Jeff Bezos just bet $106M on AI that designs proteins that don't exist in nature

Jeff Bezos just bet $106M on AI that designs proteins that don't exist in nature

Profluent raised the funding this week to scale what they call "frontier AI models" for programmable biology. Their tech is already being used by major pharma companies to design gene editors, antibodies, and enzymes from scratch.

Some of you may wonder: what exactly makes this a "frontier" model? Let me explain:

• Most traditional machine-learning type AI is specialized and narrow, it does one task well within fixed limits. Your Netflix recommendations, spam filters, even many "AI-powered" business tools.

• Frontier models are fundamentally different: Built on massive datasets where more data = genuinely smarter models (Profluent has 115B proteins, the world's largest collection)

• Ideally, they enable scientific firsts: Doing things never done before like creating the first CRISPR system entirely from AI, published in Nature

• Foundation-grade: I don't think there is a formal definition out there, but generally we talk about large, general-purpose systems that enable multiple applications, not just one narrow use case

• True scaling laws: Performance improves dramatically (or at least should...) with more compute and data, not just incrementally

The practical difference? We're moving from "screen millions of molecules hoping to find one that works" to "describe what you need and AI designs it." (one step at a time that is...)

That's the shift frontier models should enable – and why Bezos is paying attention.