GPU Credits for Startups
Which GPU clouds give AI startups free compute, how much, who qualifies and what each credit buys in H100 GPU-hours, read from each provider's own pages.
Credits and H100 prices read on each provider's own pages, all checked on or after . GPU programs and prices change often: confirm them on the provider's page before you plan a workload around a credit.
GPU credit programs for startups include Denvr AI Ascend ($1,000 to start; up to $500,000 in production credits), NetMind Elevate Program (up to $100,000 in inference credits; $100 at sign-up), Together AI Startup Accelerator (up to $50K in platform credits (tiers $15K/$30K/$50K)), Koyeb Startup Program (up to $30K in free compute credits), and Baseten AI Startup Program (up to $25K for inference or training + $2.5K model APIs). Each of the 17 programs below links to the provider's page that states its terms, with who qualifies, its own H100 price and what its largest credit buys in GPU-hours.
GPU credits are compute allowances that GPU clouds and AI platforms give startups to win their training and inference workloads. They range from a few dollars of free usage on signup to six-figure programs, and the largest usually go to startups with funding, an accelerator or a partner behind them.
A credit's headline is only half the comparison. The same NVIDIA H100 GPU-hour costs different amounts on different clouds, so each program below shows its provider's own on-demand H100 price, and we divide the largest credit by that price to show roughly how many GPU-hours it buys. The arithmetic is ours; the credits and prices are the providers'.
Which programs we picked, and their order, are our editorial choices; every term is quoted from the provider's own page with the date we read it. The general cloud programs at the end give larger credits that are not tied to GPUs.
GPU credits you can apply for
These programs take applications straight from startups. Each sets its own criteria, often a stage, a funding level or a workload.
- 1
Denvr AI Ascend
The largest GPU credit a startup can apply for directly
Denvr starts accepted teams with $1,000 of credits and says eligible teams can receive up to $500,000 in production credits, based on needs and potential, for its AI Compute and AI Inference services.
- What it pays for
- Denvr AI Compute and AI Inference Services (source: denvr.com)
- How long it lasts
- Varies by program fit; no fixed term (source: denvr.com)
- H100 on demand
- $2.45 per H100 GPU-hour (8x SXM node) (source: denvr.com)Largest credit buys about 204,081 H100 GPU-hours (our arithmetic: $500,000 at $2.45 an hour)
Terms read on denvr.com on Denvr AI Ascend: eligibility and how to apply - 2
NetMind Elevate Program
Inference-heavy products that need API credits
NetMind offers up to $100,000 in free inference credits through its Elevate Program, and signing up unlocks $100 of credit to trial the platform before you apply. Its pages state no eligibility rules, tiers or expiry.
- What it pays for
- Inference on NetMind's platform (source: netmind.ai)
Terms read on netmind.ai on NetMind Elevate Program: eligibility and how to apply - 3
Together AI Startup Accelerator
Open-model inference, fine-tuning and GPU clusters
Together AI tiers its free platform credits by funding raised: up to $15K below $5M, up to $30K from $5M to $10M, and up to $50K above $10M. The credits cover serverless inference, dedicated endpoints, fine-tuning and instant clusters, but not reserved GPU clusters.
- H100 on demand
- $3.99 per H100 GPU-hour (GPU Clusters, on demand) (source: together.ai)Largest credit buys about 12,531 H100 GPU-hours (our arithmetic: $50,000 at $3.99 an hour)
Terms read on together.ai on Together AI Startup Accelerator: eligibility and how to apply - 4
Koyeb Startup Program
Serverless GPUs for a seed-stage team
Koyeb gives seed-stage startups up to $30K of free compute credits for its Standard and GPU instances if they are in a partner accelerator or raised $1M to $5M from VCs, and reviews others case by case. Koyeb agreed in February 2026 to join Mistral AI, subject to closing conditions.
- What it pays for
- Standard CPU and GPU instances (source: koyeb.com)
- H100 on demand
- $2.50 per H100 GPU-hour (serverless GPU) (source: koyeb.com)Largest credit buys about 12,000 H100 GPU-hours (our arithmetic: $30,000 at $2.50 an hour)
Terms read on koyeb.com on Koyeb Startup Program: eligibility and how to apply - 5
Baseten AI Startup Program
Dedicated inference and training for a funded AI startup
Baseten gives AI-first startups at Seed to Series A, started less than five years ago, up to $25,000 for dedicated inference or training plus up to $2,500 for model APIs. The credits expire six months after they are granted, and only net-new customers qualify.
