We're a small team trying to do something bold: create a new kind of AI chip where the model doesn't just run on hardware — it's actually built into it.

Instead of using GPUs to process model weights over and over, we’re exploring how to hardwire those weights directly into a chip. It’s a non–Von-Neumann architecture, meaning we’re eliminating the constant back-and-forth between memory and compute that bottlenecks traditional systems. That could make huge models way faster and dramatically more efficient — especially in edge devices where power and space are tight.

What We're Doing

With your support, we’re working toward three major goals:

1. Designing the core architecture: figuring out how to represent AI models physically, how data should flow, and how to make it all actually work.

2. Training large models: including a 400 billion parameter model, plus smaller ones we'll tune specifically for our custom chip design.

3. Prototyping and testing: using FPGAs, DACs, ADCs, and custom-built circuits to simulate and build early versions of our compute elements.

We’re also building the software tools to help convert and optimize models for this new kind of hardware.

Where the Money Goes

We're raising $950,000 to get through this phase of the project. That includes:

• Researcher and engineer salaries

• Workstations, simulation rigs, and prototyping tools

• Specialized electronics (like FPGAs)

• A few good chairs

We’re not spending anything on marketing or admin. Every dollar goes directly into the tech.

What If We Don’t Fully Succeed?

This is an ambitious and challenging project, and there’s a chance we may not reach all our goals. If that happens, the funds will still be put to good use by:

• Supporting ongoing research efforts by funding other researchers in related areas

• Developing valuable prototypes and hardware designs that can guide future innovation

• Building software tools and models that the wider community can benefit from

• Ensuring the team is fairly compensated for their time and expertise

Even if we don’t achieve everything planned, your support will advance knowledge and technology, and help fuel continued progress in AI research.

Why This Matters

Right now, large AI models are expensive to run and controlled by just a few big players. If we pull this off, it could lead to a future where even huge models can run efficiently, locally, and securely — right on your own hardware.

No datacenter. No subscription. Just powerful AI that you actually control.

AI compute today consumes enormous amounts of power, contributing significantly to environmental impact. Reducing this energy use is essential for creating a more sustainable future.

If that sounds like something worth building, we’d love to have your support.