Unconventional Grant: Final Call for Proposals

Tomorrow, May 15th, is the final day to submit pre-proposals for the Unconventional Grant.

Over the past several weeks, we’ve received a wave of submissions that reinforce a core belief behind this program: there is no shortage of ambitious thinking in AI, but there are far too few venues willing to fund it early.

This fund exists to change that.

Why we created the Unconventional Grant

Modern AI systems are extraordinarily capable, but they are also extraordinarily inefficient.

The human brain operates at ~20 W. In contrast, state-of-the-art AI systems require orders of magnitude more energy to train and deploy. This gap is not just an engineering problem. It reflects deeper assumptions about the abstractions we use to build intelligent systems.

Today’s AI stack is built on digital computation, dense linear algebra, and memory hierarchies that were not designed for intelligence. As models scale, data movement increasingly dominates system cost. As architectures grow more complex, efficiency gains become incremental.

We believe meaningful progress will require rethinking the substrate itself.

The Unconventional Grant was created to support research that challenges these assumptions, particularly work that is too early, too cross-disciplinary, or too unconventional for traditional funding pathways.

Our goals are to invest in foundational research and the teams behind it, connect like-minded researchers, create a pipeline of new ideas, and help develop the next generation of experts in this field.

What we’re seeing in the proposals

The breadth of ideas we’ve seen so far has been encouraging.

Across circuits, systems, and theory, many proposals are converging on a shared intuition: that efficiency gains will not come from scaling existing approaches alone, but from fundamentally different ways of representing and computing.

A few themes we’ve observed:

• Computation as dynamics
• In-memory and in-physics compute
• Architectures that minimize data movement
• New abstractions beyond linear algebra

These are hard problems to solve. Not all of these ideas will work. Many will fail.

That is precisely the point.

What we’re looking for today

If you are considering submitting, this is the moment.

We are not looking for incremental improvements to existing architectures. We are not looking for well-trodden extensions of current scaling laws.

We are looking for ideas that:

• challenge core assumptions in how AI systems are built
• introduce new computational primitives or substrates
• connect theory, hardware, and learning in non-obvious ways
• can be explored through rigorous modeling, simulation, or early prototyping within a one-year timeframe

The strongest proposals we’ve seen are the ones that make a clear, technically grounded case for why a different approach might unlock a step-function improvement.

Final reminder

Pre-proposals are due tomorrow, May 15, 2026.

If you have an idea that does not fit cleanly into existing funding categories, that may be a signal worth paying attention to.

The pre-proposal process is intentionally lightweight and takes approximately 10 minutes to complete.

Submit your application here.