One Goldman Sachs and Moody's Veteran Is Backing This $112 Million Bet on Risk Language Models

Earthian AI is an Amsterdam-based startup building a “financial inference layer” that uses AI to translate real-world events into dollar values and risk exposure for banks, insurers, and asset managers. The company aims to replace slow disaster pricing processes that previously took months. Earthian announced a pre-seed raise valuing the company at $112 million, supported by Rabobank and NVIDIA’s Inception program, with additional backing from former executives at Goldman Sachs, Moody’s, and Accenture. The startup reports that five Fortune 500 firms use its risk models today and expects growth to 30 by 2027. It focuses on small language models trained for financial reasoning, emphasizing speed, lower energy use, regulatory suitability, and reduced hallucination risk.

"Earthian AI wants to do it in real time. The Amsterdam-based startup, founded by former Harvard researcher Shayan Shokri, is building what it calls a “financial inference layer,” an AI system designed to tell banks, insurers, and asset managers not just what's happening in the world, but also what those events are actually worth in dollars and their risk exposure."

"On April 30, Earthian announced apre-seed raise $112 million valuation - structured as equity plus debt - with backing from Rabobank and NVIDIA's Inception program. The angel round also drew former executives from Goldman Sachs, Moody's, and Accenture. “Five Fortune 500 firms are using our risk models today, and we expect that to be 30 by 2027,” Shokri said in a statement."

"Natural catastrophes caused $107 billion in insured losses globally in 2025, the sixth straight year that figure has exceeded $100 billion. The Los Angeles wildfires alone accounted for $40 billion. Add in geopolitical fragmentation, climate volatility, and AI-driven cyber threats, and the math is moving faster than the analysts trying to do it."

"While large language models dominate most public AI conversations, Earthian is focused on SLMs, or small language models, trained specifically for financial reasoning. “A large language model can write you a sonnet, but it shouldn't be pricing your catastrophe exposure,” Shokri said. “Our small language models run faster, use less energy, and hold up better in regulated environments. They don't hallucinate on the kind of financial”"

Read at Entrepreneur