Home › Guides › Compute & cloud credits › Guide › Article

Article · Compute & cloud credits · 7 min read

How to write a strong compute-credits application

Cloud and GPU credit programmes are not a lottery. Reviewers want to see a real organisation, a real workload, and an account that can actually receive the credit. The form looks short; most rejections come from a mismatch between who you are and what that programme funds.

How much free compute could you stack?Answer three questions in the credits calculator. Built from verified listings only.

Some links in this guide are affiliate links. If you sign up through them we may earn a small commission, at no extra cost to you. It never affects what we recommend.

Two different kinds of application

Startup cloud programmes and research-credit programmes look similar on a listing, but they ask for different evidence.

Do not send a startup pitch to a research-credit form, or a paper abstract to a startup programme. If you are comparing the large clouds first, use Google, AWS and Azure credits for AI startups compared. AWS Activate has its own walkthrough in AWS Activate credits: eligibility and how to apply. Our compute and cloud credits guide explains the types of programme.

What to have ready before you open the form

Most weak applications fail before anyone reads the project description, because the account or identity does not match the rules.

  • The right email. Google wants a business email that matches your public website domain, and the Google Cloud account and billing ID must be administered by a user on that same domain [source]. AWS Cloud Credit for Research requires an official institutional address (.edu, .org or equivalent) and will not accept a personal email [source]. Runpod's startup programme also asks you to use your company email [source].
  • The right cloud account, already created. Google asks for your 18-character billing account ID, because that is where approved credits are issued [source]. AWS Cloud Credit for Research requires an AWS account before you apply [source]. For AWS Activate, upgrade that account to a paid tier and link the account the credits should go to, as we set out in the Activate article.
  • A live website that describes the company or lab. Google screens for placeholder sites and for a high volume of AI-generated text, images or other content, and may reject those applications [source]. Runpod looks for a business email, a live website, LinkedIn and a social presence [source].
  • A workload you can describe in concrete terms. "We will train a model" is not enough. Say what you will run (training, fine-tuning, inference, a benchmark, a workshop), on what kind of hardware, and over what period.

Startup programmes: what they actually assess

On Google for Startups Cloud Program: AI track, Google says it assesses funding stage and type, business model, founder experience and product usage, plus company age and growth [source]. The Start tier ($2,000) is for startups founded in the past two years with a working MVP and plans to seek venture funding. Funded tiers (Scale and Scale AI) run from $200,000 to $350,000 and are aimed at startups with verifiable Pre-Seed to Series A funding. For AI startups the official breakdown is 100% of up to $250,000 in year one and 20% of up to an additional $100,000 in year two [source].

Two Google rules catch people out. First, non-venture funding — private equity, bank loans, government innovation programmes, grants, crowdfunding, or friends and family — does not qualify for the Scale tier, though the company may be considered for other tiers [source]. Second, credits apply to Google-owned, pay-as-you-go Cloud products; they do not cover third-party models in Model Garden or Marketplace, or subscription-billed first-party products [source]. In most cases you hear back in five to seven days; applications with missing information can take ten days or more [source].

Microsoft for Startups offers up to $150,000 in Azure credits for startups building software or AI. Read the official programme and FAQ for the current verification steps, and remember that a credit card is typically required for anything the credit does not cover once you move to pay-as-you-go. AWS Activate Credits is covered in detail in our Activate article: have the company profile, funding stage and the correct AWS account ready, and expect a week or two rather than an instant decision.

Research credits: write a project, not a wish

AWS Cloud Credit for Research is explicit about what it will and will not fund. Eligible applicants are full-time faculty, full-time research staff, or enrolled graduate, postgraduate or PhD students at an accredited research institution. The project must be one of three types: a proof of concept or benchmark for moving a workload or open dataset to the cloud; a cloud-hosted, publicly available science-as-a-service tool; or a workshop or tutorial that trains a wider community. It will not support ongoing or already established research, operational workloads, or general laboratory funding [source]. Student awards are capped at $5,000; faculty and staff awards are not capped. Reviews typically take 90 to 120 days, and the programme is not currently available for applications from Greater China [source].

Google TPU Research Cloud (TRC) gives accepted researchers access to Cloud TPUs at no charge, with TensorFlow, PyTorch, Julia and JAX supported. In return, participants are expected to share TRC-supported research through publications, open-source code, blog posts or similar, and to give Google feedback on the programme and the TPU platform [source]. Anthropic AI for Science program offers Up to $20,000 in API credits over 6 months, with an emphasis on biology and life sciences. OpenAI Researcher Access Program offers Up to $1,000 in API credits, valid 12 months. Lambda Research Grant offers Up to $5,000 in Lambda cloud credits, and selected work may be featured on Lambda's site.

In every case, size the request to the work. A vague "as much as you can give us" request is weaker than a short plan: the experiment, the hardware, a rough hour count, and what you will do if the credit runs out.

Specialist GPU clouds have their own startup programmes

If your workload is already on a specialist GPU cloud, or will be, apply to that provider's programme rather than only to the large clouds.

