same invoice.'” Anthropic’s share of Bedrock token volume sat at roughly 38% in Q1 2026 according to channel data; analysts expect that to compress to 25–30% by year-end as OpenAI ramps.

Google Cloud, the third hyperscaler, was conspicuously absent from the OpenAI announcement. GCP has Gemini 3, the only frontier model not available on Bedrock, but its third-party model catalogue remains thinner than AWS’s and its Vertex AI agent platform has lagged Bedrock AgentCore in enterprise adoption. Google’s $40 billion Anthropic investment – closed in March 2026 – remains the company’s primary hedge, and Google has signalled no intent to pursue OpenAI distribution rights.

Hyperscaler AI Catalog	OpenAI	Anthropic	Google Gemini	Meta Llama	Mistral
AWS Bedrock	Yes (Apr 2026)	Yes (exclusive)	No	Yes	Yes
Azure OpenAI Service / Foundry	Yes (primary)	Yes (Copilot)	No	Yes	Yes
Google Vertex AI	No	Yes	Yes (exclusive)	Yes	Yes
Oracle OCI	Yes (via partnership)	No	No	Yes	Yes
IBM watsonx	No	No	No	Yes	Yes

Pricing: How Bedrock Stacks Up Against Direct OpenAI API

One of the recurring complaints about Azure OpenAI Service has been that its pricing tracks OpenAI’s direct API, eliminating any cost advantage from going through a hyperscaler. Bedrock launches with the same constraint at the per-token level – GPT-5.5 costs the same on Bedrock as it does on OpenAI’s direct API – but the surrounding economics differ materially.

Bedrock customers can apply existing AWS Enterprise Discount Programs to OpenAI consumption, layering negotiated discounts of 5–25% on top of list pricing depending on commitment size. Provisioned throughput contracts, which guarantee dedicated capacity and predictable latency, can drive effective costs lower for high-volume workloads. AWS Trainium-served models – including some of the open-weight gpt-oss variants – are priced at a 30–40% discount to comparable NVIDIA-served pricing, giving Bedrock a meaningful cost advantage for inference-heavy use cases where the open weight model is sufficient.

For high-volume customers, the indirect savings can be meaningful. AWS’s published case studies indicate that enterprises consolidating multi-model workloads on Bedrock typically realize 15–25% lower total cost of ownership versus running the same workloads split across OpenAI direct, Anthropic direct, and a hyperscaler – primarily because they unify billing, monitoring, and identity into a single AWS contract. With OpenAI in the catalogue, that consolidation argument becomes considerably easier to make.

Five Predictions for the Multi-Cloud OpenAI Era

The Bedrock launch is the start of a multi-year shift, not its endpoint. Five predictions for how the next 12–24 months play out:

• Azure OpenAI growth will compress materially in 2026. Expect Azure OpenAI revenue growth to decelerate from its current 80%+ year-over-year pace to 40–50% as enterprise customers diversify. Microsoft’s $87 billion FY2026 capex remains intact, but the mix shifts toward MAI and first-party Azure AI Foundry workloads.
• Bedrock revenue will cross $10 billion run rate by Q4 2026. AWS’s internal forecast of $5 billion was set before OpenAI was confirmed; doubling that target is realistic given OpenAI’s existing customer base and the ease of migration through unified SDKs.
• Google Cloud will counter with an OpenAI deal of its own. By the end of 2026, Google Vertex AI is likely to announce a comparable distribution agreement for OpenAI, ending Bedrock’s third-party exclusivity. Sundar Pichai has already signalled willingness to host competing frontier models, and OpenAI has a strong incentive to maximize cloud distribution.
• Anthropic will pursue Azure OpenAI parity. Anthropic’s $200 billion valuation depends on growing past its AWS-anchored consumption base. Expect a deeper Microsoft partnership in late 2026 – possibly involving Microsoft equity in Anthropic – to bring Claude into Azure OpenAI’s primary catalog rather than just Copilot integrations.
• Stateful agent runtimes will become the default deployment surface. The Bedrock Stateful Runtime and Azure AI Foundry agent platform will commoditize agent hosting within 18 months, pushing pricing toward $0.10–0.30 per agent-hour and accelerating the shift away from raw model API consumption toward managed agent endpoints.

Historical Context: From 2019 Lock-In to 2026 Multi-Cloud

The arc of the OpenAI–Microsoft relationship is, in retrospect, one of the defining business stories of the AI era. Microsoft’s $1 billion investment in 2019 was unusual in two ways: it was paid largely in Azure compute credits rather than cash, and it came with a clause making Azure the exclusive cloud for OpenAI’s models. At the time, OpenAI was a research lab with no commercial product, and exclusivity was a reasonable price to pay for guaranteed compute. The 2023 $10 billion follow-on cemented that arrangement just as ChatGPT was crossing 100 million users.

By 2024, the constraints of exclusivity were already chafing. OpenAI publicly signed deals with Oracle for additional capacity – technically permissible under a “compute-only” carve-out – and the company’s Stargate joint venture with SoftBank and Oracle, announced in early 2025, marked the first major break from Microsoft-only infrastructure. The October 2025 restructuring that converted OpenAI’s commercial arm into a public-benefit corporation included specific language unblocking multi-cloud distribution, setting the stage for the AWS deal a month later.

