MIT and Recursion Launch an Open‑Source Protein Co‑Folding Model Accelerated by NVIDIA

• Products
  Cloud Services Data Center Embedded Systems Gaming and Creating Graphics Cards and GPUs Laptops Networking Professional Workstations Software Tools

  Cloud Services

  BioNeMo

  AI-driven platform for life sciences research and discovery

  DGX Cloud

  Fully managed end-to-end AI platform on leading clouds

  NVIDIA APIs

  Explore, test, and deploy AI models and agents

  Omniverse Cloud

  Integrate advanced simulation and AI into complex 3D workflows

  Private Registry

  Guide for using NVIDIA NGC private registry with GPU cloud

  NVIDIA NGC

  Accelerated, containerized AI models and SDKs

  Data Center

  Overview

  Modernizing data centers with AI and accelerated computing

  DGX Platform

  Enterprise AI factory for model development and deployment

  Grace CPU

  Architecture for data centers that transform data into intelligence

  HGX Platform

  A supercomputer purpose-built for AI and HPC

  IGX Platform

  Advanced functional safety and security for edge AI

  MGX Platform

  Accelerated computing with modular servers

  OVX Systems

  Scalable data center infrastructure for high-performance AI

  Embedded Systems

  Jetson

  Leading platform for autonomous machines and embedded applications

  DRIVE AGX

  Powerful in-vehicle computing for AI-driven autonomous vehicle systems

  Clara AGX

  AI-powered computing for innovative medical devices and imaging

  Gaming and Creating

  GeForce

  Explore graphics cards, gaming solutions, AI technology, and more

  GeForce Graphics Cards

  RTX graphics cards bring game-changing AI capabilities

  Gaming Laptops

  Thinnest and longest lasting RTX laptops, optimized by Max-Q

  G-SYNC Monitors

  Smooth, tear-free gaming with NVIDIA G-SYNC monitors

  DLSS

  Neural rendering tech boosts FPS and enhances image quality

  Reflex

  Ultimate responsiveness for faster reactions and better aim

  RTX AI PCs

  AI PCs for gaming, creating, productivity and development

  NVIDIA Studio

  High performance laptops and desktops, purpose-built for creators

  GeForce NOW Cloud Gaming

  RTX-powered cloud gaming. Choose from 3 memberships

  NVIDIA App

  Optimize gaming, streaming, and AI-powered creativity

  NVIDIA Broadcast App

  AI-enhanced voice and video for next-level streams, videos, and calls

  SHIELD TV

  World-class streaming media performance

  Graphics Cards and GPUs

  Blackwell Architecture

  The engine of the new industrial revolution

  Hopper Architecture

  High performance, scalability, and security for every data center

  Ada Lovelace Architecture

  Performance and energy efficiency for endless possibilities

  GeForce Graphics Cards

  RTX graphics cards bring game-changing AI capabilities

  NVIDIA RTX PRO

  Accelerating professional AI, graphics, rendering and compute workloads

  Virtual GPU

  Virtual solutions for scalable, high-performance computing

  Laptops

  GeForce Laptops

  GPU-powered laptops for gamers and creators

  Studio Laptops

  High performance laptops purpose-built for creators

  NVIDIA RTX PRO Laptops

  Accelerate professional AI and visual computing from anywhere

  Networking

  Overview

  Accelerated networks for modern workloads

  DPUs and SuperNICs

  Software-defined hardware accelerators for networking, storage, and security

  Ethernet

  Ethernet performance, availability, and ease of use across a wide range of applications

