AI is expensive in a way most software is not. A prototype that calls a large model a few thousand times is cheap. A product with real users, a fine tuning job, and a GPU hungry training run is not. For many founders, the bill arrives long before the revenue does.
That is where AWS credits for AI startups come in. AWS offers promotional credits that can offset compute, storage, and even third party foundation models on Amazon Bedrock. Used well, they buy months of runway and room to experiment.
But the rules are not obvious. Amounts differ by program. Some credits need an investor’s code. Some programs are competitive. And every credit expires. This guide explains what is on offer, how to apply, what the credits cover for AI work, and how to make them last. You will learn:
- Which AWS credit programs exist for AI startups, and how much each offers
- Who is eligible, and how the application works step by step
- What credits cover for Bedrock, SageMaker, and GPU workloads, and what they do not
- How to cut training and inference costs so credits stretch further
- Mistakes that waste credits, and how to decide what to apply for first
All figures come from AWS’s own pages. Amounts and rules change, and AWS’s pages do not always agree, so confirm current numbers before you plan around them.
What Are AWS Credits for AI Startups?
AWS credits for AI startups are promotional balances that offset eligible AWS charges, including compute, storage, and third party foundation models on Amazon Bedrock. They come from programs such as AWS Activate and the Generative AI Accelerator, and they expire, so startups must use them before the end date.
Credits are not cash. You cannot withdraw them, sell them, or move them to another company. AWS’s promotional credit terms say credit can only be used on eligible services, applies to your own account, and is forfeited if it is not used before its expiration date.
For an AI startup, that last point matters more than usual. Model training and inference costs can climb quickly, so credits can vanish early. Or, if usage is light, they can expire unused. Either way, the plan has to start with the calendar.
AWS Credit Programs for AI Startups at a Glance
There is no single “AI startup credit.” Several programs exist, aimed at different stages. This table compares the main ones, using AWS’s own figures.
| Program | Credit amount per AWS | Who it is for | How you get it |
| Activate Founders | $1,000, with up to $5,000 over time for select startups | Self-funded startups with no provider | Apply online, no Org ID needed |
| Activate Portfolio | Up to $100,000 on AWS’s credits page, up to $200,000 in AWS’s June 2026 guide | Pre-Series B startups backed by an Activate Provider | Apply with your provider’s Org ID |
| Credits above $200,000 | Available for AI startups ready to scale | AI startups that have outgrown Activate tiers | Talk to your AWS Account Manager |
| Generative AI Accelerator | Up to $1 million | Early-stage generative AI startups | Competitive application |
The differences in the Portfolio row are real. AWS’s credits page lists up to $100,000, while its application guide lists up to $200,000. Your provider decides which package level you qualify for, so ask them for your actual number.
AWS Activate for AI Startups
AWS Activate is the main door. Its credits can offset more than 200 AWS services, and AWS says they can be redeemed on third party models in Amazon Bedrock from companies such as AI21 Labs, Anthropic, Cohere, Meta, Mistral AI, and Stability AI, plus Amazon’s own models, all through one API.
There are two tiers:
- Founders: for self-funded startups. It offers $1,000, and AWS says select startups may qualify for up to $5,000 over time.
- Portfolio: for startups connected to an Activate Provider, such as an accelerator, angel investor, or venture capital firm. You need the provider’s Org ID to apply.
AWS also says generative AI startups may be eligible for additional credit offerings beyond the standard tiers. Read more about the Org ID in AWS’s guide to Activate providers.
The AWS Generative AI Accelerator
The Generative AI Accelerator is a selective program for early-stage startups building with generative AI. For its 2025 global cohort, AWS selected 40 startups, said its acceptance rate was under 2%, and offered each one up to $1 million in AWS credits plus an 8-week program.
AWS launched the program as part of a $230 million commitment to generative AI startups. The 2024 cohort ran 10 weeks, with partners including NVIDIA and Mistral AI, and also offered up to $1 million in credits per startup.
Advice for applicants:
- Treat it as a stretch goal, not your only plan. Odds are low.
- Describe what your model or product does that is hard, not just which APIs you call.
