Short answer: the programs that give a startup free GPU time are NVIDIA Inception (free to join, no published dollar figure), Runpod ($1,000 Starter credits, or a $25,000 bonus on a $50,000 commitment), Modal ($30 a month on the free plan, thousands in startup credits for venture-backed teams), Together AI (up to $15K, $30K or $50K by funding tier), Vast.ai (dollar-for-dollar credit match on your spend) and Nebius (credits through VC partners for companies that raised $5M+). For a bootstrapped team with no investor, the realistic options are Modal's free plan, Hugging Face community GPU grants, and Lambda or Nebius research grants if you do research.
Most "free GPU credits" posts blur two different things: programs that hand you credits, and programs that tell you "contact us". This one separates them, because the figure matters when you are deciding whether to apply or just pay. Everything below was read on the provider's own page on 17 October 2026 unless noted.
The comparison table
| Program | What it gives | Who qualifies | Published number? |
|---|---|---|---|
| NVIDIA Inception | Free cloud credits from NVIDIA and partners, partner offers, hardware pricing | Incorporated, under 10 years old, a working website, a clear NVIDIA use case, run by at least one founder. Any funding stage | No dollar figure |
| Runpod Starter | $1,000 in free credits | Pre-Series A teams with working products and near-term GPU needs. Venture-backed preferred | Yes, $1,000 |
| Runpod Growth | $25,000 bonus on a $50,000 upfront commit, $75,000 total, 12-month agreement | Seed to Series B+ spending $5K to $25K+ a month | Yes |
| Modal free plan | $30 a month in compute credits | Anyone, no card | Yes, $30/month |
| Modal for Startups | "Thousands" of GPU credits, engineering support | New to Modal. Seed to Series A with partner-network VC funding or $1M+ raised | No exact amount |
| Together AI accelerator | Up to $15K / $30K / $50K plus engineering hours | Tiers by funding raised | Yes |
| Vast.ai | Credit match, dollar for dollar, up to a set amount | Teams actively training or running production workloads | No cap published |
| Nebius | Introductory credits and compute savings | $5M+ raised from an approved VC partner, incorporated under 5 years | No amounts |
| Lambda research grant | Up to $5,000 in cloud credits | Qualifying researchers | Yes |
| Hugging Face GPU grants | Free GPU upgrade for a Space | A published Space that needs a GPU | No |
| Thunder Compute | $20 credit | Students at a US university | Yes |
NVIDIA Inception: the one with no number
Inception is the one everyone asks about, and the honest summary is that it is a gateway, not a grant. NVIDIA's page says Inception is free, with no application fees, membership fees or equity requirements. Benefits include free cloud credits from NVIDIA and partners, partner offers on tools and services, and preferred pricing on a subset of enterprise GPUs. That pricing works as a rebate off the supplier price, and NVIDIA does not sell hardware directly to members.
The page gives no dollar amounts. Where credits show up, they tend to come through partners that Inception connects you to. NVIDIA directs questions on amounts to inceptionprogram@nvidia.com.
Eligibility is looser than most: you must be incorporated, under 10 years old, with a working website, a clear use case for NVIDIA technology, and actively operated by at least one founder. Consulting firms, crypto-related companies, cloud providers, resellers and public companies do not qualify. Funding stage does not matter, which makes this the best first application for a bootstrapped AI company. There are no deadlines or cohorts, so you can apply today.
Runpod: two clear tiers
Runpod is the most transparent of the paid-GPU-cloud programs. The Starter Tier is $1,000 in free credits with self-serve onboarding and direct support, aimed at pre-Series A startups with working products. The credits can be used on Pods, Serverless, Clusters, Public Endpoints or Storage.
The Growth Tier is a match, not a gift: commit $50,000 upfront and receive $25,000 in bonus credits, $75,000 in total, as a one-time bonus under a 12-month agreement with no credit reloads. That only makes sense if you already plan to spend that much. The page says the program is selective, venture-backed startups are preferred, and it is for new customers. Pre-funding or bootstrapped teams are pointed to self-serve pay-as-you-go instead.