  InfiniBand

  High-performance networking for super computers, AI, and cloud data centers

  Networking Software

  Networking software for optimized performance and scalability

  Network Acceleration

  IO subsystem for modern, GPU-accelerated data centers

  Professional Workstations

  Overview

  Accelerating professional AI, graphics, rendering, and compute workloads

  DGX Spark

  A Grace Blackwell AI Supercomputer on your desk

  DGX Station

  The ultimate desktop AI supercomputer powered by NVIDIA Grace Blackwell

  NVIDIA RTX PRO AI Workstations

  Accelerate innovation and productivity in AI workflows

  NVIDIA RTX PRO Desktops

  Powerful AI, graphics, rendering, and compute workloads

  NVIDIA RTX PRO Laptops

  Accelerate professional AI and visual computing from anywhere

  Software

  NeMo Agent toolkit

  AI Blueprints

  AI Inference - Dynamo

  AI Inference Microservices - NIM

  AI Microservices - CUDA-X

  Automotive - DRIVE

  Avatar - Tokkio

  Cloud-AI Video Streaming - Maxine

  Cybersecurity - Morpheus

  Data Science - Apache Spark

  Data Science - RAPIDS

  Decision Optimization - cuOpt

  Generative AI - NeMo

  Healthcare - Clara

  Industrial AI - Omniverse

  Intelligent Video Analytics - Metropolis

  NVIDIA Mission Control

  NVIDIA AI Enterprise Platform

  Physical AI - Cosmos

  NVIDIA Run:ai

  Robotics - Isaac

  Telecommunications - Aerial

  See All Software

  Tools

  AI Workbench

  Simplify AI development with NVIDIA AI Workbench on GPUs

  API Catalog

  Explore NVIDIA's AI models, blueprints, and tools for developers

  Data Center Management

  AI and HPC software solutions for data center acceleration

  GPU Monitoring

  Monitor and manage GPU performance in cluster environments

  Nsight

  Explore NVIDIA developer tools for AI, graphics, and HPC

  NGC Catalog

  Discover GPU-optimized AI, HPC, and data science software

  NVIDIA App for Laptops

  Optimize enterprise GPU management

  NVIDIA NGC

  Accelerate AI and HPC workloads with NVIDIA GPU Cloud solutions

  Desktop Manager

  Enhance multi-display productivity with NVIDIA RTX Desktop Manager

  RTX Accelerated Creative Apps

  Creative tools and AI-powered apps for artists and designers

  Video Conferencing

  AI-powered audio and video enhancement

• Solutions
  Artificial Intelligence Cloud and Data Center Design and Simulation High-Performance Computing Robotics and Edge AI Autonomous Vehicles

  Artificial Intelligence

  Overview

  Add intelligence and efficiency to your business with AI and machine learning

  Agentic AI

  Build AI agents designed to reason, plan, and act

  AI Data

  Powering a new class of enterprise infrastructure for AI

  Conversational AI

  Enables natural, personalized interactions with real-time speech AI

  Cybersecurity

  AI-driven solutions to strengthen cybersecurity and AI infrastructure

  Data Science

  Iterate on large datasets, deploy models more frequently, and lower total cost

  Generative AI

  Instantly run and deploy Generative AI

  Inference

  Drive breakthrough performance with AI-enabled applications and services

  Cloud and Data Center

  Overview

  Powering AI, HPC, and modern workloads with NVIDIA

  AI Data Platform for Enterprise

  Bringing enterprise storage into the era of agentic AI

  Accelerated Computing

  Accelerated computing uses specialized hardware to boost IT performance

  Cloud Computing

  On-demand IT resources and services, enabling scalability and intelligent insights

  Colocation

  Accelerate the scaling of AI across your organization

  MLOps

  Accelerate AI with MLOps

  Networking

  High speed ethernet interconnect solutions and services

  Sustainable Computing

  Save energy and lower cost with AI and accelerated computing

  Virtualization

  NVIDIA virtual GPU software delivers powerful GPU performance

  Design and Simulation

  Overview

  Streamline building, operating, and connecting metaverse apps

  Computer Aided-Engineering

  Develop real-time interactive design using AI-accelerated real-time digital twins

  Digital Twin Development

  Harness the power of large-scale, physically-based OpenUSD simulation

  Rendering

  Bring state-of-the-art rendering to professional workflows

  Robotic Simulation

  Innovative solutions to take on your robotics, edge, and vision AI challenges

  Scientific Visualization

  Enablies researchers to visualize their large datasets at interactive speeds

  Vehicle Simulation

  AI-defined vehicles are transforming the future of mobility

  Extended Reality

  Transform workflows with immersive, scalable interactions in virtual environments

  High-Performance Computing

  Overview

  Discover NVIDIA's HPC solutions for AI, simulation, and accelerated computing

  HPC and AI

  Boost accuracy with GPU-accelerating HPC and AI

  Scientific Visualization

  Enables researchers to visualize large datasets at interactive speeds

  Simulation and Modeling

  Accelerate simulation workloads

  Quantum Computing

  Fast-tracking the advancement of scientific innovations with QPUs

  Robotics and Edge AI

  Overview

  Innovative solutions to take on robotics, edge, and vision AI challenges

  Robotics

  GPU-accelerated advances in AI perception, simulation, and software

  Edge AI

  Bring the power of NVIDIA AI to the edge for real-time decision-making solutions

  Vision AI

  Transform data into valuable insights using vision AI

  Autonomous Vehicles

  Overview

  AI-enhanced vehicles are transforming the future of mobility

  Infrastructure

  Essential data center tools for safe autonomous vehicle development

  Simulation

  Explore high-fidelity sensor simulation for safe autonomous vehicle development

  In-Vehicle Computing

  Develop automated driving functions and immersive in-cabin experiences

  Safety

  State-of-the-art system for AV safety, from the cloud to the car

• Industries
  Industries

  Overview

  Architecture, Engineering, Construction & Operations

  Automotive

  Consumer Internet

  Cybersecurity

  Energy

  Financial Services

  Healthcare and Life Sciences

  Higher Education

  Game Development

  Global Public Sector

  Manufacturing

  Media and Entertainment

  US Public Sector

  Restaurants

  Retail and CPG

  Robotics

  Smart Cities

  Supercomputing

  Telecommunications

  Transportation

• Shop
• Drivers
• Support

• US
• Log In LogOut

• 0
• US
• Login LogOut

• • Shop
  • Drivers
  • Support

Healthcare and Life Sciences

MIT and Recursion Launch an Open-Source Protein Co-Folding Model Accelerated by NVIDIA

Overview

MIT's Jameel Clinic and CSAIL researchers, partnering with Recursion, built Boltz-2. This first fully open-source biomolecular foundation model co-folds protein complexes and simultaneously predicts their binding affinities with accuracy rivaling physics-based FEP methods. Leveraging the BioHive-2 GPU supercomputer, cuEquivariance kernels, and a ready-to-deploy NIM™ microservice, the team trained and released an open-sourced Boltz-2 (code + weights under an MIT license), delivering accelerated structure-and-affinity predictions for broad, community-scale drug-discovery workflows.