- Show traction, a roadmap, and why AWS infrastructure fits your workload.
- Check AWS’s accelerator page for the current application window, since dates and terms change each cycle.
Credits Above $200,000
AWS says additional credits above $200,000 are available for AI startups that are ready to scale beyond Activate’s tiers, and it points founders to their AWS Account Manager. If you do not have one, AWS says you can reach it through AWS Activate Support. Expect AWS to ask about your workload, your growth, and your plan for production.
Which Program Should You Pursue First?
- Self-funded and pre-product: start with Activate Founders, and keep building.
- Backed by an accelerator or investor: ask for their Org ID and apply to Portfolio.
- Training or serving large models: apply to the Accelerator, and ask your account team about larger awards.
- Already received credits: you may be able to reapply for a higher amount if you still meet the requirements and are under the lifetime cap.
Who Is Eligible for AWS Credits for AI Startups?
Eligibility depends on your funding stage and whether an Activate Provider backs you. Here are the requirements AWS lists in its application guide.
Activate Founders
- New to Activate credits
- Self-funded
- A fully functioning company website
- Founded in the past 10 years
Activate Portfolio
- Associated with an Activate Provider, and holding their Org ID
- Has not previously received Activate credits of equal or greater value
- Pre-Series B, with the most recent funding round within the last 12 months if applicable
- A fully functioning company website
- Founded in the past 10 years
AWS’s credits page adds that startups need an AWS account on the Paid Tier Plan. If your account is still on the Free plan, upgrade it first. AWS says you can do this with the Upgrade Plan option in the console.
The Generative AI Accelerator has its own selection process, which weighs your product, team, and traction. Read the accelerator page for that year’s criteria.
How to Apply for AWS Credits for AI Startups Step by Step
This process follows AWS’s published checklist. Set aside about an hour, and have your company details ready.
- Create or sign in to your AWS Builder ID. Select Apply Now on the AWS credits page. Use a personal email for the Builder ID and verify it.
- Complete your Activate profile. Use a business email that matches your startup’s domain. AWS says consistent domain information helps validate your application.
- Choose your tier. Pick Founders if you are self-funded. Pick Portfolio if a provider backs you, and enter their Org ID.
- Provide startup details. Describe your product, target market, and traction. Give your funding stage and the date of your most recent round.
- Link your AWS account. Link the primary account where you want credits. You need administrator permissions, and only one Builder ID can link to an AWS account.
- Verify the account. Use the verification button in the console until you see that the accounts linked successfully.
- Review and submit. Double check funding stage and email domains. AWS says applications cannot be edited after submission.
What to Prepare Before You Start
A little preparation prevents delays. Gather these items:
- Your company website, with a working homepage that explains the product
- A business email on your own domain
- The exact date and amount of your latest funding round, if any
- Your provider’s Org ID, if you are applying for Portfolio
- An AWS account that is active, on the Paid Tier Plan, and set up by someone with admin rights
- A short, honest description of your AI product and how you plan to use AWS
Tips Specific to AI Startups
A few habits make your application clearer and easier to review:
- Be clear about what you are building, such as a model, an agent, or an AI feature in an existing product.
- Name the AWS services you expect to use, such as Amazon Bedrock, Amazon SageMaker, or GPU instances.
- Keep your numbers consistent across the Activate profile, your website, and your funding records.
- If you have a provider, ask them early for the Org ID and for their specific package levels.
How Long It Takes and What Happens Next
AWS says processing usually takes 7 to 10 business days, and the result arrives by email. You can follow progress on the Credit Application Status page. If you are approved, credits apply to your account automatically and show up in the AWS Billing and Cost Management console within a few hours.
If you wait more than 10 business days, check your spam folder and the status page, then contact AWS Activate Support. AWS also notes that credits typically sit at the AWS account level, so link the account that will actually run your workloads.
If You Do Not Have an Org ID
Apply for Founders first. AWS says that without an Org ID you are initially eligible for the Founders package, and with one you are eligible for that provider’s package levels. If you later join a program with a provider, you may be able to apply again for a larger amount, as long as you still qualify.
