Up to $300,000 in AWS credits is on the table if you qualify for the AWS Activate Generative AI tier. These AWS Activate credits can be spent across 200+ AWS services, including Amazon Bedrock (with third-party models) and serious training/inference infrastructure like Trainium and Inferentia.
AI startup founders training or fine-tuning models, ML engineers running big experiments, and CTOs trying to keep compute from eating the runway tend to get the most out of this program. If you’re only building an API wrapper, honestly, you’ll likely get filtered out.
This guide breaks down eligibility, the exact signup flow, what’s covered (and not), the restrictions that trip teams up, and a few practical ways to stretch the credits further.
Program at a Glance
| Provider | AWS |
| Credit Amount | Up to $300,000 promotional credits |
| Duration | 2 years from activation |
| Eligibility | Pre-Series B, VC/accelerator Org ID, foundation-model work |
| Credit Card Required? | Yes, for AWS account signup |
| Difficulty | Advanced; nomination + strict GenAI criteria |
| Best For | FM training, GPU/Trainium clusters, Bedrock model usage |
| Official Page | AWS Program Page |
What You Actually Get
The AWS Activate Generative AI tier is an elevated allocation of up to $300,000 inside the AWS Activate Portfolio framework. The credits are redeemable across 200+ eligible AWS services, including Amazon Bedrock (with third-party models from Anthropic, Meta, Mistral, Cohere, and others), plus heavy compute options like EC2 GPU instances (P5 and G5) and AWS Trainium (Trn1) and Inferentia (Inf2). You can also apply credits to common foundation pieces like S3, EBS/EFS, RDS/DynamoDB/Aurora, and services like EKS/ECS and CloudFront.
In real terms, $300K is enough budget to run real experiments, not just a weekend prototype. It can cover training runs on Trainium, inference deployments on Inferentia, and a lot of iteration through Bedrock-hosted models while you figure out what actually works. The teams that get full value usually treat it like a two-year compute plan, with billing alerts and a clear runway of milestones.
Who Qualifies (and Who Doesn’t)
This tier is for AI startups building foundation models or critical generative AI infrastructure, not “thin” products sitting on top of someone else’s hosted model. AWS is explicit about that, and reviewers look for clear technical need for compute-heavy workloads. Nomination from a recognized VC partner is typically required.
- Your startup must be pre-Series B, and if you raised recently, your most recent round should be within the last 12 months.
- The company needs to be founded within the last 10 years.
- You need an active AWS account on a paid-tier plan, and the signup flow includes confirming a credit card and phone number.
- A functioning company website with a clear product description, plus a corporate email address (matching your domain), is required.
- You must have a valid, case-sensitive Organization ID from a recognized VC, accelerator, or AWS Activate Provider.
If you’re only wrapping GPT-4, Claude, or similar hosted models behind an API, you do not qualify for the GenAI tier. Government entities are excluded as well, and applying with Gmail/Yahoo is a fast rejection.
How to Sign Up
Plan for a quick initial submission, then a longer review cycle if you’re going for the higher allocation.
- Secure an Organization ID by contacting your VC, accelerator, or startup program for a valid Org ID (it is unique and case-sensitive), and ask what credit tier they can support before you apply.
- Set up your AWS account by signing up at aws.amazon.com using a corporate email that matches your website domain, then confirm your email, credit card, and phone number.
- Create an AWS Builder ID that links your application to your AWS account, and make sure the email is not tied to another Builder ID (duplicates are rejected).
- Go to the AWS Activate application page at aws.amazon.com/startups/credits.
- Select Activate Portfolio and fill out the application with your company name, website, description, AWS account ID, Org ID, and funding details.
- Describe your AI workloads with specifics on your foundation model work, compute needs (instance types, training pipeline), and how you plan to use Bedrock, Trainium, or Inferentia, because this description directly affects your allocation.
- Submit and wait; AWS often reviews applications within about 7–10 business days, but higher-tier approvals can take about 6–10 weeks.
- Once approved, credits appear in your AWS Billing Dashboard.
Two common gotchas: applying without a valid Org ID, and using an email that doesn’t match your domain. Another one people miss is vagueness; if your workload description reads like marketing copy, expect a lower allocation or a rejection.
What the Credits Cover
These credits apply automatically across eligible AWS services, which means you don’t “assign” a budget to Bedrock vs EC2 vs storage. The scope is broad: AI/ML tools, GPU and non-GPU compute, containers, data, and the boring-but-necessary pieces like networking and databases.
| Service / Feature | What It Does | Included? |
|---|---|---|
| Amazon Bedrock (third-party models) | Hosted foundation models for inference and prototyping. | ✓ |
| EC2 (P5, G5) + Trainium/Inferentia | Compute for training and inference workloads. | ✓ |
| SageMaker | ML development and training workflows. | ✓ |
| Core cloud services (S3, EBS/EFS, RDS, DynamoDB, EKS/ECS) | Storage, databases, containers, and infrastructure backbone. | ✓ |
Notable exclusions matter here: AWS Marketplace purchases are not covered (except Bedrock third-party models). Route 53 domain registrations and Reserved Instance upfront fees are also excluded.
