about kernel grants
Tokens are the new kWh. Kernel Grants funds companies building tools to power the token factories of the future. If you are making it easier / faster / cheaper to extract tokens from the latent space, we want to help. We do not have an application deadline, nor a batch format. Any day is a great day to start building.
We invest $271,828 for 2.71828% alongside ~$1M of compute and cloud credits in 10 companies a year. We can follow on with up to $20M as you start hitting PMF. You also get access to the Kernel community space, a 24/7 coworking space in SF with everything you need to focus on building, as well as small events with the best people in AI for you to learn from.
ideas we're interested in
These are some of the problems we have been finding interesting just to give you a sense for the types of ideas we want to back.
Billions of dollars are spent on AI tokens, and most of it is thrown away at the end of a task. The spec is the new code in software. What is the equivalent for other domains? How do we turn these conversations into something more?
Every major language and dev tool has been written for humans. We need to rethink them to be ergonomic to the autonomous coding workflow rather than to a human-in-the-loop centric one.
The current behavior of starting new conversations and resetting context is not going to work in the mainstream. AI systems will need to modify their context while feeling seamless to the end user. This requires changes from the hardware level all the way to the frontend.
Not every token is worth the same. In the future, inference pricing will likely be tied to the economic impact of the tokens it generates; how would that pricing system be justified?
The new computer-based agents architecture needs a much more sophisticated system for agent-to-agent communication. Dispatching tasks to other agents, interacting with their custom systems, securely sharing data, etc are all very brittle at the moment.
The latent space is too vast for us to predict all good ideas; we're excited to hear about yours regardless.
learn from top AI builders
We host events throughout the year for our founder community. These are some of the speakers for our 2026 events.
Michele Catasta
President & Head of AI
Replit
Soumith Chintala
CTO
Thinking Machines
Jeremy Howard
Founder
Answer.ai & Fast.ai
Nader Khalil
Director of DevTech
NVIDIA
Logan Kilpatrick
MOTS
DeepMind
Chris Lattner
Founder
Modular
Dylan Patel
Founder
SemiAnalysis
swyx
Editor
Latent Space
And more...
Get in the arena
Apply
Any day is a great day to get started. Tell us what you want to build and why.
FAQ
Who should apply?
We are looking for founders with a technical insight they want to explore and the ability to turn that insight into something real. Ideally, you have not raised institutional capital before.
How long is the program?
Since we don't have batches, we just run events through the year. We don't do one large demo day, but rather launch companies as they are ready. We think the current batch system is fundamentally incompatible with how AI companies are being built.
How much do you invest?
We invest $271,828 for 2.71828% equity on a SAFE. We have a $315M fund from which we can invest up to an additional $20M as you start scaling.
Do you offer any credits for my company?
Yes. We work with a set of infrastructure and platform partners that can help cover early cloud, model, payments, and developer tooling costs.
OpenAI
Up to $10,000: OpenAI API credits for experimentation and production usage.
Stripe
$5,000: Stripe Startups access, $5k in credits, and waived Stripe fees for the first year.
AWS
Up to $100,000: AWS credits to cover early infrastructure and compute spend.
Microsoft Azure
Up to $250,000: Azure credits including Azure AI Studio access.
Google Cloud
Up to $350,000: Google Cloud credits for infrastructure and AI workloads.
Cloudflare
Up to $250,000: Cloudflare credits for deployment, networking, storage, and security.
Do I need to be in San Francisco?
Yes. You do need to be in San Francisco, but you do not need to come into Kernel every day to be part of the program.
Why is the amount such a weird number?
Softmax is one of the most important kernels, and e is at the core of its formula. It's also a natural exponential function, which every founder aspires to grow into.