Key Takeaways

From AlphaFold to an Open Foundation Model

Boltz-2 marks a decisive inflection point for AI-driven drug discovery. Developed by MIT's Jameel Clinic and CSAIL in partnership with Recursion, the model unifies complex-structure prediction and binding-affinity estimation in a single, open-source package. Trained on Recursion's BioHive-2 supercomputer and accelerated by NVIDIA cuEquivariance kernels, it approaches the chemical accuracy of physics-based free-energy perturbation (FEP) simulations while returning results in about 20 seconds on one A100 GPU. With Boltz-2 available (code, weights) under an MIT license, any organization can retrain, fine-tune, and deploy Boltz-2 without legal or operational friction. Enterprises looking to scale inference can also deploy the accelerated NVIDIA Boltz-2 NIM for production use cases under an NVIDIA AI Enterprise license.

The success of DeepMind's AlphaFold2 in 2020 commoditized single-protein structure prediction, with its successor, AlphaFold3, enabling the advanced modeling of multiple molecules simultaneously (DNA, RNA, protein, and small molecules). MIT released an open-source alternative to AlphaFold3 in late 2024 with Boltz-1, the first open model capable of co-folding biomolecular complexes at near-AlphaFold3 accuracy. Six months later, the same team introduced Boltz-2, adding molecule-target affinity prediction functionality and retraining on ~3 million assay-labeled examples to predict both pose and potency in one forward pass.

Across a community benchmark for affinity predictions, Boltz-2 achieves a mean Pearson correlation of roughly 0.62-0.66, much closer than previous comparable models, yet 1,000 times faster than state-of-the-art physics pipelines. On CASP16 affinity data, the model ranked first overall, and in retrospective high-throughput screens (MF-PCBA), it doubled the average precision relative to docking and prior ML approaches. A single A100 processes a ligand-protein pair in ~20 GPU-seconds; scaling across BioHive-2 lowers wall-clock times for million-compound libraries to hours, not weeks.

Engineering Boltz-2 at Supercomputer Scale

Boltz-2's leap in capability rests on two engineering pillars.

First, training ran on BioHive-2, which ranks #35 on the TOP500 list of the most powerful supercomputers in the world and delivers a four-fold speedup versus its predecessor. BioHive-2 is an NVIDIA DGX SuperPOD™ AI supercomputer, powered by 63 DGX™ H100 systems with NVIDIA H100 Tensor Core GPUs interconnected by NVIDIA Quantum-2 InfiniBand networking. BioHive-2 builds on Recursion's success with its first supercomputer, BioHive-1, and was designed to handle the immense workloads that Recursion puts it through, processing the 50 petabytes of biological, chemical, and patient data that the company has amassed. The goal is to direct Recursion scientists to wet-lab experiments that have the most potential for success.

"With AI in the loop today, we can get 80% of the value with 40% of the wet lab work, and that ratio will improve going forward," said Recursion Chief Technology Officer Ben Mabey.

Second, scaling a model like Boltz-2 required highly efficient training and inference. NVIDIA engineers profiled the model pre-release and pinpointed triangle attention and triangle multiplication as the key bottlenecks. Partnering closely with the Boltz-2 research teams, NVIDIA built and validated custom cuEquivariance kernels that deliver up to 5x faster triangle operations, reduce memory usage, and integrate seamlessly into any structure-prediction model, starting with Boltz-2 itself, delivering up to 2x-3x training and inference speedups. These kernels now power the trunk and PairFormer layers, ensuring quadratic scaling across GPUs compared to cubic complexity in the original implementation. cuEquivariance is available for developers on GitHub and is already becoming widely adopted in the ecosystem.

"These kernels are long-awaited and will become an integral part of the Boltz family of models, helping address the bottlenecks in speed and memory consumption," said Gabriele Corso, researcher at MIT.

Enterprise-Ready NIM Deployment

Boltz-2 NIM is NVIDIA's day-zero, production-ready packaging of MIT and Recursion's open-source Boltz-2 foundation model, delivered as an accelerated API microservice. The microservice ingests protein, DNA, RNA, or ligand sequences and returns high-confidence 3D complex structures and binding affinity predictions at up to 2x-3x acceleration over the open-source alternative.

For enterprise biopharma, this translates into higher throughput and materially lower compute spend, compressing structure-affinity design cycles and accelerating your R&D pipeline. While the Boltz-2 NIM can be tested in R&D environments at no cost here, product-grade deployment requires NVIDIA AI Enterprise support.

Learn about the Bolz-2 NIM and the rest of the NVIDIA AI platform for healthcare and life sciences today.

Learn more about NVIDIA solutions for healthcare and life sciences.