Simulacra AI uses Nebius distributed compute to train a large wavefunction model for quantum chemistry. Nebius lists its model size as 100M+ parameters. The model aims to predict how electrons behave in molecules and to generate data for drug and materials research without repeating a full quantum calculation for every molecule. Its pre-training approach incorporates first-principles quantum mechanics without requiring labelled data. The stated eventual goal is billion-parameter models and infrastructure for calculating molecular properties, including time dynamics. Nebius also reports that Simulacra generates high-precision datasets for molecular dynamics models and delivers molecular data for other companies’ in silico discovery pipelines. Nebius says this enables companies to refine those pipelines without maintaining broad internal infrastructure to train models.
Nebius's customer story describes Simulacra splitting model weights and data across GPUs to train beyond its earlier 10M-parameter limit. The June 2025 article identifies single program, multiple data (SPMD) partitioning. The customer story also describes ahead-of-time compilation and caching that cut some compilation runs from more than two hours to minutes. Nebius's life-sciences page reports that its largest models take 10–20 minutes to compile for pre-training, compared with more than two hours previously.
A June 2025 Nebius article reports access to 16 NVIDIA H200 GPUs and NVIDIA BioNeMo for developing generative chemistry models. These models are intended to produce synthetic training data or predict molecular properties at scale. Simulacra's CEO describes using BioNeMo NIM microservices, including GenMol and MolMIM, to generate structurally diverse molecules for training datasets. A paper benchmarking Simulacra AI’s quantum-accurate synthetic data generation also acknowledges Nebius resources and services supporting the research project. The paper attributes improved scalability to GPU parallelization and describes a sampling scheme that decouples spectral-gap size from system dimension.
Nebius life sciences and healthcare: customer workloads
27 Sept 2026 · Nebius · Web page
Nebius reports that Simulacra AI used its infrastructure to scale molecular wavefunction models and generate molecular datasets for drug and materials discovery. The page lists 100M+ model parameters and reports pre-training compilation times of 10–20 minutes, compared with more than two hours previously.
Nebius life sciences and healthcare: customer workloads27 Sept 2026 · Nebius · Web pageNebius reports that Simulacra AI used its infrastructure to scale molecular wavefunction models and generate molecular datasets for drug and materials discovery. The page lists 100M+ model parameters and reports pre-training compilation times of 10–20 minutes, compared with more than two hours previously.
Nebius From genome analysis to quantum chemistry: Nebius powers the next generation of biotech research with…
25 Jun 2025 · Nebius · Web page
Nebius describes Simulacra AI distributing its molecular wavefunction model across GPUs using SPMD partitioning. It reports access to 16 NVIDIA H200 GPUs and BioNeMo through Nebius infrastructure. Simulacra AI CEO Aleksei Ustimenko describes using GenMol and MolMIM microservices to generate structurally diverse molecules for training datasets.
Nebius From genome analysis to quantum chemistry: Nebius powers the next generation of biotech research with…25 Jun 2025 · Nebius · Web pageNebius describes Simulacra AI distributing its molecular wavefunction model across GPUs using SPMD partitioning. It reports access to 16 NVIDIA H200 GPUs and BioNeMo through Nebius infrastructure. Simulacra AI CEO Aleksei Ustimenko describes using GenMol and MolMIM microservices to generate structurally diverse molecules for training datasets.
Nebius Customer story
14 Apr 2025 · Nebius · Web page
Nebius describes Simulacra AI using its distributed compute infrastructure to pre-train molecular wavefunction models. Scaling beyond 10 million parameters requires model partitioning across GPUs.
Nebius Customer story14 Apr 2025 · Nebius · Web pageNebius describes Simulacra AI using its distributed compute infrastructure to pre-train molecular wavefunction models. Scaling beyond 10 million parameters requires model partitioning across GPUs.
Benchmarking simulacra AI’s Quantum Accurate Synthetic Data Generation for Chemical Sciences
16 Oct 2026 · arxiv.org · Web page
The authors acknowledge Nebius resources and services supporting their research benchmarking Simulacra AI’s quantum-accurate synthetic data generation for chemical sciences. The paper attributes improved scalability to GPU parallelization and describes a sampling scheme that decouples spectral-gap size from system dimension.
