Editorial overview
Why it matters
What they do
Purpose-built, high-memory hardware for transformer inference.
Why it could matter
An inference architecture optimised around memory capacity and energy use could loosen the industry's dependence on general-purpose GPU clusters.
Why now
Serving trained models is becoming a larger infrastructure problem than training them, creating room for specialised inference systems.
Recent update
Positron raised a $230 million Series B after shipping Atlas and is developing its next-generation Asimov silicon and Titan systems.
Leadership
Meet the founders
T
Thomas Sohmers
Founder & CTO
Applied mathematician and hardware engineer who founded Positron to build memory-first systems for AI inference.
Keep exploring