Limitations to Know About
Every free-credit program has catches. This one is generous, but it’s picky, and the “fine print” affects budgeting.
- Credits expire 2 years from activation, and AWS cannot extend expiration dates even if you have unused credits.
- Credits are non-transferable, so you can’t pass them through to end customers.
- Credits only apply to new usage, not prior or retroactive bills.
- You cannot manually allocate credits to specific services; they are applied automatically across eligible services.
- AWS Marketplace purchases aren’t covered (Bedrock model usage is the exception).
- Some services have carve-outs like Route 53 domain registrations and Reserved Instance “All Upfront” fees.
When credits run out (or expire), AWS does not pause your infrastructure. You are charged standard pay-as-you-go rates for any running services, and AWS has no spending caps by default. AWS does send email reminders at about 60 days before, 30 days before, and after expiry, plus usage alerts when you hit about 75% and 100% consumption, but you still need to watch billing.
Have Unused AWS Credits?
AWS startup credits can pile up faster than teams can spend them, especially when priorities shift or a training plan gets simplified. And once the clock is running, unused credits are basically a decaying asset. If you have surplus AWS credits you won’t use before they expire, AI Credit Mart lets you sell unused credits instead of watching them go to zero.
Need More AWS Credits?
Once your free allocation is gone, paying full retail for GPUs, storage, and Bedrock calls adds up fast. AI Credit Mart lists discounted AWS credits from organizations with surplus allocations, so you can often extend your runway without changing providers. Discounts typically land around 30–70% below retail, depending on supply.
Tips for Getting the Most Out of Your Credits
- Ask your VC/accelerator what tier they can support before you apply, because not all providers can sponsor the same credit amount.
- Write the workload description like an engineer, not a pitch deck: name instance types (P5, G5, Trn1, Inf2), outline your training pipeline, and call out Bedrock usage if relevant.
- Don’t accidentally lock yourself into a lower allocation; the program notes that upgrading after triggering a low tier can be administratively difficult.
- Set up billing alerts immediately, because AWS has no spending caps and you will pay full price after credits are consumed.
- Use Bedrock batch inference pricing (about 50% lower than on-demand) for large evaluations to stretch your credits further.
- Check region availability for Bedrock models and Trainium instances before you commit your architecture, since not everything is available everywhere.
- If you want help maximizing allocation, AWS consulting partners (for example, Cloudvisor or nOps) can guide the application and billing setup.
Frequently Asked Questions
They’re worth up to $300,000 in promotional credits that automatically offset eligible AWS usage. In practice, that can cover large chunks of Bedrock model usage (including third-party models), EC2 GPU capacity (P5/G5), and Trainium/Inferentia workloads, plus the supporting stack like S3 and databases. The “worth it” part depends on whether you actually run compute-heavy training or inference over the next two years. If your product doesn’t need that level of infrastructure, you’ll leave value unused.
Yes. The AWS account setup step includes confirming a credit card.
These Activate credits expire 2 years from activation, and AWS won’t extend them.
Yes. If you have AWS credits you won’t use before they expire, you can list them on AI Credit Mart and sell them at up to 70% of face value. Companies regularly list surplus credits from startup programs and enterprise agreements.
AI Credit Mart has discounted AWS credits available from companies with surplus allocations. Prices are typically 30-70% below retail.
After expiration, you’re charged standard pay-as-you-go rates for any running services. AWS sends reminder emails around 60 days before, 30 days before, and after expiry, but it won’t pause your infrastructure for you.
Not exactly. Credits do not stack: if you already received $100K in Portfolio credits and later qualify for the Generative AI tier, AWS grants the difference (an additional $200K) to reach the higher allocation.
Using a personal email (Gmail/Yahoo) and looking like an API wrapper are the big ones. AWS also flags vague product/workload descriptions, missing or invalid Org IDs, and prior Activate credits at an equal or greater value. One more: your Builder ID email cannot be tied to another Builder ID, or AWS rejects it as a duplicate. Treat the application like a technical spec, not a branding exercise.
If you’re genuinely building foundation-model tech, the AWS Activate Generative AI tier is one of the bigger credit pools out there. Apply carefully, get the allocation you deserve, and don’t let unused credits expire.
Your AI credits are losing value every day
Join the marketplace and start trading unused credits today.