Resources // AI Infrastructure
Modal for Startups
Offer Up to ~$25k credits
Suits for Non-VC-backedVC-backed
Updated Jun 2026
The offer
- Thousands of free compute credits, up to ~$25,000 (Modal publishes tiers, not a fixed cap)
- Credits apply to actual GPU/CPU/storage usage (on-demand), not subscription fees
- Covers ML/AI workloads: training, inference, fine-tuning, batch, sandboxes
- One-time grant, valid ~12 months (you can't reapply once granted in a tier)
- Plus: direct access to Modal's engineering team, GTM promotion, founder community
Who qualifies
- New to Modal (no prior credits)
- Funding-gated tiers:
– Seed–Series A: raised from a VC in Modal’s partner network, or > $1M from any fund – Scaling (Series B+): raised > $30M or post-Series B, and a partner-network VC
- Company-domain email + a payment method required
- Partner network includes Y Combinator, a16z, Sequoia, Khosla, Neo, HF0, Pear, Lux
Community Insights
Modal is a developer-experience favorite for serverless GPU: scale-to-zero, fast cold starts, infra-as-code, pay only for what you use, minimal ops. Users report real savings (one cites ~3x fewer GPU-hours) and production use at companies like Suno, Substack and Quora. The trade-offs: a learning curve on Modal’s function/config model, and at sustained scale it runs pricier than raw-GPU shops like RunPod. Strongest fit is serving/inference with autoscaling; one-off fine-tuning runs are a weaker fit.
Best Practices (from community tips)
- Develop locally, offload the GPU-heavy parts to Modal. Keeps iteration fast and spend low.
- Lean into autoscale-to-zero for inference services; use it less for one-off fine-tuning runs (community says that’s not the sweet spot).
- Match the GPU to the job and use auto-stop. High-capacity instances are powerful but pricey if left idle.
- For sustained high-volume, compare raw-GPU pricing (RunPod) before burning credits; your funding stage and partner-network VC drive the credit tier.