Listing an AI model on a cloud marketplace is not one path but several: a trained model, an autonomous agent and a dataset each list through a different mechanism, with different pricing and a different deploy flow. Picking the wrong one is the most common reason a first AI listing stalls.
If you are an AI-native ISV preparing your first marketplace listing, the instinct is to treat “list our AI product” as a single task. It is not. The marketplaces have pulled apart into separate listing paths for separate kinds of AI: a fine-tuned model is not listed the way an agent is, and neither is listed the way the data behind them is. Each path has its own product type, its own pricing options, and its own way a buyer consumes what you shipped.
Get the mapping right and the listing is a well-worn process with vendor documentation for every step. Get it wrong — list a model where the marketplace expects an agent, or price a dataset as if it were software — and you burn a review cycle discovering the shape does not fit. Here are the current paths on AWS and Google Cloud, what each one is for, and how to decide which your product is.
How do you list an AI model on AWS Marketplace?
You list a trained AI model on AWS Marketplace as an Amazon SageMaker AI product, in one of two types. AWS’s seller guide is explicit that there are exactly two SageMaker AI product types a seller can publish:
- Model package — “a pre-trained model for making predictions that does not require any further training by the buyer.” This is the path for a model you have already trained and want buyers to run as-is.
- Algorithm — “a model that requires the buyer to supply training data before it makes predictions. The training algorithm is included.” This is the path when the value is your training method, applied to the buyer’s own data.
The distinction is not cosmetic. It decides what the buyer does after subscribing, and — as the pricing section below shows — what you are allowed to charge for. A model package is inference only; an algorithm can also charge for training runs.
Either way, the buyer never receives your weights as a file to run on their own infrastructure. They subscribe through the SageMaker AI console or AWS Marketplace, the product lands in their SageMaker product list, and they “create a fully managed REST inference endpoint or perform inference on batches of data” inside their own AWS account. Your model runs where the buyer’s data already sits, under their account, on infrastructure they pay AWS for directly.
What can you charge for an AI model on AWS?
AWS splits an ML product’s cost into two parts, and you only set one of them. The buyer always pays AWS directly for the SageMaker infrastructure the model runs on; you set the software price on top. For the software price, AWS documents three models plus a trial:
| Pricing model | How it bills | Best for |
|---|---|---|
| Free | Buyer pays only AWS infrastructure costs | Adoption, open models, lead generation |
| Hourly | Price per hour per instance type, prorated to the minute | Long-running or high-throughput inference |
| Inference | Price per endpoint invocation (only requests returning 2XX) | Pay-per-prediction, sporadic usage |
| Free trial | 5–31 days, buyer pays infrastructure only, then converts | De-risking evaluation before commitment |
Two mechanics matter when you model the economics. With inference pricing, if a single invocation processes a mini-batch, you can charge for more than one inference by returning a custom metering header — AWS’s example, X-Amzn-Inference-Metering: {"Dimension": "inference.count", "ConsumedUnits": 3}, bills three inferences for one call. And for algorithm products, on top of inference you also set an hourly price for training jobs, per instance type — the extra revenue line a model package does not have. Batch transform jobs, asynchronous inference endpoints and algorithm training jobs always bill hourly, independent of your inference price.
Comparison table: model, agent and data-product listing paths
The single most useful thing to internalize before you start is that “AI product” resolves to at least three different listings, each with its own type, pricing and consumption model. Here is how the paths line up across AWS and Google Cloud.