Benchmarking simulacra AI’s Quantum Accurate Synthetic Data Generation for Chemical Sciences16 Oct 2026 · arxiv.org · Web pageThe authors acknowledge Nebius resources and services supporting their research benchmarking Simulacra AI’s quantum-accurate synthetic data generation for chemical sciences. The paper attributes improved scalability to GPU parallelization and describes a sampling scheme that decouples spectral-gap size from system dimension.
Nebius Describes Simulacra quantum-chemistry model training on Nebius
3 Jan 2025 · Nebius · Web page
Nebius states that Simulacra AI used Nebius infrastructure to overcome scalability and efficiency challenges while developing a molecular wavefunction model and generating high-precision datasets for molecular dynamics and drug and materials discovery pipelines. It says Simulacra's molecular data enables companies to refine in silico pipelines without maintaining broad internal infrastructure to train models.
Nebius Describes Simulacra quantum-chemistry model training on Nebius3 Jan 2025 · Nebius · Web pageNebius states that Simulacra AI used Nebius infrastructure to overcome scalability and efficiency challenges while developing a molecular wavefunction model and generating high-precision datasets for molecular dynamics and drug and materials discovery pipelines. It says Simulacra's molecular data enables companies to refine in silico pipelines without maintaining broad internal infrastructure to train models.
Simulacra AI, operating as Simulacra Research Inc., develops AI models for quantum simulation and molecular property prediction. Its Large Wavefunction Models combine deep learning with quantum mechanics to support research in pharmaceuticals, materials science and chemical engineering. Its open-source Hamilton-Zero model learns quantum ground states across different physical systems. The company also develops high-fidelity molecular property datasets for drug and materials discovery, including data used to train molecular dynamics models and predict chemical properties.
Simulacra releases Hamilton-Zero quantum simulation model
Simulacra Research released Hamilton-Zero with public model weights, code and datasets for computing quantum-system ground states. The team reports fine-tuning on systems up to 1,024 qubits and evaluation on systems up to 8,100 qubits.
Simulacra AI uses Nebius distributed compute to train a large wavefunction model for quantum chemistry. Nebius lists its model size as 100M+ parameters. The model aims to predict how electrons behave in molecules and to generate data for drug and materials research without repeating a full quantum calculation for every molecule. Its pre-training approach incorporates first-principles quantum mechanics without requiring labelled data. The stated eventual goal is billion-parameter models and infrastructure for calculating molecular properties, including time dynamics. Nebius also reports that Simulacra generates high-precision datasets for molecular dynamics models and delivers molecular data for other companies’ in silico discovery pipelines. Nebius says this enables companies to refine those pipelines without maintaining broad internal infrastructure to train models.
Nebius's customer story describes Simulacra splitting model weights and data across GPUs to train beyond its earlier 10M-parameter limit. The June 2025 article identifies single program, multiple data (SPMD) partitioning. The customer story also describes ahead-of-time compilation and caching that cut some compilation runs from more than two hours to minutes. Nebius's life-sciences page reports that its largest models take 10–20 minutes to compile for pre-training, compared with more than two hours previously.
A June 2025 Nebius article reports access to 16 NVIDIA H200 GPUs and NVIDIA BioNeMo for developing generative chemistry models. These models are intended to produce synthetic training data or predict molecular properties at scale. Simulacra's CEO describes using BioNeMo NIM microservices, including GenMol and MolMIM, to generate structurally diverse molecules for training datasets. A paper benchmarking Simulacra AI’s quantum-accurate synthetic data generation also acknowledges Nebius resources and services supporting the research project. The paper attributes improved scalability to GPU parallelization and describes a sampling scheme that decouples spectral-gap size from system dimension.
Nebius life sciences and healthcare: customer workloads
27 Sept 2026 · Nebius · Web page
Nebius reports that Simulacra AI used its infrastructure to scale molecular wavefunction models and generate molecular datasets for drug and materials discovery. The page lists 100M+ model parameters and reports pre-training compilation times of 10–20 minutes, compared with more than two hours previously.
