# ChatGPT Enterprise on AWS Marketplace: Pricing, Commitments and Data Residency

Buying ChatGPT through AWS simplifies procurement, but the inference-residency restriction deserves a separate review.

By Paras Madan | 2026-11-10


ChatGPT Enterprise became available through AWS Marketplace on November 9. You can buy through an existing AWS relationship and apply eligible spending toward commitments, but this route has a significant technical restriction: inference residency is available only in the United States for these workspaces.

That makes this a procurement decision with an architecture review attached. A team that wants a simpler invoice may find it useful. A team that requires European GPU processing should resolve that requirement before asking procurement to move the purchase.

## What changes when you buy through AWS Marketplace?

[OpenAI's announcement](https://openai.com/index/chatgpt-enterprise-aws-marketplace/) says Marketplace customers receive the same models, features and functionality as direct customers, with new features arriving at the same time. The offering includes ChatGPT, ChatGPT Work, Codex and dots. Eligible customers can also access the Healthcare and Financial Services offerings through this channel.

The practical gain is using a purchasing relationship that already exists. That can reduce the number of commercial steps between a team's evaluation and an approved rollout. It does not remove the need to review what employees and agents can access.

## Pricing and AWS commitment eligibility

The [official Marketplace listing](https://aws.amazon.com/marketplace/pp/prodview-sscra4m7iracg) uses private offers rather than a public per-seat price. Pricing depends on requirements and eligibility. Eligible consumption can count toward AWS EDP or PPA commitments, subject to the customer's agreement.

There is no defensible universal discount to calculate from that listing. Before comparing a direct proposal with a Marketplace proposal, put the actual offer terms side by side: total commitment, included usage, overages, renewal terms and which charges count toward the AWS agreement. Ask the account team to confirm those points in writing rather than treating the Marketplace badge as the answer.

Consider a company with an existing AWS spending commitment and a separate budget for employee AI tools. Moving an otherwise identical purchase into an eligible channel could help use that commitment. It does not automatically reduce the invoice, and it would be poor economics to buy more seats than the business needs just to consume a budget.

## Does purchasing through AWS keep everything inside AWS?

**No. Buying through AWS Marketplace does not guarantee that all customer data stays on AWS infrastructure or inside your AWS environment.** The listing explicitly identifies OpenAI as the data processor and AWS among its subprocessors. The product experience remains managed by OpenAI.

That distinction matters when someone describes the purchase internally as “ChatGPT in our AWS account.” A billing relationship is not evidence of a private deployment. Security reviewers should ask where the particular workload runs, which subprocessors handle its data and what the contract actually guarantees.

## Data residency versus inference residency

OpenAI's [residency documentation](https://help.openai.com/en/articles/9903489-data-residency-and-inference-residency-for-chatgpt) distinguishes storage at rest from GPU execution. Marketplace workspaces can use the listed data-residency regions for in-scope content, but their inference residency is US-only. The broader page lists Europe, the US and the UAE for supported inference-residency configurations; those broader options should not be assumed to apply to this purchase route.

Residency also has a defined scope. External integrations and some non-GPU processing can take place outside the chosen region. The documentation lists feature exclusions, and dots and Sites have separate qualifications. A storage-region selection is therefore insufficient evidence that every step of an agent workflow stays there.

A useful review separates three questions:

| Requirement | What to verify |
| --- | --- |
| Store supported customer content in a chosen region | The workspace's data-residency configuration and covered features |
| Keep supported GPU inference in-region | Separate inference-residency eligibility, including the Marketplace US-only restriction |
| Keep every tool call and processing step in-region | Each external integration and processing path, rather than assuming residency covers them |

For example, a team in Europe could have one requirement about saved conversation content and another about where GPU inference happens. Approving the first requirement does not approve the second. That is the conversation to have before the commercial offer is accepted.

## ChatGPT Enterprise or OpenAI models through Bedrock?

These purchases serve different jobs. ChatGPT Enterprise gives employees an application and agent experience. OpenAI models through Amazon Bedrock support developers building applications and agents. The new Marketplace announcement describes them as complementary.

A useful decision starts with the user of the system. If colleagues need a shared enterprise assistant experience, evaluate ChatGPT Enterprise. If your product needs model calls inside its own workflow, evaluate the model-serving route and its controls. A company can need both, but one budget line should not obscure two different technical evaluations.

## What to confirm before buying

Request the private offer, confirm commitment treatment with the AWS account team, and have the workspace owner document the required storage and inference regions. Then review the actual features the pilot will use, especially connected apps and agent workflows, against those requirements.

The strongest reason to choose this channel is a better purchasing fit for an already suitable product. **Do not let procurement convenience decide your data architecture.**

I'm Paras. I work at the intersection of AI, engineering and distribution.

I help enterprise AI teams explain technical products and their tradeoffs clearly to developers and decision-makers.

[Book a 30-minute call](https://cal.com/paras-madan/30min).