Profluent

Profluent has become the marquee proof that generative AI can design working biology, and in 2026 that reputation converted into franchise-scale validation. In April 2026 it announced a strategic partnership with Eli Lilly to design AI-generated site-specific recombinases for genetic medicine, with Profluent eligible for up to $2.25 billion in milestones plus tiered royalties, an upfront payment, and committed R&D funding — a top-tier pharma effectively endorsing the thesis that protein design is a language-modeling problem. That followed OpenCRISPR-1, the first AI-designed gene editor to edit the human genome, released openly and drawing over 1,200 license requests, and an October 2025 Corteva partnership extending the platform into agricultural gene editing. The technical foundation scaled dramatically with ProGen3 (announced April 2025): sparse protein language models from 112M to 46B parameters, pre-trained on 1.5 trillion tokens across 3.4 billion sequences, demonstrating that AI scaling laws hold for protein generation the way they do for text. The November 2025 $106M Series B, co-led by Altimeter and Bezos Expeditions (roughly $150M total raised), capitalizes a capital-light model: Profluent designs proteins and licenses them across therapeutics, agriculture, and biomanufacturing rather than running its own trials. If generative protein design becomes the default way novel enzymes and editors are created, Profluent — founded by ProGen's originator Ali Madani — owns the frontier.

Profluent's headline numbers flatter a young, roughly 50-person company: the Lilly deal is "up to $2.25 billion" in milestones-and-royalties — a biobucks structure whose realized value depends entirely on preclinical recombinases surviving years of development that Lilly, not Profluent, controls, with the upfront a small fraction of the total. OpenCRISPR-1 generated over 1,200 license requests but was released openly, which builds mindshare while giving away the crown jewel; converting free interest into durable revenue is unproven. The competitive field is the most intense in science: every major AI lab, cloud provider, and a wave of AI-protein startups are chasing generative biology, and the moat around a protein language model is thin when the underlying techniques, datasets, and even architectures diffuse quickly through open publication — Profluent itself competes against far larger compute budgets. Scaling laws for proteins are a compelling narrative, but a 46B-parameter model is expensive to train and does not guarantee that generated proteins are developable, manufacturable, or safe in vivo — designing a sequence is not the same as delivering a therapy. The company's value rests on partner programs it does not own, in a field where a single high-profile clinical failure of an AI-designed protein could reprice the entire thesis. Profluent is the frontier's most visible name, which means it also carries the frontier's disappointment risk.