Nebius life sciences and healthcare: customer workloads27 Sept 2026 · Nebius · Web pageNebius reports that Simulacra AI used its infrastructure to scale molecular wavefunction models and generate molecular datasets for drug and materials discovery. The page lists 100M+ model parameters and reports pre-training compilation times of 10–20 minutes, compared with more than two hours previously.
Nebius From genome analysis to quantum chemistry: Nebius powers the next generation of biotech research with…
25 Jun 2025 · Nebius · Web page
Nebius describes Simulacra AI distributing its molecular wavefunction model across GPUs using SPMD partitioning. It reports access to 16 NVIDIA H200 GPUs and BioNeMo through Nebius infrastructure. Simulacra AI CEO Aleksei Ustimenko describes using GenMol and MolMIM microservices to generate structurally diverse molecules for training datasets.
Nebius From genome analysis to quantum chemistry: Nebius powers the next generation of biotech research with…25 Jun 2025 · Nebius · Web pageNebius describes Simulacra AI distributing its molecular wavefunction model across GPUs using SPMD partitioning. It reports access to 16 NVIDIA H200 GPUs and BioNeMo through Nebius infrastructure. Simulacra AI CEO Aleksei Ustimenko describes using GenMol and MolMIM microservices to generate structurally diverse molecules for training datasets.
Nebius Customer story
14 Apr 2025 · Nebius · Web page
Nebius describes Simulacra AI using its distributed compute infrastructure to pre-train molecular wavefunction models. Scaling beyond 10 million parameters requires model partitioning across GPUs.
Nebius Customer story14 Apr 2025 · Nebius · Web pageNebius describes Simulacra AI using its distributed compute infrastructure to pre-train molecular wavefunction models. Scaling beyond 10 million parameters requires model partitioning across GPUs.
Benchmarking simulacra AI’s Quantum Accurate Synthetic Data Generation for Chemical Sciences
16 Oct 2026 · arxiv.org · Web page
The authors acknowledge Nebius resources and services supporting their research benchmarking Simulacra AI’s quantum-accurate synthetic data generation for chemical sciences. The paper attributes improved scalability to GPU parallelization and describes a sampling scheme that decouples spectral-gap size from system dimension.
Benchmarking simulacra AI’s Quantum Accurate Synthetic Data Generation for Chemical Sciences16 Oct 2026 · arxiv.org · Web pageThe authors acknowledge Nebius resources and services supporting their research benchmarking Simulacra AI’s quantum-accurate synthetic data generation for chemical sciences. The paper attributes improved scalability to GPU parallelization and describes a sampling scheme that decouples spectral-gap size from system dimension.
Nebius Describes Simulacra quantum-chemistry model training on Nebius
3 Jan 2025 · Nebius · Web page
Nebius states that Simulacra AI used Nebius infrastructure to overcome scalability and efficiency challenges while developing a molecular wavefunction model and generating high-precision datasets for molecular dynamics and drug and materials discovery pipelines. It says Simulacra's molecular data enables companies to refine in silico pipelines without maintaining broad internal infrastructure to train models.
Nebius Describes Simulacra quantum-chemistry model training on Nebius3 Jan 2025 · Nebius · Web pageNebius states that Simulacra AI used Nebius infrastructure to overcome scalability and efficiency challenges while developing a molecular wavefunction model and generating high-precision datasets for molecular dynamics and drug and materials discovery pipelines. It says Simulacra's molecular data enables companies to refine in silico pipelines without maintaining broad internal infrastructure to train models.
Simulacra AI, operating as Simulacra Research Inc., develops AI models for quantum simulation and molecular property prediction. Its Large Wavefunction Models combine deep learning with quantum mechanics to support research in pharmaceuticals, materials science and chemical engineering. Its open-source Hamilton-Zero model learns quantum ground states across different physical systems. The company also develops high-fidelity molecular property datasets for drug and materials discovery, including data used to train molecular dynamics models and predict chemical properties.
Simulacra releases Hamilton-Zero quantum simulation model
Simulacra Research released Hamilton-Zero with public model weights, code and datasets for computing quantum-system ground states. The team reports fine-tuning on systems up to 1,024 qubits and evaluation on systems up to 8,100 qubits.