- How long it lasts
- Use within 6 months of the grant date (source: baseten.co)
- H100 on demand
- $6.50 per H100 GPU-hour (dedicated, billed per minute) (source: baseten.co)Largest credit buys about 3,846 H100 GPU-hours (our arithmetic: $25,000 at $6.50 an hour)
Terms read on baseten.co on Baseten AI Startup Program: eligibility and how to apply - 6
Anyscale AI Startup Program
Ray workloads on your own cloud
Anyscale offers up to $20K in credits to startups at various stages, weighing funding, technical maturity and current cloud setup. It runs on your own cloud, so the credits stack with the credits your cloud provider already gave you.
Terms read on anyscale.com on Anyscale AI Startup Program: eligibility and how to apply - 7
E2E Networks Startup Program
A free $5K to start, with discounts as you grow
E2E gives AI startups incorporated in any country $5K in free cloud credits in its entry tier, and usage-based discounts of up to $50K or $100K in tiers for startups that raised $2M or $5M. The credits last 12 months and cover GPU, CPU, storage and networking.
- H100 on demand
- $2.69 per H100 GPU-hour on demand (source: e2enetworks.com)Largest credit buys about 1,858 H100 GPU-hours (our arithmetic: $5,000 at $2.69 an hour)
Terms read on e2enetworks.com on E2E Networks Startup Program: eligibility and how to apply - 8
Lambda Research Grant
Research projects that need H100 time
Lambda offers qualifying researchers up to $5,000 in cloud credits to develop and showcase their work on its on-demand instances, by application.
- Who qualifies
- Qualifying researchers (source: lambda.ai)
- What it pays for
- Lambda's on-demand cloud instances (source: lambda.ai)
- H100 on demand
- $3.99 per H100 GPU-hour (8x SXM; 1x is $4.29) (source: lambda.ai)Largest credit buys about 1,253 H100 GPU-hours (our arithmetic: $5,000 at $3.99 an hour)
Terms read on lambda.ai on Lambda Research Grant: eligibility and how to apply - 9
Runpod Startup Program
A small free start, or a bonus on a prepaid commitment
Runpod's Starter tier gives $1,000 in credits, and its Growth tier adds $25K of bonus credits when a startup commits $50K upfront under a 12-month agreement. Runpod says venture backing is strongly preferred.
- How long it lasts
- Growth credits run under a 12-month agreement (source: runpod.io)
- H100 on demand
- $2.69 per H100 GPU-hour (SXM, Community Cloud) (source: runpod.io)Largest credit buys about 371 H100 GPU-hours (our arithmetic: $1,000 at $2.69 an hour)
Terms read on runpod.io on Runpod Startup Program: eligibility and how to apply - 10
FlexAI Startup Program
Teams with a production inference workload
FlexAI starts accepted startups with $1,000 of Token Factory credit, its serverless per-token tier, then takes 50% and 30% off for six months, with up to $20K of dedicated savings later. It asks for $1M+ in venture or accelerator backing, or a validated production workload.
- How long it lasts
- Discounts run 6 months; credit expiry not stated (source: flex.ai)
Terms read on flex.ai on FlexAI Startup Program: eligibility and how to apply - 11
Modal for Startups
Serverless GPUs for a VC-backed startup
Modal does not publish an amount: credits depend on traction, funding raised and whether a VC or accelerator in its network backs you. They cover all compute including GPUs, last one year, and are a one-time grant per tier.
- How long it lasts
- 1 year; cannot be reinstated (source: modal.com)
Terms read on modal.com on Modal for Startups: eligibility and how to apply - 12
Vast.ai Startup Program
Teams already spending on GPUs
Vast matches verified compute spend dollar for dollar in credits, up to an amount set when you are approved, for startups actively training, serving or scaling AI workloads. Its prices are marketplace rates that change hourly.
Terms read on vast.ai on Vast.ai Startup Program: eligibility and how to apply - 13
Gcore for Startups
Matching credits as your GPU spend grows
Gcore's startup program, open to startups up to Series A, gives grant credit back for every dollar spent and says the cashback-style grants apply to its AI and GPU infrastructure. Applications are reviewed quarterly.
- Who qualifies
- Startups up to Series A (source: gcore.com)
Terms read on gcore.com on Gcore for Startups: eligibility and how to apply
GPU credits that need a partner
These programs give credits only to startups referred by a partner the provider names: an investor, an accelerator or a membership program.
- 1
fal startup deals (a16z and Entrepreneur First)
Generative media teams in an a16z or EF portfolio
fal gives a16z portfolio companies one package, $50,000 in serverless credits or $20,000 for hosted model APIs, released in stages over one to two years; Entrepreneur First companies get $10,000. Serverless runs custom models on H100, H200, B200 and B300 GPUs.
