Sooth Labs Raises $50M Seed Led by Felicis to Build AI Models That Forecast Geopolitical and Market Events (April 2026)

Sooth Labs, an AI startup founded by former Meta Platforms employees to build models that forecast geopolitical and market events, closed a $50 million seed round at a $335 million post-money valuation on April 22, 2026. Felicis led the round, with personal checks from Meta’s former chief AI scientist Yann LeCun and Google’s chief scientist Jeff Dean, and advisory support from Meta CTO Andrew Bosworth.

What Happened

Bloomberg first reported the round on April 22, 2026. According to filings and people familiar with the deal, the company — which operates publicly under the domain sooth.inc — is building what it calls a world model for global foresight: an AI system that ingests events, news, market signals, and multimodal data to deliver “calibrated, auditable long-horizon forecasts” for institutional clients.

The pitch is squarely targeted at finance, defense, insurance, and real-estate firms whose decisions depend on probability estimates of specific future outcomes — rate moves, election outcomes, supply-chain shocks, conflict escalations — rather than on chat-style assistance. Sooth has reportedly already published two probabilistic forecasts publicly to demonstrate the system, though full product details remain under wraps.

Key Details

• Round: $50 million seed, post-money valuation $335 million.
• Lead investor: Felicis Ventures, a firm with a strong recent track record in frontier AI seed deals (Periodic Labs, Together AI, Adept).
• Notable backers: Yann LeCun (Turing Award winner, former Meta chief AI scientist) and Jeff Dean (Google chief scientist) participated personally.
• Advisor: Andrew Bosworth, Meta’s CTO, is advising the founders.
• Founders: Former Meta Platforms employees; specific names have not been disclosed publicly. Coverage notes connections to AI research backgrounds.
• Use cases: Capital allocation, risk management, geopolitical and market-event probability scoring for finance, defense, insurance, and real estate.

What Developers and Users Are Saying

Reaction on Twitter/X and Hacker News has been a mix of curiosity and skepticism. The most-shared X post about the deal — @Discoplomacy — framed the launch as “a new AI lab, this one looking at forecasting geopol and market events,” noting the LeCun/Dean stamp of approval.

Skeptics point out that calibrated long-horizon forecasting has been an unsolved problem since Philip Tetlock’s Good Judgment Project in the 2010s, and that LLM-based forecasters have historically performed at chance on questions where the ground truth lies more than a few months out. Optimists counter that Sooth’s plan to combine multimodal training data (video, audio, text) with proprietary client signals is closer in spirit to bespoke risk models at hedge funds than to consumer chatbots — and that institutional buyers tolerate higher prices and lower precision than consumer markets demand.

What This Means for Developers and Enterprises

For enterprise builders, Sooth Labs is an early signal that the next AI category after coding agents is structured forecasting: systems whose output is a probability distribution over future outcomes, not a chat completion. Expect a wave of competitors over the next 12 months pitching variations of the same idea — especially given how much of finance, insurance, and supply-chain software already runs on hand-built scenario models that an LLM-plus-signals stack might compress.

For developers, the practical near-term takeaway is narrower: open-source forecasting evaluation harnesses (Metaculus benchmarks, ForecastBench, the Good Judgment Open archive) are about to become the new SWE-bench — the standard datasets every model release cites. Watch this space.

What’s Next

Sooth Labs has not announced a public product launch date or pricing. The company’s sooth.inc landing page collects email addresses for early access and indicates a focus on institutional rather than consumer customers. Expect a more detailed technical paper, a benchmark publication, or a marquee design-partner announcement within the next two quarters as the team deploys its seed capital.

Sources

• Bloomberg: AI Pioneers Back Startup Building Models to Predict Events — primary breaking-news source by Rebecca Torrence.
• Techmeme aggregation of the Sooth Labs round — cross-referenced reporting and reactions.
• Sooth Labs official site (sooth.inc) — the company’s public positioning and email list.
• Ben’s Bites coverage — AI-industry newsletter framing.
• @Discoplomacy on X — early developer/researcher reaction thread.
• PitchBook profile of Sooth Labs — valuation and investor cross-reference.