Be careful with anyone offering to sell you credits or an Org ID. AWS describes Org IDs as confidential codes provided by your accelerator, investor, or startup organization, and it says they must be entered on the official AWS Activate application page.
What AWS Credits Cover for AI Workloads
Credits offset eligible AWS charges, and AWS says Activate credits can apply to more than 200 services, including machine learning and AI. Here is how that plays out for the workloads AI startups run most.
Amazon Bedrock and Third Party Foundation Models
Amazon Bedrock gives you access to foundation models through a single API. AWS says Activate credits can be redeemed on third party models in Bedrock, from providers such as AI21 Labs, Anthropic, Cohere, Meta, Mistral AI, and Stability AI, along with Amazon’s own models.
This is a notable exception. AWS Marketplace charges are normally excluded from credits, but AWS makes an exception for charges that come from using third party foundation models on Bedrock.
Bedrock offers several billing modes, described on its pricing page: on-demand, batch inference, provisioned throughput, and prompt caching. Each mode has a different cost profile for the same model, which is why choosing the right one can stretch your credits much further.
Amazon SageMaker for Training and Hosting
Amazon SageMaker AI lets you build, train, and deploy models on managed infrastructure. Training jobs, notebooks, and hosted endpoints are billed as AWS usage, so they generally fall under eligible charges.
Two cost traps to watch. Real time endpoints bill for every hour they run, even when traffic is quiet. And notebooks left running keep charging. Credits will absorb these costs for a while, but they will still burn your balance.
GPU Compute on Amazon EC2
If you train or serve models on your own instances, GPU capacity is your biggest cost. Credits can offset eligible EC2 usage, including accelerated instances.
GPU supply can be tight. AWS offers EC2 Capacity Blocks for ML, which let you reserve GPU instances for a future date and a short duration. According to AWS, you can reserve a start time up to eight weeks ahead, and the total price is charged up front. Because AWS excludes upfront fees for Savings Plans and Reserved Instances from credits, confirm with AWS whether credits apply to a Capacity Block before you rely on them.
AWS Trainium and Inferentia
AWS designs its own AI chips. AWS said its Trainium based Trn1 instances offer up to 50% cost-to-train savings over comparable GPU-based EC2 instances, and its Inferentia2 chip can reduce inference costs by up to 50% against comparable instances. You can read more on the AWS Trainium page.
These are “up to” figures for specific models and settings. They also require using the AWS Neuron software, so test your model before you plan around the savings. For some teams the switch is easy. For others it is a real engineering project.
Data, Storage, and Networking
AI work is data work. Datasets in Amazon S3, volumes in Amazon EBS, and traffic between services all produce eligible charges. Data transfer is easy to forget, especially if you serve models to users across Regions or send results out to the internet. Track it from the start.
What Credits Do Not Cover
Per AWS’s credits FAQ, credits cannot be applied to:
- Amazon Mechanical Turk
- AWS Managed Services
- Ineligible AWS Support plans
- AWS Marketplace, except third party foundation model use on Bedrock
- AWS Professional Services, AWS Training, and AWS Certification
- Amazon Route 53 domain registration or transfer
- Cryptocurrency mining
- Upfront fees for Savings Plans and Reserved Instances
AWS’s terms also say credit is not applied to transaction-based taxes.
How Fast Can an AI Startup Burn Through Credits?
Very quickly, if you are not watching. The table below shows how long a hypothetical $50,000 credit balance would last at different monthly spend levels.
| Monthly eligible AI spend | Months of runway from $50,000 |
| $3,000 | About 16.7 months |
| $6,000 | About 8.3 months |
| $12,000 | About 4.2 months |
| $25,000 | 2 months |
Notice the tradeoff. Low spend risks hitting the expiration date with credits left over. High spend risks running out in a few months. Compare your balance and your expiration date, then plan to land in the middle.
How AWS Credits Are Applied and When They Expire
AWS’s Activate guide explains the mechanics:
- Credits apply monthly to new usage charges on eligible services.
- They do not retroactively cover past bills.
- You cannot choose which services they apply to.
- They typically expire within one to two years, depending on the package.
