If your biggest line item is GPU hours, the programme you apply to first matters more than almost any other fundraising decision you'll make this quarter. Cash is fungible. Compute credits are not, they're tied to a specific cloud, a specific tier, and often a specific stage of company. Here's what twelve real programmes actually hand over, where the eligibility quietly cuts people off, and the order I'd apply in if I were burning GPU budget today.
The two that aren't really about cash
NVIDIA Inception gives free cloud credits and preferred pricing on GPUs and software. There is no cash grant, and there's no cohort deadline either, intake is rolling year round, so there's no reason to wait for a batch window that doesn't exist. Because it's a pricing and credit layer rather than a single cloud's allocation, it sits underneath whatever cloud you actually run on, which is why I'd apply here before anywhere else on this list.
HubSpot for Startups isn't compute at all, it's go to market discounting: up to 90% off year one, 50% off year two and 25% off year three for venture-backed startups, or 30% off for the other track. Worth having, irrelevant to your GPU bill. Mentioned here only so you don't confuse it with a compute programme when you're triaging where to spend application time.
The big cloud credit grants, and where the eligibility bites
Google for Startups Cloud Program is the largest pure cloud grant on this list: up to $350,000 in Cloud credits for Seed to Series A startups that are AI-first, and $200,000 for other early-stage companies. That split is the trap. If you're not classified AI-first, or you're past Series A, you drop straight to the $200,000 tier or out of the programme entirely, so get your positioning right before you apply, not after a rejection.
Microsoft for Startups Founders Hub offers up to $150,000 in Azure and Startup credits. It's smaller than the Google AI-first tier but it has no equivalent stage split described in the programme's own figures, which makes it the more predictable of the two if you want a number you can plan a budget around rather than a number you might qualify for.
AWS Activate splits even harder: up to $5,000 USD for self-funded founders, versus up to $200,000 USD for pre-Series B startups with institutional backing. If you're bootstrapped, don't expect the headline number, you're capped at $5,000 regardless of how much GPU you're burning. What AWS does well is speed, you hear back within 5 to 10 business days of applying, which beats every other programme here on turnaround.
Snowflake for Startups offers up to $1 million in investments, the largest single figure on this list, but it's structured as investment rather than a straight credit drawdown, so it fits a data-warehouse-heavy workload more than a raw training compute problem.
Narrower programmes worth a look depending on your stack
- Google for Startups Accelerator: AI First Latin America, equity-free, includes Google product credits and the option to apply for Cloud credits on top. Geography-gated, but if you qualify it's worth stacking against the main Cloud Program.
- Together AI Startup Accelerator, up to $15K on the Build tier, up to $30K on Scale, up to $50K on Growth. Tiered by stage, so know which tier you're applying into before you submit.
- Greylock Edge, $500K+ in credits, rolling applications through the form on their site, no cohort to wait for.
- Pinecone for Startups, amount not published by the organiser, but it includes free access to the Standard tier with Pro support, useful if vector search is a real cost centre for you.
- ElevenLabs Grants Program, more than $5,500+ in value, rolling with no fixed deadline, worth it only if voice is actually in your product.
- Google for Startups Accelerator, equity-free mentorship and technical support rather than a published credit figure. The programme site doesn't currently state application dates, so check status before you build a timeline around it.
Can you combine them?
Yes, but only in one direction that actually works. NVIDIA Inception isn't tied to a single cloud, it's a pricing layer plus credits that sit on top of whatever infrastructure you choose, so it stacks cleanly with a Google, Microsoft or AWS credit grant. What doesn't stack is two cloud-specific credit pools against one bill, AWS Activate credits only burn down your AWS invoice and Google's only burn down GCP, so running multi-cloud to use both means running two infrastructures, not one cheaper one. The practical move is one primary cloud credit grant, plus NVIDIA Inception underneath it, plus whichever narrow tool-specific programme (Pinecone, Together AI, ElevenLabs) matches a cost centre you actually have.
What happens when credits expire
None of the programmes in this catalogue publish an expiry date in the figures we checked, which is itself worth noting rather than guessing past. Treat every credit grant as a clock you can't see: draw down the largest, shortest-fuse pool first (your Google or AWS allocation) and keep NVIDIA Inception's preferred pricing as the steady-state discount you lean on once the one-time grants are spent, since it isn't a fixed pool that depletes the same way.
Apply in this order
Start with NVIDIA Inception because it's rolling, free, and compatible with every other programme on this list, there's no reason to delay it. Apply to Google for Startups Cloud Program next if you can honestly claim AI-first and Seed through Series A, the $350,000 ceiling is the single biggest cloud number you can realistically land here. If you're self-funded and need money fast rather than big, file AWS Activate for the 5 to 10 business day turnaround, just budget for $5,000, not $200,000. Everything else on this list is a layer you add once those three are in motion, not a substitute for them.