| AI model | AI agent | Data / knowledge product | |
|---|---|---|---|
| What it is | A trained model buyers run for predictions | An autonomous system that pursues goals on the buyer’s behalf | A dataset or knowledge source the product reasons over |
| AWS path | SageMaker AI product — model package or algorithm | AI/ML solution listed as SaaS or container; agents via the AI Agents & Tools category | Data product (data exchange listing) |
| Google Cloud path | Deployed via Vertex AI / SaaS listing | ”Offer AI agents” — the dedicated A2A-protocol path | ”Offer data products” |
| How the buyer consumes it | Runs inference in their own account (SageMaker) | Invokes the agent, discoverable through surfaces like Gemini Enterprise | Subscribes to and queries the data |
| Pricing shape | Free, hourly, or per-inference (+ training for algorithms) | Free, subscription, usage-based, or combined | Subscription or usage-based |
The takeaway is not that one marketplace is better; it is that your product almost certainly fits one of these rows cleanly and fits the others badly. A fine-tuned scoring model is an AWS SageMaker model package. A multi-step assistant that calls tools is a Google Cloud AI agent. The dataset that either one depends on is a data product. Trying to force all three into a single listing is where first attempts stall.
How do you list an AI agent on Google Cloud Marketplace?
Google Cloud Marketplace treats agents as their own listing path, separate from SaaS. The “Offer AI agents” path publishes agents that “use AI to pursue goals and complete tasks on behalf of users” and interoperate through the Agent2Agent (A2A) protocol — a different product than a SageMaker model, with different rules.
Google’s eligibility bar for an A2A agent listing is specific. The agent must be “customer-ready, secure, and built for enterprise ecosystems,” align with Google’s AI Principles, use Google or third-party models from Model Garden, and be discoverable through primary Google surfaces such as Gemini Enterprise. It must also exclude professional services, hardware, or non-software billable components — an agent listing is the software only.
The pricing options are broader than the AWS ML set because an agent is a running service, not just an inference endpoint: free (the customer pays only for the Google Cloud resources the agent uses), subscription (a flat recurring fee), usage-based (pay per measured metric), or a combination of a base subscription plus usage. The listing steps run through the Producer Portal — create the agent and its Agent Card, add the agent in Producer Portal, upload the Agent Card, add product and pricing details, integrate with Cloud Marketplace, and submit for review.
Google Cloud’s partner documentation is worth reading as a menu, because it names the paths side by side: “Offer AI agents,” “Offer software as a service (SaaS) products,” “Offer virtual machine (VM) products,” “Offer Kubernetes apps,” “Offer data products,” “Offer professional services,” and “Offer free container images.” Choosing among them is the first decision, and it is the one this post exists to make deliberate.
Where do data products fit?
A data product is the third path, and often the one an AI listing quietly depends on. A data product lists the dataset or knowledge source your model or agent reasons over, rather than the model itself — and both AWS and Google Cloud give it a dedicated listing type distinct from software.
This matters for AI vendors specifically because a model is frequently only as valuable as the proprietary data behind it. If your differentiator is the corpus, listing it as a data product can be the better commercial move than burying it inside a model package — the buyer subscribes to the data directly, and you keep the model listing focused on the inference. It is also the natural home for retrieval and enrichment products that feed a buyer’s own AI. The mechanics of the data-product path, and how it transacts against committed spend, are covered in Snowflake Marketplace for data and AI products, which is the clearest of the three on how a data listing differs from a software one.
How do you sell across all three marketplaces at once?
You sell across all three by treating the AI product type — model, agent, or data — as one catalogue entry that you list, price, meter and reconcile in one place, instead of running each marketplace as its own project. The operational cost of AI marketplace distribution is not the first listing; it is keeping the fifth in sync with the first, across three different consoles, three metering formats and three disbursement schedules.
That is the problem Suger’s AI marketplace distribution solution is built for: package your AI product once — agent, MCP server, or knowledge base — then list, meter, and sell it through AWS and Google Cloud marketplaces without building the marketplace infrastructure yourself, so a buyer procures it with their existing cloud commitment. The same catalogue, offers and usage metering that run a software listing run the AI ones, described on the product listing and offer management page. And because the metering for a per-inference model or a usage-based agent has to reconcile against what each marketplace actually disburses, that side runs through marketplace billing and metering rather than a spreadsheet per marketplace.
Suger transacts on six marketplaces — AWS, Microsoft, Google Cloud, Snowflake, Alibaba Cloud, and Oracle — which is what lets a single AI product ship to more than one buyer surface without a new project each time. For the broader operating model that keeps a multi-marketplace listing from fragmenting, running one GTM motion across AWS, Azure and Google Cloud walks through the catalogue-and-reconciliation pattern in full.
