LOW LATENCY AI CHIPS FOR HEDGE FUNDS

Elevate your risk management with custom hardware that keeps your book long gamma with low jitter at sub-microsecond speeds.

Elevate your risk management with custom hardware that keeps your book long gamma with low jitter at sub-microsecond speeds.

Avronna designs custom intelligence hardware for hedge funds. Commodity chips take hundreds of nanoseconds to just round trip a single number on-die from the CPU to the GPU and back. FPGAs are fast enough for market making, but naively chasing speed typically creates a net short gamma position. Our CPU-TPU hybrid chips give your traders the flexibility to make complex calculations at timescales nobody else can touch.

Up to

7.4x faster

7.4x faster

inference & prediction than same package CPU+GPU (like AMD's MI300A APU).

It’s rigorous thinking, physics, and bold action. Just like what you demand from yourself.

It’s rigorous thinking, physics, and bold action. Just like what you demand from yourself.

The first business decision is whether, not how.

The first business decision is whether, not how.

Use our air-gapped simulator at your site to assess the optionality landscape: what sorts of chips would integrate into your environment?

Use our air-gapped simulator at your site to assess the optionality landscape: what sorts of chips would integrate into your environment?

We're offering a low-friction engineering experience (e.g. you still use JAX, GCC. LLVM etc.) so that we can isolate the main question: since swaps exchange timescales for when you have to be right, and committing to a policy is entering a payer swap whose tenor is your optimization timescale, how much alpha can be unlocked by small language models at the 20-100 microsecond timescale (for the full inference and round trip).


This isn't HFT, and this isn't Warren Buffet levels of breadth vs information tradeoffs. This is something new for us to build together.

We're offering a low-friction engineering experience (e.g. you still use JAX, GCC. LLVM etc.) so that we can isolate the main question: since swaps exchange timescales for when you have to be right, and committing to a policy is entering a payer swap whose tenor is your optimization timescale, how much alpha can be unlocked by small language models at the 20-100 microsecond timescale (for the full inference and round trip).


This isn't HFT, and this isn't Warren Buffet levels of breadth vs information tradeoffs. This is something new for us to build together.

Rumi Naik, CEO of Avronna

YOUR SPECIFICS ARE OUR OBSESSION

Email our CEO, Rumi.

You have to make a business decision, not a technical one, and we're here to help you make an efficient, high quality decision even if the right answer is "don't buy our chips."


There are risks, they can be managed.


Let's talk.

EMAIL RUMI

We have bandwidth for two more custom chip clients

SAY ALOHA TO ALPHA

NYC — SAN DIEGO — PITTSBURGH — BOSTON

NYC — SAN DIEGO — PITTSBURGH — BOSTON