What the AWS Startup Program Gives You
The AWS startup program awards up to $300,000 in AWS credits, applied against your ordinary bill across compute, storage, databases, networking and managed AI services.
The important word is "up to". $300,000 is a ceiling reached through specific routes, not a number every applicant receives. Most teams land somewhere well below it, and the size of the award tracks how you come in rather than how good your product is.
Credits are not cash and not a discount. They sit as a balance that your monthly invoice draws down until it is empty, at which point you pay list price for whatever you built in the meantime. That single mechanic drives almost everything worth knowing about the program.
Eligibility depends on stage, funding and whether you arrive through a partner, accelerator or investor. Those specifics change often and are listed per program at AI Perks, which tracks $7.7M in credits across 194 companies.
What Cloud Infrastructure Credits Are Actually For
Cloud infrastructure is the layer every other part of your product sits on, so credits here offset a bill you cannot avoid rather than one you chose to take on.
For an AI startup in 2026, AWS usually carries four things at once:
Compute. Containers, serverless functions and virtual machines running your API, your workers and your background jobs.
State. Managed Postgres, object storage, caches and queues. This is the part that is genuinely hard to move later.
Networking. Load balancers, DNS, CDN and the bandwidth between all of it.
Model access. Amazon Bedrock resells hosted foundation models inside AWS, including Anthropic's Claude. Credits can therefore absorb inference spend if you route through Bedrock rather than calling a model vendor directly.
That last point is the one founders underuse. Cloud credits and model credits normally cover separate bills, but AWS is one of the few places where a cloud grant can also absorb inference spend.
How an AWS Bill Behaves at Scale
AWS costs do not scale with users. They scale with what you leave running, how much data you keep, and how much data you move.
The bill is not one meter. It is dozens of small ones, each with a different growth curve:
| Cost driver | How it grows | Why it surprises teams |
|---|---|---|
| Compute (on demand) | With provisioned capacity, not traffic | An idle instance costs the same as a busy one |
| Object storage | Monotonically, almost never down | Nobody deletes anything, so it only ratchets up |
| Managed databases | Step function on instance size | A single resize can double the line item |
| Data transfer out | With usage, priced separately | Egress is invisible until it is the second biggest line |
| Managed service premiums | With adoption | You pay for operations you no longer perform |
| Inference and GPU capacity | With request volume | The one cost that actually tracks product usage |
Only the last row behaves the way founders instinctively expect. Everything above it grows from decisions made months earlier, which is why cloud bills tend to jump in steps rather than curves.
The practical consequence: a credit grant covers precisely the period when your architecture is least efficient. That is what makes it valuable, and also what makes it so easy to waste. A grant spent on an inefficient architecture buys far fewer months than the same grant spent on a lean one.
What AWS Credits Stack With
Cloud credits cover where your code runs. They do not cover model APIs you call from outside AWS, and that is the gap most teams forget to fund.
If your product calls a model vendor's API directly, that invoice arrives separately and AWS credits do nothing for it. The strongest position is holding credits at both layers.
| Program | Commonly published ceiling | Layer it covers |
|---|---|---|
| AWS startup program | Up to $300,000 | Cloud infrastructure |
| Google Cloud for Startups | Up to $350,000 (AI track) | Cloud infrastructure |
| Microsoft for Startups Founders Hub | Up to $150,000 Azure | Cloud infrastructure |
| Model vendor programs | Varies widely by vendor | Inference and API calls |
Ceilings shown are the widely published maximums, not typical awards, and hyperscaler terms are revised frequently. Treat them as the shape of the opportunity and verify current numbers at AI Perks before planning around any of them.
Two caveats worth knowing. Hyperscaler programs are generally not stackable with each other on the same account, so this is a choice between providers rather than a pile. And some routes into a program exclude others, meaning the order in which you apply can quietly cost you the larger tier.
What Founders Get Wrong About Cloud Credits
The expensive mistake is treating a credit grant as free runway instead of a clock that has already started.
Every cloud credit balance is time boxed. Claiming one before you have a workload to spend it on burns the most valuable months against an empty bill.
The recurring errors:
Activating too early. The window opens the day you accept, not the day you start scaling.
Reading the ceiling as the award. Planning a budget around the maximum figure when your route qualifies for a fraction of it.
Building on services you cannot afford at list price. Credits make expensive managed services feel free, and the architecture you pick under those conditions is the one you inherit when the balance hits zero.
Ignoring egress. If you ever want to leave, moving your accumulated data out is a real line item, and it grows every month you stay.
Leaving non production environments running. Staging, dev and abandoned experiments consume credits at exactly the same rate as production.
Applying once. Multiple routes into the same program exist at different tiers, and which one you use determines the size of the award. That mapping is what AI Perks maintains.
The healthiest way to hold credits is to run as if you were paying. Then the grant becomes extra months of runway rather than a spending habit you cannot sustain.
How to Work Out What You Qualify For
Eligibility turns on stage, funding raised and how you enter the program, and the thresholds are revised often enough that last year's write up is usually wrong.
Step 1: Start at getaiperks.com and filter to Cloud Infrastructure. That category covers AWS alongside every competing hyperscaler and compute program.
Step 2: Map your route before applying. The entry path is the single biggest determinant of award size, and it is easier to change before you apply than after.
Step 3: Decide your provider deliberately. Because the large cloud programs do not combine, picking one is an architecture decision with a six figure price tag attached.
Step 4: Sequence the start date. Claim the balance when there is a workload ready to consume it, not when you first qualify.
Step 5: Re-check quarterly. Infrastructure program terms move more than any other category on the site.
Frequently Asked Questions
How much is the AWS startup program worth?
Up to $300,000 in AWS credits, drawn against your normal bill across compute, storage, databases, networking and managed AI services. That figure is a ceiling reached through specific entry routes rather than a standard award, and most teams receive considerably less. Current tiers are tracked at getaiperks.com.
Do AWS startup credits expire?
Yes. Every cloud credit balance is time boxed, and the window starts when you accept rather than when you begin spending. The exact length varies by tier and changes periodically, which is why claiming credits before you have a real workload wastes the most valuable part of the grant.
What can you actually spend AWS credits on?
Credits apply across the AWS catalogue: compute, object storage, managed databases, networking and Amazon Bedrock model inference. Some categories are typically excluded, including third party marketplace purchases and certain support plans. Exclusions vary by tier, so check the current terms at getaiperks.com.
Do AWS credits cover OpenAI or Anthropic API bills?
No. Calling a model vendor's API directly produces a separate invoice that cloud credits do not touch. Claude accessed through Amazon Bedrock is different, because that usage lands on your AWS invoice and draws on the credit balance. Teams calling vendors directly need model credits too.
Can you get AWS and Google Cloud credits at the same time?
Generally not in a way that helps. The large hyperscaler programs are not designed to stack, and you would be splitting one workload across two providers to use both. Choosing a provider is usually the better decision. Compare the current ceilings at getaiperks.com.
Is it worth applying before you have a product?
Applying is worth it, claiming usually is not yet. Understanding which tier you qualify for costs you nothing, but activating a time boxed balance against zero usage burns months you will want later. Line the grant up so it starts when your workload does.
Build on the cloud. Let someone else pay for the first $300,000 of it.