EvolutionaryScale

High-Level Overview

EvolutionaryScale is an AI biotech startup that develops generative AI models to design novel proteins, making biology programmable for applications in human health, such as cancer treatments and carbon capture.[1][2][4] Headquartered in New York, the company builds the ESM3 platform—a family of models trained on 2.78 billion proteins using unprecedented compute (1 trillion teraflops)—which generates proteins like esmGFP that simulate 500 million years of evolution.[1][2][4] It serves scientists and pharma companies (including nine of the top ten global players via AWS integrations), solving the challenge of creating proteins beyond natural evolution to accelerate drug discovery and biotech innovation.[2][3] The company raised $142 million in seed funding led by Nat Friedman, Daniel Gross, Lux Capital, Amazon, and NVIDIA's NVentures, but was acquired by Chan Zuckerberg Biohub in 2025, transitioning its team to focus on long-term scientific research.[2][3]

Origin Story

Founded in 2023 by former Meta AI researchers, including co-founder and chief scientist Alexander Rives, EvolutionaryScale emerged after Meta disbanded its protein AI team.[1][2][3] The idea stemmed from their work at Meta on evolutionary scale modeling (ESM), evolving into ESM3 as a generative frontier model for biology.[2][3][4] Early traction came swiftly: in June 2024, they launched ESM3 and Cambrian models, alongside the massive $142M seed round, signaling investor confidence in AI-driven protein design.[2][3][4] A pivotal moment arrived with the 2025 acquisition by Chan Zuckerberg Biohub, where the full team joined as staff—Rives as Head of Science—shifting from commercial startup to a research-focused entity building AI for cellular digital twins and rapid discovery.[3]

Core Differentiators

• Generative AI for Biology: ESM3 is the first model reasoning simultaneously over protein sequence, structure, and function, generating novel proteins like esmGFP—a bright, stable fluorescent protein far from natural variants—trained on Earth's full protein diversity.[1][2][4]
• Unmatched Scale and Efficiency: 98-billion-parameter models using 1T teraflops compute; ESM Cambrian adds state-of-the-art representation learning for faster, efficient protein modeling.[3][4]
• Open and Accessible Tools: ESM3 family (small/medium/large) available via API, AWS platforms, and open-source ESM3-open on GitHub (non-commercial license), enabling fine-tuning for researchers and pharma.[2][4]
• Responsible Framework: Public benefit corporation emphasizing safe AI, scientific partnerships, and non-commercial openness, now amplified by Biohub's end-to-end data-AI-lab pipeline.[3][4][5]

Role in the Broader Tech Landscape

EvolutionaryScale rides the convergence of generative AI and biotech, accelerating protein design amid booming demand for AI in drug discovery and climate solutions.[2][4] Timing is ideal post-AlphaFold era, with market forces like massive VC interest ($142M seed) and cloud compute from AWS/NVIDIA enabling biology's "programmability" akin to software engineering.[2][3] It influences the ecosystem by open-sourcing models for global researchers, partnering with top pharma, and now via Biohub acquisition, fostering virtuous cycles of data, AI, and validation to compress decades of discovery into months—pioneering AI for cellular simulations.[3][4]

Quick Take & Future Outlook

Post-acquisition, EvolutionaryScale's team at Chan Zuckerberg Biohub will prioritize foundational AI for biology, developing datasets and models for digital cell twins to drive exponential breakthroughs in health and beyond.[3] Trends like multimodal AI (sequence/structure/function) and scalable compute will shape progress, potentially unlocking engineered proteins for pandemics, sustainability, and personalized medicine. Its influence evolves from startup innovator to open-science powerhouse, embedding AI deeply in biotech R&D and redefining how we "program" life. This positions it at the vanguard of making biology as malleable as code, echoing its origins in evolutionary simulation.[3][4]