Tomorrow.io
Tomorrow.io: Redefining weather forecasting
A storm is forming offshore, beyond the reach of ground-based radar. In traditional forecasting, the first useful observation may not arrive for hours—time when conditions can change fast.
Boston-based weather intelligence company Tomorrow.io is closing that gap by combining proprietary space-based observations with rapid-update physics and AI forecasting, helping industries and governments make time-critical decisions.
To do this, the company built its own microwave sounder constellation to capture frequent, timely atmospheric observations—especially where conventional sensing is limited. Those observations feed Tomorrow.io’s Resilience Platform and its physics- and AI-based models, delivering hyperlocal forecasts across 15+ industries and to meteorological services and governments worldwide. The platform runs on Microsoft Azure with NVIDIA accelerated computing to meet global-scale processing demands, reducing the time from observation to actionable forecast.
The hard-fought science of prediction
Atmospheric scientist and Director of Data Science at Tomorrow.io, Dr. Forest Cannon is pushing the limits of weather prediction. While AI models have only recently begun to outperform the physics-based approaches that led the field for ~70 years, performance still varies by variable, geography, lead time, and the quality and freshness of incoming observations.
Tomorrow.io’s response was to build its own AI forecasting capabilities designed for rapid, observation-driven updates. The company developed its FOCUS model architecture in partnership with NVIDIA and Microsoft. NIVIDIA’s Earth-2 platform and PhysicsNeMo framework, tailored for specific customer use cases as well as CorrDiff model, which learns complex atmospheric physics by training on large volumes of observational data we essential to FOCUS.
FOCUS is designed to help address practical real-time forecasting challenges, including initializing from diverse observations and representing uncertainty in rapidly evolving storms. Rather than presenting a single deterministic answer, the system generates probabilistic guidance through ensembles, helping users make decisions when timing and confidence matter as much as the central forecast. The platform lets Cannon and his team grab containerized applications, modify them, deploy them into training, and iterate, all without needing to create the underlying infrastructure.
“NVIDIA Earth-2 on Azure has been fundamental to our development,” Cannon says. “At my core, I’m an atmospheric scientist. I’m not a software engineer.”
The first operational FOCUS deployment for the contiguous United States runs on Microsoft Azure using NVIDIA H100 infrastructure. It updates every 15 minutes at a 13-kilometer resolution and produces guidance for variables including precipitation, near-surface temperature and humidity, winds, gusts, visibility, cloud cover, and severe weather indicators such as hail and lightning. At interference time, FOCUS integrates larger-scale model guidance with observation driven analyses that combine stations, radar, geostationary satellites and the company’s microwave sounder constellation.
Since that initial deployment, FOCUS has expanded well beyond a single U.S. domain, with regional deployments across the globe. In practice, it has become a flexible operational framework for downscaling larger-scale guidance and combining it with locally available observations to support the weather decisions customers face every day.
That observation emphasis matters because severe weather forecasting is not just about sharper maps; it is also about better framing uncertainty. In severe weather applications, FOCUS is designed to provide more precise risk guidance, helping users make faster, better-informed decisions in rapidly evolving conditions.
A constellation born from necessity
Tomorrow.io is a very different company than it set out to be, “We were actually a software-only business,” says CEO and co-founder Shimon Elkabetz. “And then I think very early we understood that the holy grail is really the lack of data.”
Tomorrow.io’s first constellation, completed January 2026, and designed to fill the global data gap from space, consists of microwave sounders achieving a 60-minute revisit rate for every point on earth.
Developing a satellite once took 10 to 15 years; today, a single-instrument satellite can be developed and launched in a couple of years, then replaced as technology improves. As a result of the Tomorrow.io Microwave Sounder (TMS) constellation, Tomorrow.io is able to observe every point on the globe every hour and feed that data directly into active forecasts. As more satellites come online, hyperlocal accuracy will extend to regions that have long been underserved.
Together with Microsoft and NVIDIA, Elkabetz says, Tomorrow.io is excited to continue working with some of the world’s most transformative companies and global meteorological agencies to advance a more intelligent weather operating system.
Weather intelligence for every citizen
The decisions Tomorrow.io’s platform supports saving lives, reducing accidents, improve worker safety, and protect operations that can’t afford to be caught off guard. That mission is scaled from a utility company pre-positioning repair crews to a government issuing evacuation orders ahead of a storm that no ground radar has yet seen. In Elkabetz’s framing, it is work that touches everyone.
“We’re doing something that every citizen in the world is going to benefit from and is already benefiting from,” he says. “It’s a huge privilege.”