Why Exa Was Founded

Image 0: CEO Elias (right), and CTO Prithvi (left) at the infamous YC sign.

The demand for compute is on an exponential curve, and it is not going to slow down anytime soon. Increased computational power leads to better AI, which in turn leads to more advancements in science and technology at a rate that would have seemed impossible just a couple of years ago. It is amazing how fast AI is progressing, but it does come at a cost unless we do something about it.

Compute is generally very expensive when we look at the energy consumed. A large cluster of the current flagship GPUs is generally in the MEGAWATT range these days and is drastically inefficient. The energy usage is only going to increase, and it's going to follow that exponential curve, leading humanity into an energy crisis if we cannot supply these large clusters with enough energy, which in turn would impair the AI industry, leading to a stagnation in technological progress and scientific discovery.

The most optimal way for us to combat this is to:

• Create a more sustainable and efficient energy source (e.g., fusion).
• Create hardware capable of supporting AI with high levels of efficiency.

We are creating the latter. We believe that future AI can actually run on today's energy production levels, but not without the right hardware. This is why we founded Exa, and we are super excited to announce to the world what we are building.

We are making energy-efficient, reconfigurable chips for AI. Essentially, hardware that can reconfigure itself for each AI model, making it both faster and more energy-efficient than traditional GPUs. This area is called "polymorphic computing", and we believe it's the future of AI. The ideal hardware provides these three things:

• Versatility - different architectures are better at different things, and new ones are coming out every day. Besides, custom neural network design is entirely necessary for important real-world problems like the climate crisis or cancer (ref. AlphaFold or any SciML model).
• Interpretability - AI should be used to advance human knowledge, not replace it. It must remain a tool for human progress, not an enigmatic black box to throw at problems.
• Efficiency - AI should be able to run on today's power levels, at the edge, and decentralized.

Humanity will stagnate if it continues to be constrained by fixed computing architectures. We're breaking those chains. Our technology doesn't just optimize chips - it empowers people to experiment, discover, and innovate without the same hardware limitations, aiming for full reconfigurability. We're decentralizing AI, giving people the power to create whatever they want at sustainable costs. No more gatekeepers. No more resource monopolies. No more asking for permission.

Our vision goes beyond just creating efficient hardware. AI should be a tool for humanity to evolve, discover more, push the boundaries of our knowledge and understanding, and start solving the problems it was originally promised to address.

We believe open source is the key to creating safe and accessible AI for everyone. That's why we're committed to open-sourcing significant portions of our firmware, research, and software surrounding the hardware. We're convinced this approach will accelerate innovation in the AI industry. Let hackers hack. Let builders build. You don't spark true innovation by shoving proprietary licenses down people's throats.

We can't wait to reveal more about our hardware and see what people create with it in the future.

Stay tuned.

Per aspera ad astra.

// <3 Exa

Message signature

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA512

The demand for compute is on an exponential curve, and it is not going to slow down anytime soon. Increased computational power leads to better AI, which in turn leads to more advancements in science and technology at a rate that would have seemed impossible just a couple of years ago. It is amazing how fast AI is progressing, but it does come at a cost unless we do something about it.

Compute is generally very expensive when we look at the energy consumed. A large cluster of the current flagship GPUs is generally in the MEGAWATT range these days and is drastically inefficient. The energy usage is only going to increase, and it's going to follow that exponential curve, leading humanity into an energy crisis if we cannot supply these large clusters with enough energy, which in turn would impair the AI industry, leading to a stagnation in technological progress and scientific discovery.

The most optimal way for us to combat this is to:

- - Create a more sustainable and efficient energy source (e.g., fusion).
- - **Create hardware capable of supporting AI with high levels of efficiency.**

We are creating the latter. We believe that future AI can actually run on today's energy production levels, but not without the right hardware. This is why we founded Exa, and we are super excited to announce to the world what we are building.

**We are making energy-efficient, reconfigurable chips for AI. Essentially, hardware that can reconfigure itself for each AI model, making it both faster and more energy-efficient than traditional GPUs.** This area is called *"polymorphic computing"*, and we believe it's the future of AI. The ideal hardware provides these three things:

- - **Versatility** - different architectures are better at different things, and new ones are coming out every day. Besides, custom neural network design is entirely necessary for important real-world problems like the climate crisis or cancer (ref. [AlphaFold](https://github.com/google-deepmind/alphafold) or any *SciML model*).
- - **Interpretability** - AI should be used to advance human knowledge, not replace it. It must remain a tool for human progress, not an enigmatic black box to throw at problems.
- - **Efficiency** - AI should be able to run on today's power levels, at the edge, and decentralized.

**Humanity will stagnate if it continues to be constrained by fixed computing architectures**. We're breaking those chains. Our technology doesn't just optimize chips - it empowers people to experiment, discover, and innovate without the same hardware limitations, aiming for full reconfigurability. **We're decentralizing AI**, giving people the power to create whatever they want at sustainable costs. No more gatekeepers. No more resource monopolies. No more asking for permission.

Our vision goes beyond just creating efficient hardware. AI should be a tool for humanity to evolve, discover more, push the boundaries of our knowledge and understanding, and start solving the problems it was originally promised to address.

We believe open source is the key to creating safe and accessible AI for everyone. That's why we're committed to open-sourcing significant portions of our firmware, research, and software surrounding the hardware. We're convinced this approach will accelerate innovation in the AI industry. Let hackers hack. Let builders build. *You don't spark true innovation by shoving proprietary licenses down people's throats.*

We can't wait to reveal more about our hardware and see what people create with it in the future.

Stay tuned.

*Per aspera ad astra.*

// <3 Exa
-----BEGIN PGP SIGNATURE-----

iHUEARYKAB0WIQTqH4Le1ZB+MwLlm9jtpSZBtyZXuAUCZrznngAKCRDtpSZBtyZX
uI5ZAQDCnquwbcxt+3rcLA49lv0s+rF9FWCeKUAVOX5mSOYgVQD/bGLcVw++yBqA
SinhPAd1phSm0x6KvqNok5o60INS4Aw=
=wu0S
-----END PGP SIGNATURE-----
Copy