- You can see the balance and expiration date in the Billing and Cost Management console.
AWS also emails Activate members 60 and 30 days before credits expire and again when they have expired, plus when credits are 75% and 100% used. When credits run out or expire, you pay standard rates for ongoing services. AWS says it cannot extend the expiration date, even if credits remain unused. For a deeper look, see AWS’s promotional credit terms.
How to Make AWS Credits for AI Startups Last Longer
Credits are a head start, not a plan. AI costs scale with usage, so every efficiency you add now stretches the balance and shrinks the shock on the first full price invoice. Start with the levers below.
Cut Amazon Bedrock Costs
AWS’s Bedrock pricing page describes the main ways to spend less on inference:
- Batch inference. For select models, AWS offers batch inference at a 50% lower price than on-demand. It suits work that does not need an instant answer, such as nightly document processing.
- Prompt caching. If many requests share the same long context, caching it reduces both cost and response time.
- Prompt optimization. Bedrock can rewrite prompts for better performance and shorter responses.
- Provisioned throughput. Reserved capacity can fit steady, high volume traffic, and fine tuned or distilled models need it.
Beyond those features, the biggest lever is model choice. Use a smaller, cheaper model for simple classification or extraction, and keep your largest model for tasks that truly need it. Keep prompts short, cap output length, and avoid sending the same documents again and again.
Cut SageMaker Training and Hosting Costs
AWS says Managed Spot Training in SageMaker can lower training costs by up to 90% compared with on-demand instances. Spot instances can be interrupted, so AWS recommends checkpointing unless the job finishes quickly.
A few more habits help:
- Shut down notebooks and endpoints you are not using.
- Pick the smallest instance type that meets your latency target.
- Delete old checkpoints, model artifacts, and experiment outputs you no longer need.
Cut GPU Costs on EC2
For self-managed training, EC2 Spot Instances offer savings of up to 90% compared with On-Demand prices. They suit fault tolerant training jobs with regular checkpoints. For short, planned runs that need guaranteed GPUs, Capacity Blocks reserve capacity for a set period.
For steady inference or training that runs all year, Savings Plans offer up to 72% savings in exchange for a one or three year commitment. But AWS says credits do not cover upfront Savings Plans fees, so compare the cash effect before you commit.
Consider AWS’s Own AI Chips
If your models run on supported frameworks, Trainium and Inferentia can lower training and inference costs, based on AWS’s “up to 50%” claims for specific cases. Pilot one workload first, measure real cost and quality, and then decide.
Watch Data and Storage Too
Apply lifecycle rules to datasets and logs. Keep training data and compute in the same Region and Availability Zone when you can. Remove old snapshots and unused volumes. These are small items, but they add up when your pipelines run all day.
AI Cost Levers at a Glance
| Lever | Where it applies | AWS stated savings | Trade-off |
| Bedrock batch inference | Asynchronous model calls | 50% lower price for select models | Results are not instant |
| Bedrock prompt caching | Repeated context across calls | Lower cost and latency | Works best with large, stable prompts |
| SageMaker Managed Spot Training | Model training | Up to 90% vs on-demand | Interruptions, so use checkpoints |
| EC2 Spot Instances | Fault tolerant GPU or CPU jobs | Up to 90% vs On-Demand | Capacity can be reclaimed |
| Trainium and Inferentia | Supported training and inference | Up to 50% in AWS’s examples | Needs the Neuron software and testing |
| Savings Plans | Steady compute usage | Up to 72% vs On-Demand | Commitment, and upfront fees are not covered by credits |
These are maximums that depend on the model, Region, and workload. Treat them as a guide to what to test first.
Plan Your First 90 Days With AI Credits
A short plan beats improvisation. Try this sequence.
- Record the facts. Write down each credit’s amount, expiration date, and eligible services from the Credits page.
- Set budgets. Create AWS Budgets alerts at 50%, 75%, and 90% of your credit balance, and turn on Cost Anomaly Detection.
- Tag everything. Tag resources by project, model, and environment so AWS Cost Explorer can show where money goes.
- Measure your baseline. Track cost per request, per training run, and per active user.
