AWS credits for AI startups: what you can get in 2026, and who files each one.

IWhich AWS programs fund AI startups, and who files each one?

Six AWS programs matter. A partner files three of them for a specific project. The other three are AWS Activate tiers, and no partner can file those.

IIWhich funding can a partner file for an AI startup?

Your situation decides the route. Your matched partner files it, and AWS decides.

• You want to test an AI feature: That is a proof of concept (POC). The test qualifies and AWS approves it. Examples: a support assistant on Claude in Amazon Bedrock, tested on real past tickets, or a fine-tuned model on Amazon SageMaker. You do not need to run on AWS yet. More on AWS POC credits.
• You already run on AWS and are launching new AI work: That is incremental workload (IW) funding. The new workload qualifies and AWS approves it. It fits a new AI feature, product, customer or contract. AWS does not describe IW on a public page, so ask your partner how it applies. More on IW funding.
• You run on another cloud or GPU host and want to move to AWS: That is MAP (Migration Acceleration Program) funding. The migration qualifies and AWS approves it. It fits moving training, inference or the whole product. More on MAP for startups.

AWS says partner funds help "explore proof of concepts, and accelerate migration projects" (AWS partner funding page). The money is tied to a real project the partner delivers with you. So the request names the work, the AWS services and the expected usage.

IIIWhich route gives an AI startup more money?

For most startups without a VC, the partner route. If an Activate Provider backs you, apply for Portfolio as well.

Most startups have no VC or accelerator behind them. Their Activate tier is Founders: $1,000 to start, up to $5,000. A POC or IW request a partner files for a real AI project is $50K to $150K for startups in CloudRoute's partner records. Depends on the workload; AWS approves. The amount follows the project.

If an Activate Provider backs you, AWS Activate Portfolio is up to $200,000. Raised from $100,000 to up to $200,000 in June 2026. Apply for it too, and tell your partner which credits you already hold. AWS Credits for AI Startups is $200,000 or more, by invitation, for AI startups AWS sees as ready to scale after Portfolio. Ask your account manager.

Some sites list a $300K Activate tier for AI startups. AWS's credits page lists three tiers, and none of them is $300K. Partner funding programs reach about $1M, but that tier is for enterprises such as banks, hospitals and municipalities, not startups.

IVHow do you apply for AWS Activate as an AI startup?

You apply yourself on AWS's site. No partner is involved.

• Check that you are eligible: Pre-Series B, founded in the last 10 years, with an AWS account on a paid plan. You must be new to Activate credits or ask for more than you got before. (AWS credits page)
• Create an AWS Builder ID and pick your tier: Self-funded: AWS Activate Founders. Backed by a VC, accelerator or angel that is an Activate Provider: AWS Activate Portfolio, with their Org ID. Activate Portfolio applicants must apply within 12 months of their latest funding date.
• Describe the startup, link your AWS account and submit: AWS says Activate applicants hear back within 5 to 10 business days.

Since April 2024, Activate credits also pay for third-party models on Amazon Bedrock, such as Claude and Llama (AWS Startups blog).

VWhat can AWS credits pay for on an AI workload?

Most of an AI stack on AWS. Credits pay eligible charges on your AWS bill until they run out or expire.

AWS says Activate covers Amazon Bedrock, Kiro and 200+ AWS services (AI for startups on AWS). For an AI startup that means model calls on Bedrock, training and hosting on SageMaker or EC2, vector search, storage and the rest of the product.

The AWS promotional credit terms (updated September 24, 2026) exclude some charges unless AWS authorizes them. The ones AI teams meet most: AWS Marketplace, AWS Professional Services, and the upfront fees for Savings Plans and Reserved Instances. Plan to pay those in cash.

Funding a partner files covers the project in the request. Ask which services and which period it covers before the work starts.

Model-provider credits are separate. Anthropic says Claude for Startups credits cannot be used on AWS Bedrock (Anthropic program page). On Bedrock, AWS credits pay for Claude. More on AWS Bedrock credits and Claude for Startups.

VIHow long will AWS credits last on an AI workload?

Divide your credits by your monthly AWS bill. The bill is the part to estimate.

Bedrock prices model calls per million input tokens and per million output tokens, and the price depends on the model (Amazon Bedrock pricing). For a month, multiply tokens per request by requests per day and by 30. Divide by one million, then multiply by the model's price per million tokens. Do input and output separately. Then add vector search, storage, GPU or SageMaker hours, and logs.

With round numbers: a monthly AWS bill of $2,000 against $5,000 of credits lasts about two and a half months. Against $50,000, it lasts about two years, if the credits do not expire first.

• Pick the smallest model that passes your test: Prices vary a lot by model. Test a smaller model first and move up only where it falls short.
• Use batch for work that can wait: AWS prices batch below on-demand for select models. Evaluations and nightly jobs fit. More on Bedrock batch inference.
• Cache repeated context: Prompt caching lowers the cost of resending the same long instructions or documents. More on Bedrock prompt caching.
• Try AWS chips for steady work: AWS builds Trainium for training and Inferentia for inference. Test them on your model first. More on Trainium and Inferentia.
• Watch the burn rate: AWS Cost Explorer forecasts spend up to 18 months ahead (monthly view). (AWS docs) Set a budget alert too.
• Know your expiry date: AWS says Activate credits usually expire within 1 to 2 years, depending on the package. Credits you have not used by then are lost.

VIIWhat does a partner need from you to scope an AI project?

A clear plan for the test or the launch. The partner turns it into the funding request.

• What you want to test or launch: One feature or workload, in a few plain sentences.
• How you will judge it: A test set built from real data, a baseline to beat, and the score that would make you ship. "We will evaluate as we go" is hard to scope.
• Models, services and usage: The models you plan to try on Bedrock or SageMaker, what else the workload needs, and a monthly estimate of tokens, GPU hours or AWS spend.
• Data and region: Where the data comes from, whether it holds personal data, and which AWS Region it must stay in.
• Who builds it, and when: Your team, the partner or both, and when you want to start.

The partner then files a fund request in AWS Partner Central. For POC funding, AWS's fund request guide lists an AWS review and a technical approval stage, where a Partner Solutions Architect checks that the plan is feasible. The partner can start once the request is pre-approved. AWS publishes no review time for it.

Ask about funding before the work starts, not after.

VIIIWhen does an AI startup not need a partner?

Often. Here is the honest list.

• You are self-funded and still prototyping: Apply for AWS Activate Founders yourself ($1,000 to start, up to $5,000). Come back when you have a feature to test or launch.
• An Activate Provider backs you: Apply for AWS Activate Portfolio with their Org ID (up to $200,000).
• You have an AWS account manager: Ask about AWS Credits for AI Startups. It is invite only.
• You only want to try models: Start with the AWS Free Tier, then apply for Activate Founders when you build for real.
• You are moving to AWS and want to start alone: AWS offers migration investments, as AWS credits or partner investments, to offset one-time migration costs. Start with AWS's migration assessment for startups.
• You want to pick a partner yourself: Use AWS's Find an AWS Partner directory. Ask each firm whether it can file POC, IW or MAP funding today, and what it charges for the work.

If none of these fit and you have a real AI project, one conversation with a partner tells you whether it qualifies. CloudRoute picks that partner for you, for free.