The April 2026 Bedrock launch closes a 7-year arc. OpenAI now has compute relationships with all three major U.S. hyperscalers (Microsoft, AWS and, indirectly through Stargate, Oracle), plus dedicated capacity from CoreWeave, Lambda Labs and a growing set of neoclouds. The exclusivity that defined the early commercial AI era is gone; the multi-cloud era has begun.

What Enterprise Customers Should Do Now

For enterprise architects and procurement teams, the launch creates immediate decisions. The first is whether to migrate existing Azure OpenAI workloads to Bedrock. For most production deployments, immediate migration is not warranted – Azure OpenAI remains stable, deeply integrated with Microsoft tooling, and competitively priced. But new workloads being scoped in Q2 and Q3 2026 should evaluate Bedrock as a serious primary option, particularly if they involve multi-model orchestration, agentic patterns, or AWS-native data sources.

The second decision is contract structure. Both AWS and Microsoft will be willing to negotiate aggressively over the next 12 months as they fight for share of OpenAI consumption. Customers with material AI spend should request multi-year, multi-cloud quotes from both vendors and use the new competition to extract better commercial terms. AWS in particular is offering bridge incentives – including portability credits and migration funding – to accelerate adoption.

The third decision concerns architecture. The Stateful Runtime Environment, when it ships, will create a fork in agent deployment patterns. Customers building agents today should design their orchestration layer to be runtime-portable – using OpenAI’s Responses API rather than Bedrock-specific calls where possible – to preserve optionality between Bedrock and Azure AI Foundry as both platforms mature.

Frequently Asked Questions

When did OpenAI’s models become available on Amazon Bedrock?

OpenAI’s models entered Amazon Bedrock in limited preview on April 28, 2026, with general availability rolling out “within weeks.” The launch followed the expiration of Microsoft Azure’s exclusive hosting rights on April 27, 2026.

Which OpenAI models are on Bedrock?

At launch, Bedrock hosts GPT-5.5, GPT-5.4, gpt-oss-20b and gpt-oss-120b. Codex is documented in the Bedrock catalog but ships as part of the OpenAI Frontier agent platform rather than the standard model catalog. GPT-5.5 Pro and o-series reasoning models are expected to follow.

Does this end Azure OpenAI Service?

No. Azure remains the “primary cloud partner” for OpenAI through 2032 under the restructured agreement and retains deep integrations with Microsoft 365 Copilot, GitHub Copilot and Dynamics 365. What ended on April 27, 2026 was Azure’s exclusivity – Bedrock is now the second hyperscaler authorized to distribute OpenAI’s frontier models.

Is Bedrock cheaper than Azure OpenAI for GPT-5.5?

Per-token pricing is identical to OpenAI’s direct API on both platforms, so list prices for GPT-5.5 match. Total cost of ownership tends to favor the platform where the customer already has volume commitments, since both AWS and Microsoft offer enterprise discounts on top of list pricing. Bedrock can be meaningfully cheaper for open-weight gpt-oss workloads served on AWS Trainium.

What is the Stateful Runtime Environment?

The Stateful Runtime Environment is a co-developed managed agent runtime, built on Bedrock AgentCore, designed to host long-running OpenAI-built agents with persistent context, compute and memory. It is the technical foundation of the OpenAI Frontier enterprise agent platform and is scheduled to launch within “the next few months” after April 28.

How does this affect Microsoft’s $13 billion OpenAI investment?

Microsoft retains an approximately 27% equity stake in OpenAI’s restructured public-benefit corporation following the October 2025 conversion. The Bedrock launch does not affect equity, but it removes Microsoft’s exclusive distribution control over OpenAI’s API revenue, which should compress Azure’s AI growth premium while increasing OpenAI’s overall enterprise reach – a mixed outcome for Microsoft as both customer and shareholder.

What does this mean for Anthropic on AWS?

Anthropic remains a major partner for AWS, with $13.8 billion in committed investment and dedicated Trainium capacity. However, OpenAI’s arrival on Bedrock creates direct catalog competition for Claude. Channel analysts expect Anthropic’s share of Bedrock token consumption to compress from roughly 38% in Q1 2026 to 25–30% by year-end as OpenAI ramps.

Will Google Cloud get OpenAI models too?

Not in the initial agreement. AWS holds exclusive third-party distribution rights for OpenAI Frontier, the enterprise agent platform. Standard model distribution is non-exclusive on AWS, so a future Google Vertex AI deal is technically possible – and is widely expected within 12–24 months – but has not been announced.

Related Coverage

• Amazon’s $5B Anthropic Bet: 5GW Trainium and $100B AWS Pact
• Google’s $40B Anthropic Bet: 5GW TPU Compute, $350B Valuation
• Anthropic vs OpenAI 2026: 30x Revenue Gap and 4x Context Divide
• AWS vs Azure 2026: 31% vs 24% Market Share and a 75% Archive Cost Gap
• Microsoft’s In-House AI Models: Inside the MAI Strategy
• Claude vs ChatGPT 2026: 80.8% vs 77.2% SWE-Bench and a 2x API Price Gap
• Best AI Models 2026: The Complete Comparison Guide

Sources: AWS Bedrock OpenAI announcement, OpenAI Amazon partnership, AWS $38B compute deal announcement, OpenAI GPT-5.5 launch, The Stack reporting on the Bedrock launch.