Electric Plant Raises $11.2M to Decode Plant Signals with AI Model Bombadil

A neuroscience-inspired startup has closed an $11.2 million seed round to turn plant bioelectricity into actionable farm intelligence. Electric Plant, founded by Justin Mast and DARPA-experienced neuroscientist Lee von Kraus, has spent three years building Bombadil — a large language model trained on microvolt-level signals from over 125 species, ranging from tomatoes to redwoods.

The company's stainless-steel sensors screw directly into the cambium layer, streaming real-time data via low-power wireless to cloud inference endpoints where Bombadil classifies stress states: water deficit, nutrient imbalance, light hunger. Growers receive irrigation and fertilizer adjustments through a dashboard, with early pilots claiming 20% water savings and yield gains.

Why it matters for GPU / AI infrastructure

Training a multimodal biological LLM across 125+ taxa demands heterogeneous compute: GPU clusters for transformer pre-training, CPU-heavy feature extraction from noisy time-series, and edge-optimized inference for field-deployed base stations. Electric Plant's architecture — cloud training, edge inference, continuous retraining loops — mirrors the hybrid workflows AiGpu cloud customers run daily. Low-cost sensors generating high-value inference is a textbook case for GPU-accelerated MLops at scale.

Seed investors Onto Ventures and Climactic joined Ponderosa, Endeavor Catalyst, and others backing the round. The capital funds sensor hardening, species expansion beyond the current hundred-plus, and multilingual grower interfaces — Arabic among them — as the company targets MENA greenhouse operators next.