- How long it lasts
- Released in stages over 1-2 years (source: fal.ai)
- H100 on demand
- $4.50 per H100 GPU-hour list (as low as $2.49) (source: fal.ai)Largest credit buys about 11,111 H100 GPU-hours (our arithmetic: $50,000 at $4.50 an hour)
Terms read on fal.ai on fal startup deals (a16z and Entrepreneur First): eligibility and how to apply - 2
Nebius offer for NVIDIA Inception members
Well-funded AI startups in NVIDIA Inception
Nebius gives NVIDIA Inception members $5,000 in credits for on-demand training or inference plus up to $150,000 in compute savings. Selection is temporarily limited to startups with $5M or more in venture backing, and every application joins a waitlist.
- What it pays for
- Nebius on-demand training or inference (source: nebius.com)
- H100 on demand
- $4.50 per H100 GPU-hour (from October 1, 2026) (source: docs.nebius.com)Largest credit buys about 1,111 H100 GPU-hours (our arithmetic: $5,000 at $4.50 an hour)
Terms read on nebius.com on Nebius offer for NVIDIA Inception members: eligibility and how to apply - 3
Lambda offer for NVIDIA Inception members
A first $1,000 of GPU time for Inception members
Lambda gives NVIDIA Inception members $1,000 in credits for its on-demand B200, H100 and A100 instances, one grant per company, valid for 12 months after they are applied. Lambda reviews each application.
- How long it lasts
- 12 months from when credits are applied (source: lambda.ai)
- H100 on demand
- $3.99 per H100 GPU-hour (8x SXM; 1x is $4.29) (source: lambda.ai)Largest credit buys about 250 H100 GPU-hours (our arithmetic: $1,000 at $3.99 an hour)
Terms read on lambda.ai on Lambda offer for NVIDIA Inception members: eligibility and how to apply
Small free credits to start with
Free usage on signup, enough to test a workload before you pay. After it, usage is billed at the provider's prices.
- 1
Crusoe Cloud introductory credits
Trying serverless inference at no cost
Crusoe gives new and existing users $5 in free credits for serverless inference and fine-tuning in its Intelligence Foundry. The credit is for those serverless services; Crusoe's GPU instances bill at its listed prices.
- H100 on demand
- $3.90 per H100 GPU-hour on demand (source: crusoe.ai)
Terms read on docs.crusoecloud.com on Crusoe Cloud introductory credits: eligibility and how to apply
Cloud startup programs with bigger credits
General cloud credits from the hyperscalers, larger than most GPU programs but not tied to GPUs. Check each program's terms for the services its credits cover. Largest tracked value first.
How this comparison is built
The list covers GPU clouds and compute platforms that give startups credits for GPU time, grouped by how a startup reaches the credit: by applying, only through a partner, or as a small free credit on signup. The order and the "best for" notes are our editorial reading of each provider's terms. Every credit, condition and H100 price comes from the provider's own program, pricing or documentation page on the date shown, never from a third-party list; follow each link for the full conditions. The GPU-hours figure is our arithmetic, the largest stated credit divided by the provider's own on-demand H100 price, and it is not a promise of what a startup will be granted. The general cloud programs at the end are tracked like every program in the catalog, with their own sources and check dates. No provider pays for a placement on this page.
Read the full methodologyCommon questions
How can a startup get free GPU credits?
Three ways: apply to a GPU cloud's startup program, get referred by a partner the provider works with (an investor, an accelerator or a membership program), or start with the small free credit many GPU clouds give on signup. Each program on this page states who qualifies, and the largest usually ask for funding or a partner.
Which GPU cloud gives startups the most credits?
The comparison lists each program's largest stated credit. A bigger credit on a more expensive cloud can buy fewer GPU-hours than a smaller one on a cheaper cloud, so each program also shows its provider's own H100 price and what its largest credit buys in H100 GPU-hours.
Can GPU credits pay for H100s?
Usually, where the provider rents them, but terms differ: some credits cover any GPU on the platform, others only certain products, such as serverless inference or a specific service. Each program on this page quotes what its credit pays for from the provider's own page.
Do GPU credits expire?
Almost always. Most startup credits must be used within a set period after they are granted, often between a few months and a year. Each program on this page states its expiry where the provider publishes one, so plan the workload before you apply.
Should I take the hyperscalers' credits instead?
They are larger but general: Google Cloud, AWS and Microsoft give startups cloud credits that cover many services, and their top tiers often need funding or an investor (AWS Activate Portfolio needs an Org ID from an Activate Provider, and Google's Scale tier is for VC-funded startups). Many AI startups use both, general credits for their cloud stack and a GPU program for training and inference, which is why the general programs are listed at the end.
More best-of guides
New startup perks, in your inbox
Get an email when a new credit, discount, or deal for founders is added. No spam, unsubscribe anytime.