How they use Nebius

SieveStack uses Nebius and TractoAI infrastructure to generate high-fidelity synthetic data through molecular dynamics simulations and train a multi-layer stack of foundational models on the resulting data. The work supports research into interactions between drugs and biological targets. Its goal is to unlock treatment pathways for hard-to-treat diseases. It combines physics-based molecular modeling with machine learning to simulate molecular interactions and predict laboratory experiment results. SieveStack’s own site and public posts describe the collaboration. Nebius describes SieveStack as building the world’s largest molecular simulation dataset.

Nebius’s customer story describes TractoAI’s unified compute platform, orchestration and parallelization that maintain GPU utilization above 90%. Training uses FP32 for accuracy and BF16 for performance. The story reports 30–50% shorter delays between test runs and production and training accelerated by up to four times. Nebius’s life-sciences page also reports 50% lower memory use and drug-target interactions revealed beyond the reach of laboratory experiments.

SieveStack worked with Nebius and TractoAI support teams to prototype, debug and scale its models. SieveStack says these services let the team focus on science instead of infrastructure. A later Nebius post describes reliable GPU access through Nebius as enabling the team to focus on discovery.