Modal: free every month, bigger for funded teams
Modal's pricing page lists $30 a month in free credits on the Starter plan and $100 a month on Team. At Modal's listed rates that $30 buys real GPU time:
| GPU | Price per hour (our math from the per-second rate) | Hours in $30 |
|---|---|---|
| T4 | about $0.59 | about 50 |
| L4 | about $0.80 | about 37 |
| A100 80 GB | about $2.50 | about 12 |
| H100 | about $3.95 | about 7.5 |
Hourly figures are the per-second rates multiplied by 3,600, so treat them as approximate. For a prototype or a weekend fine-tune, that is enough to test an idea without a card.
Modal for Startups is the larger program: "thousands" of GPU credits, but no exact amount per tier on the page. It is for companies new to Modal. Seed to Series A teams qualify with any amount of funding from Modal's partner network or more than $1M raised from any fund. The credits are a one-time grant, and you cannot reapply once you have received credits in a tier. Modal's pricing page also lists up to $10k in free compute for graduate students, labs and researchers.
Together AI, Vast.ai and Nebius
- Together AI publishes tiers: up to $15K for companies that have raised up to $5M (3 hours of forward-deployed engineering), up to $30K for $5M to $10M (6 hours), up to $50K above $10M (10 hours). Credits cover serverless inference, dedicated endpoints, fine-tuning and instant clusters, but not Reserved GPU Clusters. More in Together AI and Fireworks startup programs.
- Vast.ai is explicitly not free: its page says it matches your verified compute spend in credits, dollar for dollar, up to an amount set during review, and that it is not a fit for "free compute with no committed spend". Think of it as a 2-for-1 on money you were going to spend anyway.
- Nebius gates its startup credits behind an approved VC partner. You need to have raised at least $5M from one, be incorporated under five years ago, and apply from a company email. The page gives no amounts and says applicants may be waitlisted. Nebius also runs research grants for postgraduate students, PhD students, postdocs and faculty, covering GPU cloud credits, Token Factory tokens and Tavily credits, with the amount set per proposal.
DigitalOcean: read the exclusion
One thing the "up to $100,000" headlines hide. DigitalOcean's startup page states that its core credits do not cover GPU Droplets, NVIDIA H100 GPU Kubernetes, Bare Metal GPUs or Dedicated Inference. GPU credit packages are separate, offered to selected startups who are invited to apply, and acceptance does not guarantee them. Some third-party posts describe a Hatch H100 offer with a December deadline. We did not find that on DigitalOcean's page today, so do not plan around it until you see it in writing.
No funding? What is actually free
- Modal's free plan: $30 a month, no card.
- Hugging Face community GPU grants: you need a published Space that needs a GPU, and the application sits in the Space's hardware settings. Hugging Face does not state an award size or review time. For reference, its paid tiers run from $0.40 per hour for a small T4 to $23.50 per hour for 8x L40S.
- Lambda's research grants: up to $5,000 in cloud credits for qualifying researchers.
- Thunder Compute: $20 for students at a US university, usable on A6000 and L40 GPUs.
For the broader picture beyond startup programs, where to get free GPU compute for AI and ML covers notebooks and free tiers, and cloud GPU pricing shows what H100 and B200 time costs once credits run out.
How to apply without wasting a month
- Apply to Inception first. Free, any stage, no deadline.
- Claim the self-serve credits (Modal, Thunder Compute) while you wait.
- Apply to Runpod Starter and Together if you are funded. Both publish their numbers, so you can judge the value.
- Do not stack on size alone. Some programs exclude existing customers, so claim the one you most want first.
- Ask about expiry. Fireworks credits expire after one year, and other programs set their own limits.
Credits are valued at list price, which is not what you would pay on a cheaper host. Compare the dollar value against a specialist provider before you chase a hyperscaler credit.
The full verified list is in the Perkstack catalog, with the date we last read each page. See also the $500K startup credits checklist, or create an account for each program's application steps.
rest of this guide
The rest of this guide picks up at "Runpod: two clear tiers".
- Runpod: two clear tiers
- Modal: free every month, bigger for funded teams
- Together AI, Vast.ai and Nebius
Free account. No card. 350+ verified perks.
Not ready for an account? Find out what you can claim first.
Already have an account? Log in