Frequently asked questions
How do you list an AI model on AWS Marketplace? You list it as an Amazon SageMaker AI product, in one of two types: a model package (a pre-trained model buyers run as-is) or an algorithm (buyers supply training data first). Buyers subscribe through the SageMaker AI console and run inference in their own AWS account.
What is the difference between a model package and an algorithm on AWS? A model package is pre-trained and makes predictions with no further training by the buyer. An algorithm requires the buyer to supply training data, and includes the training algorithm. The algorithm path also lets you charge an hourly price for training jobs; a model package is inference only.
How much does it cost to list an AI model versus an AI agent? On AWS, an ML model uses free, hourly, or per-inference software pricing, and the buyer separately pays AWS for SageMaker infrastructure. A Google Cloud AI agent uses free, subscription, usage-based, or combined pricing. There is no listing fee to publish either.
Can you list an AI agent on Google Cloud Marketplace? Yes. Google Cloud has a dedicated “Offer AI agents” path for agents that interoperate through the Agent2Agent (A2A) protocol. The agent must use Model Garden models, be enterprise-ready, and exclude professional services or hardware. It lists through the Producer Portal with an Agent Card.
Should an AI product be listed as a model, an agent, or a data product? Match the product to the path: a trained model buyers run for predictions is a model listing, an autonomous system that completes tasks is an agent listing, and the dataset either one reasons over is a data product. Most first-time listings stall by choosing the wrong one of the three.
How do you list an AI product on more than one marketplace? Package it once as a catalogue entry, then list, meter and reconcile it across marketplaces from one place rather than running each console separately. Suger’s AI marketplace distribution does this across AWS and Google Cloud so a buyer procures with their existing cloud commitment.
Takeaways
- “Listing an AI product” is not one path. A trained model, an autonomous agent, and the data behind them each list through a different mechanism — match the product to the path before you start.
- On AWS, an AI model is a SageMaker AI product: a model package (pre-trained, inference only) or an algorithm (buyer supplies training data, and you can also charge for training). Buyers run it in their own account.
- AWS ML software pricing is free, hourly, or per-inference, with a 5–31 day free trial; the buyer always pays AWS separately for SageMaker infrastructure.
- Google Cloud has a dedicated “Offer AI agents” path for A2A-protocol agents — enterprise-ready, Model Garden-backed, software only — priced free, subscription, usage-based, or combined.
- The cost of AI marketplace distribution is keeping the fifth listing in sync with the first. Run one catalogue, one metering model and one reconciliation surface across marketplaces instead of a project per console.
Suger lets an AI-native ISV package a model, agent, or data product once and list, meter and sell it across AWS and Google Cloud without building marketplace plumbing. See how on the AI marketplace distribution solution, or start with product listing and offer management.
Sources
Primary sources for the platform rules cited above. Last verified August 19, 2026. Cloud providers change fees, eligibility, and program terms without notice — check the source before relying on a figure.
- Machine learning products in AWS Marketplace — The two SageMaker AI product types — model package (pre-trained) and algorithm (buyer supplies training data) — and that buyers deploy them through the SageMaker AI console as a REST inference endpoint or batch job.
- Machine learning product pricing for AWS Marketplace — Free, hourly (per instance type, prorated to the minute), and inference (per invocation, 2XX only) software pricing, the buyer-pays-infrastructure split, and the 5–31 day free trial window.
- Offer AI agents through Google Cloud Marketplace — The A2A-protocol agent listing path, its eligibility rules (Model Garden models, no professional services or hardware), the free/subscription/usage/combined pricing options, and the Producer Portal Agent Card steps.
- Google Cloud Marketplace Partners documentation — The distinct listing paths a partner can choose — AI agents, SaaS, VM, Kubernetes apps, data products, professional services, and free container images.
Stay Updated
Get the latest Cloud GTM insights, product updates, and marketplace strategies delivered to your inbox.