- Pick two levers. Choose the two optimizations with the biggest effect, such as batch inference and Spot training.
- Run a cost review monthly. Compare actual burn with the plan, and adjust.
- Model the full price month. Estimate your bill without credits so the first full invoice is not a surprise.
- Plan the next funding source. If your balance ends before your next milestone, apply for more credits early.
Common Mistakes AI Startups Make With AWS Credits
- Spending as if credits were cash. They expire, and they cover only eligible services.
- Ignoring the expiration date. AWS says it cannot extend it.
- Running idle GPUs. An endpoint or notebook left running quietly burns the balance.
- Using the biggest model for everything. Most tasks do not need it.
- Buying commitments with credits in mind. Upfront Savings Plans fees are not covered.
- Skipping tags and budgets. Without them, you notice the problem after the money is gone.
- Applying with inconsistent details. Mismatched emails or funding dates slow reviews.
- Forgetting the post credit budget. Your product has to work at full price.
One honest limit applies. Amounts, rules, and eligibility change, and AWS decides each application. Treat every number here as a starting point, not a promise.
Decision Guide: What Should You Do First?
- You have no investors and no revenue yet: apply for Activate Founders now, then keep building.
- An accelerator or VC backs you: ask for the Org ID and apply for Portfolio.
- You build or fine tune models: apply to the Generative AI Accelerator and talk to your account team.
- Your workload runs on Bedrock: test batch inference and prompt caching before you scale.
- You train on GPUs: try Spot with checkpointing, and price Capacity Blocks for short runs.
- Your credits end within three months: build the full price budget and cut idle spend now.
Questions to Ask Your AWS Account Team or Provider
A short conversation can clear up most uncertainty. Bring these questions:
- Which credit package level am I eligible for, and what is the exact amount?
- When do the credits expire, and which accounts receive them?
- Which models and services on Bedrock and SageMaker are eligible?
- Do credits apply to EC2 Capacity Blocks or other prepaid purchases?
- Are there additional programs for my stage, sector, or workload?
- If I use an AWS Organization, how will credit sharing work across accounts?
- What happens to my support plan when credits end?
Get answers in writing where you can, and note the date you asked. Programs change often.
What to Do When Your AWS Credits Are About to End
The last 60 days matter. AWS sends expiration emails at 60 and 30 days, but your own plan should start earlier.
- Build the full price forecast. Use Cost Explorer to see your spend as if credits did not exist.
- Apply for more if you qualify. AWS says you can reapply for a higher amount if you meet the requirements and are under the lifetime cap.
- Optimize first, commit second. Cut waste before you consider Savings Plans, since commitments last one to three years.
- Review architecture choices. Revisit model sizes, caching, batching, and instance types.
- Watch your support plan. If your credits covered support, choose your next plan before the date arrives.
- Tell your investors and team. A funding or runway conversation goes better with a clear cost picture.
AWS says it cannot extend credit expiration dates, so the best defense is a plan that makes the end date uneventful.
Practical Examples
These three scenarios are illustrative, not real customer cases.
The self-funded builder. Two founders building an AI writing tool qualify for Activate Founders. They use the credits for prototype calls to Bedrock models, tag each experiment, and set budget alerts so the small balance lasts through their first pilot.
The backed startup. A seed stage company in an accelerator receives its provider’s Org ID and applies for Portfolio. It moves nightly document processing to batch inference and runs training on Spot with checkpoints, so its balance covers a longer stretch of development.
The scaling lab. A team training its own models wins a place in a competitive program and talks to its account manager about larger awards. It pilots Trainium on one workload, compares cost and quality, and keeps GPUs for tasks that need them.
Conclusion
AWS credits for AI startups can fund months of experimentation, but only if you pick the right program, apply carefully, and spend them with a plan. Start with Activate, ask your investors for an Org ID, consider the Generative AI Accelerator if your work fits, and talk to your account team about larger awards when you are ready to scale.
Then make the credits last. Use Bedrock batch inference and caching, Spot for training, and budgets and tags for control. Open your Billing console today, record your expiration date, and start building your first 90 day plan.
