Problem

Somite AI addresses the central bottleneck in regenerative medicine: turning pluripotent stem cells into reliably pure, scalable, mature human cell types. Today, researchers largely discover differentiation protocols through slow, manual trial and error; the process can take years and tens of millions of dollars without guaranteeing a usable result. The problem is compounded by poor purity, limited manufacturing scale, and batch-to-batch reproducibility. The company has framed the gap starkly: researchers can reliably produce only about 10 cell types, versus more than 5,000 types in the human cell atlas.

The killer use case is cell replacement: manufacturing healthy tissue to replenish cells lost or damaged by disease. Somite began with somite-derived tissues such as muscle, cartilage, bone, and tendon, and targeted conditions including muscular dystrophy, diabetes, and obesity. Its most concrete current application is Parkinson’s disease, where the renamed company, Cellular Intelligence, acquired global rights to Novo Nordisk’s clinical-stage stem-cell-derived dopaminergic progenitor program.

Product / Service

The company is building an AI-native TechBio platform rather than a conventional standalone software product. Its proprietary capsule technology places cells in semi-permeable capsules and exposes them to controlled sequences of biological signals. The resulting cell-state transitions are barcoded and measured at very large scale, creating a dataset of how combinations, timing, and strength of signals steer cells toward different fates. Somite reported generating this data at roughly 1,000 times the efficiency of existing methods and scaling from a 27,000-condition pilot to more than one million sequential signal combinations in 2025.

That data trains a foundation model—originally described as DeltaStem and now presented within Cellular Intelligence’s universal virtual cell-signaling model—which predicts the protocols needed to create particular cell types. An active-learning loop uses model uncertainty to select the next experiments, allowing the system to improve as it generates therapies and manufacturing data. The practical benefit is to replace empirical protocol hunting with more predictive design, improving purity, scalability, reliability, and development speed. Public disclosures indicate that the platform is being applied to the company’s own therapeutic programs and strategic clinical collaborations, not sold as a clearly documented self-serve SaaS product.

Market

Somite competes in AI-native TechBio, virtual-cell modeling, regenerative medicine, and cell-therapy development and manufacturing. The opportunity is unusually broad: the company’s initial disease areas were associated with a roughly $250 billion therapy market, while the platform could ultimately support protocol design, disease modeling, drug-effect prediction, and production of many human cell types. Its closest competitive set is better understood as adjacent rather than identical. Cellarity uses multi-omics AI to model cellular behavior for drug discovery; Arc Institute and Tahoe Therapeutics are building large virtual-cell datasets; and Recursion is developing a computational “virtual cell.” These efforts overlap in AI-driven biological modeling, although the evidence does not identify a direct like-for-like competitor to Somite’s capsule-based sequential-signaling approach.

The company is not best characterized as a proven commercial software vendor or publicly established revenue business; available disclosures do not report product revenue. It has, however, moved beyond a purely preclinical concept. Founded in 2023, it raised more than $47 million in a 2025 Series A and reported more than $60 million raised by May 2026. After rebranding as Cellular Intelligence in January 2026, it secured global rights to Novo Nordisk’s Phase 1/2-ready Parkinson’s cell-therapy program; Novo also became an equity investor and remains eligible for milestones and royalties. The platform itself had reported in vitro and in silico proof-of-concept results and million-condition experimental validation, making clinical translation and platform validation its principal disclosed traction.