Runpod Startup Program is applied for with a short form; you do not need a Runpod account to apply [source]. Official tiers, checked on 18 November 2026, are: Starter, $1,000 in credits, for startups that have raised under $1 million; Venture, $5,000, for portfolio companies of a verified VC (a16z, Neo and South Park Commons are named) that have raised under $5 million; and Growth Track, for startups that have raised over $5 million, where you commit $25,000 or $50,000 and receive $10,000 or $25,000 in bonus credits [source]. Credits can be used across Pods, Serverless, Clusters and Storage, and expire one year after they are issued. There is one free offer per company. Existing Runpod customers are not eligible for the free Starter or Venture credits, but can apply for Growth Track [source].

Vast.ai Startup Program (GPU credit matching) is not a free allowance. Vast.ai matches verified compute spend dollar for dollar, up to an amount agreed after a review, and is aimed at companies already training, running inference or scaling a production workload — not at teams still deciding whether to use AI, or looking for free compute with no committed spend [source].

Both programmes are selective. If you are not accepted, you can still use the platforms on standard pricing: Runpod and Vast.ai are two of the specialist GPU clouds we compare in the cheapest GPU clouds for AI research and in on-demand vs reserved GPUs.

Common mistakes, and what to do after you are approved

  1. Applying with a personal Gmail or a university email to a startup programme that wants a company domain, or the other way around.
  2. Naming a billing account you do not control, or one that does not belong to the company on the form.
  3. Describing a project the programme has already said it will not fund: ongoing lab work on AWS research credits, or third-party Marketplace models on Google Cloud credits.
  4. Treating the headline amount as a guarantee. Tiers, caps and "up to" figures are the maximum, not an award you should budget as if it were already in the account.
  5. Letting jobs keep running when the credit is used up. Set a budget before the first large run; we cover this in How to spend Google, AWS and Azure credits without wasting them.

Open credit programmes we track are on our compute and cloud credits list.

Open now: compute & cloud credits closing soonest

See all 28 open compute & cloud credits →

More on compute & cloud credits

Frequently asked questions

Should I apply to several credit programmes at once?

Yes, if you meet the rules of each one. Startup and research programmes can complement each other, and specialist GPU clouds are a separate application. Just do not assume credits stack inside the same cloud account without reading that provider's rules on prior credit use. Google, for example, treats past or present Google Cloud credit usage as part of eligibility.

How long does a decision take?

It depends. Google says most startup-programme decisions come in five to seven days. AWS Activate is usually 5–10 business days, sometimes quoted as 7–10. AWS Cloud Credit for Research is much slower: typical review cycles are 90 to 120 days, and expedited reviews are not offered.

Can a stealth or unfunded startup apply?

Sometimes, but not to every tier. Google currently supports stealth founders through investor referrals, and unfunded stealth founders are asked to wait until they have a working website that describes the business, team and product. Runpod's Starter tier is aimed at companies that have raised under $1 million, with a real GPU workload and a live company presence.

Do I need to pay once the credit runs out?

Yes, if services keep running. Several programmes require a payment method for usage the credit does not cover. Set a spending limit before you start a large job, and see our article on spending credits wisely.

Where can I see every open compute-credit programme we track?

Our compute and cloud credits list, soonest deadline first. Each listing links to the organiser's page and shows when we last checked it.

Sources

Programme details are quoted from the organisers' pages on the dates shown. advance.to isn't affiliated with any organiser.

  1. Google Cloud: Google for Startups Cloud Program (checked 18 November 2026)
  2. AWS: Cloud Credit for Research (checked 18 November 2026)
  3. Google: TPU Research Cloud, about the programme (checked 18 November 2026)
  4. Runpod: Startup Program (checked 18 November 2026)
  5. Vast.ai: Startup Program (checked 18 November 2026)
  6. AWS Startups: everything you need to know about AWS Activate Credits
  7. Google Cloud: Google for Startups Cloud Program: AI track (checked 18 Nov 2026)
  8. Amazon Web Services: AWS Activate Credits (checked 18 Nov 2026)
  9. Microsoft: Microsoft for Startups (checked 18 Nov 2026)
  10. Amazon Web Services: AWS Cloud Credit for Research (checked 18 Nov 2026)
  11. Google Research: Google TPU Research Cloud (TRC) (checked 18 Nov 2026)
  12. Anthropic: Anthropic AI for Science program (checked 18 Nov 2026)
  13. OpenAI: OpenAI Researcher Access Program (checked 18 Nov 2026)
  14. Lambda: Lambda Research Grant (checked 18 Nov 2026)
  15. Runpod: Runpod Startup Program (checked 18 Nov 2026)
  16. Vast.ai: Vast.ai Startup Program (GPU credit matching) (checked 18 Nov 2026)

Share this page:

Free guide + the Monday email

Get the free guide. Never miss a deadline.

Your First AI Funding Application shows you which AI funding fits you and how to write an application that gets read. Then, every Monday, one short email with the deadlines closing in the next 14 days.