WindBorne Systems has raised $37 million to sell something it has already proven it can build: weather forecasts that beat the government’s. The Series B, reported by TechCrunch, was co-led by Khosla Ventures and Galvanize and values the company at $250 million. It takes WindBorne’s total funding past $62 million.

The forecasting is the easy part now. Turning it into a business is the hard one.

The planetary nervous system

WindBorne flies balloons. About 600 of them are in the air at any moment, launched from 20 sites worldwide. They drift for weeks, gathering readings where satellites and weather stations cannot. The eye of a typhoon is one such place. The company calls the network a planetary nervous system, and it is the point of the whole business.

That data feeds WeatherMesh, WindBorne’s AI forecasting model. The latest version, released two months ago, tops public benchmarks as the most accurate weather model available. It beats the leading AI systems and the traditional physics-based models that governments run. It does so on far less computing power to train.

The moat is the data, not the model. Anyone can train a forecasting AI on public datasets. WindBorne also owns readings nobody else has. “The value per data point is much stronger than satellites,” chief executive John Dean told TechCrunch. Growing revenue while proving it, he added, “de-risked the demand signal to VCs.”

Sold to the government, betting on business

For now, WindBorne’s customers are mostly public. The US National Weather Service buys its data. The Air Force and Navy pay through research partnerships, including work on forecasting models that can run onboard ships with patchy connections. Its readings are already folded into NOAA’s Global Forecast System, one of the backbones of everyday weather apps.

The growth bet points the other way, at the private sector. Trading firms and investment funds were the earliest commercial takers, using sharper forecasts to price commodities and weather-sensitive markets. Expanding that base is what the new money is for. The round also funds more compute and a mesh radio network to replace the balloons’ satellite links. It even pays for ocean buoys that keep sensing after they splash down.

Why this is the hard part

Selling sensing data to businesses has broken startups before. Earth-observing satellite networks spent a decade trying and mostly retreated to government buyers. Extracting value from raw data takes workflows and expertise most companies lack. Private weather firms have survived by repackaging government forecasts, de-icing planes, routing ships, and serving speculators.

WindBorne’s bet is that AI changes the maths. “Integrating weather forecasts into broader business decision-making has traditionally been expensive and difficult,” said Galvanize’s Saloni Multani, who co-led the round. Better forecasts make the effort worthwhile, she argued, and AI makes wiring them into decisions far easier. “We think AI changes that equation.”

The backers are betting on the pairing. Khosla Ventures, fresh from co-leading Jeff Dean’s new AI venture, has held its stake since the first cheque. “As AI models become increasingly capable, the quality of the underlying observations becomes even more important,” said Khosla’s Sven Strohband. WindBorne also fits a wider pattern: investors pouring capital into AI bound to the physical world, from autonomous mining to fusion power.

The demand is not in doubt. Extreme weather is getting more frequent and more expensive, and the appetite for early warning is growing on both sides of the public-private line. The question WindBorne has to answer is the one every sensing startup before it hit. Whether better data, this time wrapped in AI, is finally something enough customers will pay for.

Get the TNW newsletter

Get the most important tech news in your inbox each week.