Skip to content
Book a demo
An inference company

Building the next generation of
inference.

For the next generation of hardware. The same intelligence, on far less compute.

(01) Research

Technical notes, white papers and essays, written by the people building it. The math lives here.

All research
White paperWhat compute is for.Our thesis. Every use of compute is a choice, and how we see the next few years of AI playing out.Coming soon · Thesis
EssayInference is the market.Training happens once. Serving is paid on every request, forever, and agents multiply the requests.Coming soon
Technical note 01The memory layer, redesigned.How we rebuilt the memory layer of inference, and what it does to the cost of serving a model.Coming soon · White paper
EssayThe substrate is moving.Inference designed for the hardware that's powering tomorrow, not the hardware of the last decade.Coming soon
(02) Company

Lightwave is an inference company. A group of Princeton and MIT graduates rebuilding inference for the hardware of tomorrow.

The people building it
(03) Inference

A model is only as good as what it can remember.

Everything a model reads has to be held in memory, and on a GPU that memory runs out long before the math does. VAULT is our proprietary attention mechanism, built to change that. With it, the same model on the same hardware holds far longer context, picks a conversation back up where it left off instead of starting over, and serves more agents from every machine. Three results so far, and each one keeps scaling.

Read Technical Note 01
Agents[2×]per nodeTwice the agents on the hardware you have. Twice the efficiency for the same compute.
RecallComing soon
ContextNew heights of context capability.
(04) Book a demo

Experience Lightwave

Tell us about your company, what you serve today and why working together could be interesting. We will set up a demo around it.

Book a demoWe reply within 24 hours.