How NP Co. Is Using AI Transformers to Simulate Physics 1,000x Faster

Transformer Architecture Applied to Physics

The company is building pre-trained foundational models using a transformer architecture — the same underlying approach that powers large language models — but applied to physical simulation rather than text . The models are trained on industrial physics data and designed to replace legacy simulation software across aerospace, defence, energy, electronics, data centres, and automotive industries .

A traditional physics simulator must run computationally expensive calculations from scratch for every design iteration. NP Co.’s approach pre-trains a model once on the relevant physics, then delivers results at inference time when an engineer wants to test a new design . There is no need to restart the full simulation chain for each change.

Demonstrated Speed Advantage

The performance difference is stark. Legacy simulation tools typically require days to weeks per design evaluation. NP Co.’s pre-trained models deliver results in seconds . The startup has demonstrated 1,000× speedups on industrial benchmarks — including those run by Safran, the aerospace engine manufacturer — and claims a path to 50,000× acceleration on full-assembly problems .

This speed jump changes what is possible in the design loop. Where an engineering team might previously have tested a handful of configurations, they can now explore thousands of design variants in the time it once took to run a single simulation .

What the €6 Million Will Fund

The pre-seed capital will primarily support expansion of the research team and continued development of the foundational models . Longer-term, NP Co. aims to build automated design tools and real-time operational simulators for industrial infrastructure . The goal is not simply to accelerate existing workflows, but to unlock entirely new ways of designing complex physical systems.

Mistral’s Strategic Neighbourhood

The investment arrived in a highly contextual window. On May 19, 2026 — just 13 days before the NP Co. announcement — Mistral AI acquired Emmi AI, an Austrian startup building physics models for industrial simulation . Two of Mistral’s own co-founders then personally invested in a Paris startup targeting nearly the same problem space. Coverage described the investor lineup as a “remarkably blue-chip set of believers” for a little-known company that had only recently rebranded .

Whether this amounts to an intentional hedging strategy or simply a bet on the strongest technical team, the signal is clear: physics simulation is being pulled into the foundation-model era, and some of Europe's most prominent AI figures are placing early bets on who will lead it.