The people who built Facebook's feed are now building AI to predict the future, and they just raised $50 million to do it.
The Summary
- Sooth Labs, founded by ex-Meta engineers, is raising ~$50 million to build AI models that forecast geopolitical and market events
- This isn't content recommendation. It's probabilistic forecasting for businesses trying to price risk in real time.
- The founding team includes AI pioneers who built Meta's recommendation systems, now applying those architectures to predict real-world outcomes instead of clicks.
The Signal
Sooth Labs is betting that the same transformer architectures that learned to predict what video you'd watch next can learn to predict whether a country will default on its debt. The founding team spent years at Meta training models on billions of user interactions. Now they're training models on geopolitical data, market signals, satellite imagery, and anything else that might contain predictive information about future events.
The pitch is simple: businesses make billion-dollar decisions based on gut feelings about what might happen next. Will sanctions hold? Will a trade route stay open? Will a currency collapse? Right now, most companies pay analysts to read reports and make educated guesses. Sooth wants to replace that with probabilistic forecasts updated in real time.
"The same architectures that learned human behavior patterns can learn event patterns in complex systems."
Here's what makes this different from existing forecasting tools:
- Training on multimodal data: not just financial time series, but news, satellite feeds, social signals, trade data
- Real-time updating: models adjust probabilities as new information arrives, not monthly reports
- Specific event targeting: "Will X happen by Y date" instead of broad trend analysis
The $50 million round suggests serious conviction from investors who've seen these architectures work at scale. Meta's recommendation systems process trillions of signals to predict individual user behavior. The bet here is that geopolitical and market events follow patterns just as learnable as human clicking patterns, just with different training data.
The Implication
If this works, it changes how companies price risk and make strategic bets. Insurance, logistics, finance, any business exposed to geopolitical volatility suddenly has a new tool that updates faster than human analysts can read. The models won't replace judgment, but they'll make it harder to ignore probabilities when your AI is screaming "30% chance of disruption in the next 60 days."
Watch for early customers in reinsurance and commodities trading. Those are industries already comfortable buying predictions and have clear ways to validate accuracy. If Sooth's models beat human forecasters consistently, expect every Fortune 500 risk team to have a subscription by 2027.