Credit Stacking: How Founders Legally Combine $200k or More
August 6, 2026 · 3 min read
The six-figure credit totals founders quote are almost never one programme. They are four or five, held at once, from companies that compete with each other and do not care that you also took the other one.
This is not a loophole. It is how the programmes are designed, and the only reason it feels like one is that nobody explains it.
Why stacking works
Credit programmes are customer acquisition. AWS gives you credits hoping you build on AWS and stay after they run out. Google does the same. Neither has any incentive to make you choose, and neither asks.
We have not found a major credit programme whose terms require exclusivity. What they do require is that you meet their own eligibility rules, individually.
A realistic stack
For a seed-stage AI startup with an accelerator affiliation, the published amounts alone reach:
| Programme | Published amount | Covers |
|---|---|---|
| Google for Startups Cloud, AI track | $350,000 | Training, inference, infrastructure |
| AWS Activate, partner tier | $200,000 | Production infrastructure |
| Microsoft for Startups Founders Hub | $150,000 | Azure, plus OpenAI access and tooling |
| NVIDIA Inception | $100,000 | GPU access and hardware discounts |
Add the tooling perks (analytics, error tracking, support, compliance) and a well-affiliated startup is running on a materially different cost base than one paying list price.
Two honest caveats. Those are the top tiers, which are referral-gated, and most startups will not clear all four at the maximum. And the majority of programmes in our catalog publish no amount at all, so any total assembled from published figures is a floor rather than a ceiling.
How to stack in the right order
Pick your primary cloud on architecture, not credits
Decide where you actually want to run, then take the others as bonus capacity. Migrating a production system to chase a larger credit almost never pays back: the credit expires and the migration does not.
Use secondary clouds for the workloads that suit them
The credit you are not running production on is ideal for batch jobs, model training, staging environments and experiments. This is where founders leave the most value unclaimed, because they treat the secondary credit as unusable.
Layer model credits on top
If you call hosted models rather than running your own, OpenAI and Anthropic credits cover a cost the cloud programmes do not. They are smaller and they stack cleanly.
Claim the tooling perks in one sitting
Individually small, collectively often most of a seed-stage tooling budget, and the applications are short by design.
The thing that actually costs founders money
Not stacking too little. Letting credits expire.
Every one of these is time limited, typically 12 to 24 months. A $200,000 credit you consume 30% of is worth $60,000. A $50,000 credit you fully use is worth more than that.
Track the expiry date the day you are approved. It is the single most valuable number in the whole stack, and it is the one nobody writes down.
Our credit wallet tracks expiry across every programme you hold, which is the boring feature that turns a stack into runway.
What not to do
- Do not misrepresent your company to different programmes. They occasionally share partner networks, and the inconsistency is what gets noticed rather than the stacking.
- Do not apply to every programme indiscriminately. A rejection for an obvious ineligibility wastes a reapplication window you may want later.
- Do not count unpublished amounts in your own planning. Most programmes do not state a figure. Assuming a number and building a budget on it is how founders end up surprised.
Work out what you can stack with our eligibility check, or see the runway maths on the credits calculator.