AI Grant Accelerator

We update this page when results are published, and when the next round opens.

AI Grant invests US$250,000 in seed-stage AI startups on a no-cap, no-discount MFN SAFE, and stacks on roughly US$640,000 in cloud and tool credits. It is run by Hersh Desai and Lenny Bogdonoff, and funded with $10M from Nat Friedman and Daniel Gross.

Applications for Batch 4 are closed and no Batch 5 date has been announced. This page is here for the record — the terms, the bar and the ~92 companies they have backed — so you know what to prepare if a batch reopens.

They say plainly: "we only invest in Delaware corporations."

That is a condition of taking the money, not of applying. Apply with whatever you have — an Indian Pvt Ltd, an LLP, or no company at all. If you are accepted, you then need a Delaware C-corp, and they point out you can create one from almost anywhere in the world in a few days using Stripe Atlas or similar.

So an existing Indian company is no barrier to applying, but it does mean a flip or a fresh US incorporation before the money lands, with the tax and FEMA consequences that carries. Worth talking to a CA once you are in the process — not before you apply.

Otherwise it is wide open: "Can anyone apply? Yes — no credentials required." Single founders can apply. Founders who have already raised can apply. They will fund multiple companies doing the same thing, on the view that the AI product market is big enough for both Uber and Lyft.

Anything that uses AI models in a useful or engaging way, at pre-seed or seed. In their words they want "technical & pragmatic founders who want to build great products" — people who get a thrill from making something others love to use, and who understand that a new product is "1% idea and 99% iteration".

Demos matter. On selection they say they want smart, energetic people with actionable, clearly useful ideas, and that demos are a big plus — "not just because they're fun to look at, but because they hint at the type of person that cares about the end-user experience." If you apply with one thing, apply with a working demo.

The credit stack, in their own list: PostHog $55k, Replicate $50k, Anthropic $30k, Modal $25k, OpenAI $15k, Vercel $12k, ElevenLabs $11k, Vapi $10k, Hugging Face $10k, AssemblyAI $10k, Perplexity $10k, Fireworks AI $10k, Together AI $7.5k, Zilliz $5k, Stripe $2.5k, Clerk $2.5k, Pilot $2.5k, Linear $1k, Render $1k, plus 10 free seats each from Weights & Biases and Julius AI, 25% off Cohere models and 20% off Deel fees.

Advisors and speakers have included David Holz (Midjourney), Andrej Karpathy (OpenAI founding member), Patrick Collison (Stripe), Tobi Lütke (Shopify), Alexandr Wang (Scale), Noam Shazeer (Character), Mike Krieger (Instagram), Guillermo Rauch (Vercel), Dylan Field (Figma) and Aravind Srinivas (Perplexity).

Bigger cheques exist separately. Nat and Daniel invest on their own account, generally leading rounds between $1M and $100M, and companies taking those also get access to the Andromeda Cluster.

No — it is an investment, and they address the name head-on. AI Grant started in 2017 handing out genuine research grants, awarding 36 individuals who went on to found Cohere, Helia and Cresta. They relaunched it around products rather than papers, on the view that AI research is now abundant while product innovation is just beginning. The name stuck; the instrument is a SAFE.

Around 92 companies across four batches, including Perplexity, Cursor, Replicate, Pika, Granola, Browserbase, MatX, Lindy and Chroma.

Batch 4 — Goodfire (interpretability tools), Silurian (weather forecasting foundation models), Granola (AI notepad for meetings), Pulse AI (document extraction), InspectMind AI (construction plan checking), Hamming (testing for voice agents), Agency (B2B customer success agents), Artificial Analysis (independent model benchmarks), Pursuit (public sector sales), Graylark (geospatial intelligence), Haize Labs (model hazing), Sync (lipsync), Rowan (molecular simulation), Vapi (voice agent API), Juicebox (recruiting), Supertrace AI (network engineers), Inception Labs (diffusion LLMs).

Batch 3 — Antimetal, Matic Robots, Omnivore, Momentic, Reworkd AI, Spark AI, K-Scale Labs, Matrices, Browserbase, Didero, Consensus, Method Security, Yuma AI, Abel, Virtual Staging AI, Symphonic Labs, DayDream Dental, Creatify AI, Womp Labs.

Batch 2 — Pika, ggml, MatX, Lindy, Freed, Mathpix, Abstract AI, Julius, Hume AI, Cofactory, Induced, Zeta Labs, Common Sense Machines, Guru AI, Curio, Echo Labs, Reality Defender, Andiron AI, Lightpage, Portola, Espresso AI, Jenni, AutogenAI, Listen Labs, Tutor Intelligence, RunPod, Akool, Coframe, OpusClip, Kino AI.

Batch 1 — Perplexity, Cursor, Replicate, Animato, Lexica.art, Minion.ai, Recraft, Flair, ValueBase, WOMBO, Chroma, Poly Corp, Sieve, Sameday, Play.ht, Ghostwrite, BuildShip, Birch, Vizcom, Circle Labs, Samaya AI, Secret Weapons, Pixelcut, AniML, Dust, Forefront.

The spread is worth reading as a signal: developer tools and infrastructure dominate, with consumer AI, vertical SaaS and a handful of robotics and hardware bets alongside.

Watch aigrant.com. There is no mailing list on the page and no announced Batch 5 date. Questions go to support@aigrant.org. Note the separate open-source grant programme at aigrant.org, which is a different thing from this accelerator.