It’s AI Versus The Weather Now

WindBorne Systems is chasing the once mythical 21-day forecast

As Hurricane Ian spiraled off the coast of Florida in September 2022, an object resembling an errant plastic bag drifted through its eye, cavorting to the beat of convection physics. This plastic bag—also known as a Global Sounding Balloon, or GSB—was on a one-way mission to gather meteorological data as it surfed along the 80 mph winds.

Over 2,000 miles away, a Harvard-trained physicist named Dr. Andrey Sushko piloted this contraption from inside a mission control (and climate-controlled) suite at the Palo Alto headquarters of WindBorne Systems, a company he co-founded in 2019.

Though typically run by an automated program, this GSB was on an atypical path and required the nimble fingers of a pilot. WindBorne’s team worked in shifts to steer the GSB from its launch site in Gainesville, Florida into the hurricane's orbit. When it reached the western eyewall 50 hours later, Sushko and his crew were locked in.

They typed commands to release pinches of sand from the ballast bag hanging beneath the balloon, lightening the GSB so it could rise above the hurricane’s eyewall. They then issued another order to vent helium with a slow hiss and plunge the GSB into its eye.

Real-time moisture and wind speed measurements from the GSB let Sushko and his co-founders—John Dean, Kai Marshland, and Joan Creus-Costa—know that it had entered the hurricane’s hollow heart. Despite the frenzied journey, Sushko, the company’s CTO, says that the conditions weren’t horrible: “Flying inside a thunderstorm in Oklahoma is arguably worse in terms of altitude control.”

Satellites relayed the data gathered from the GSB to the company’s principal product: an AI-forecasting model called WeatherMesh. This balloon-to-software pipeline might just be the first actual cloud-to-cloud service, and it is WindBorne’s edge in the latest high-tech race to answer one of civilization’s oldest questions: what’s the weather today?

The cost of an incomplete answer is steep. On July 4th, severe rainfall abruptly turned into major floods in Central Texas, which have so far claimed over 130 lives. Amid the grief, questions circulate: Why were key roles at the National Weather Service left vacant? Why did the area lack adequate warning systems? What can be done to prevent this from happening again?

In recent years, the weather forecast has found itself caught between an increasingly unpredictable climate, an alert-fatigued public, and now DOGE’s decision to gut the federal agency responsible for distributing life-saving warnings. Together, these three factors form a Bermuda Triangle of sorts, where reliable forecasts vanish.

Extreme weather events accounted for 2,520 deaths in the US between 2020 and 2024, and the country spent nearly $1 trillion on disaster recovery in a single 12-month period (ending in May 2025). Those already high figures are only expected to rise.

But where some spot a crisis, others view an opportunity. Tech giants like Google, Microsoft, and NVIDIA are eager to put their AI forecasting models to use, with Google DeepMind leading the charge this summer after they announced a partnership with the National Hurricane Center. WindBorne is in tow with a fraction of the funding. The humble weather forecast is gearing up for perhaps its biggest transformation since the invention of the computer.