Atmo
Deep-learning global weather models, claimed far cheaper to run than legacy supercomputers — acquired by Vaisala in 2026.
Acquired — still operating
Acquired by Vaisala in June 2026 for $70M plus up to $60M earnout; folded into the Xweather business.
About
Atmo was a Berkeley AI-weather startup that tried to out-model the supercomputers. Co-founded by CEO Alex Levy and CTO Johan Mathe, it trained deep-learning models on more than 30,000 terabytes of global weather data and claimed its global forecasts ran over 10,000 times more efficiently than legacy numerical models — while matching or beating them in accuracy, starting with a Singapore demonstration and later work with the US Air Force.
In June 2026 the Finnish weather-instrument giant Vaisala acquired Atmo for $70M in cash and shares, plus up to $60M in earnouts tied to 2026–2028 performance, folding the team into its Xweather business. At the time Atmo reported 2025 net sales of just $2M — an R&D-stage company bought for its models, not its revenue. Its public forecasting tools live on under Vaisala.
What it can do
- Deep-learning global forecast models
- Vectorized, near-instant 3D weather rendering
- Hyperlocal forecast variants
- Self-learning model updates
- Defense and government deployments
Setup
Products are now delivered through Vaisala's Xweather business — contact Vaisala for access.
Prerequisites
Enterprise or government engagement via Vaisala.
Use cases
Works with
Trust score — 5/10
Scored 5/10 (TIER metrics, metrics-v1). Evidence: Vaisala's June 2026 acquisition ($70M + up to $60M earnout, $2M 2025 sales) documented in Vaisala's own inside-information release; founders and technical approach verified via Freethink's independent profile. Strengths: strong AI-meteorology team now backed by a 150-country sensor network, real government deployments. Docked for: (1) '10,000x more efficient, same accuracy' claims are provider-stated with no independent benchmark; (2) company no longer exists independently — roadmap now follows Vaisala; (3) tiny revenue base means the product was pre-scale. License: proprietary; now part of Vaisala Xweather commercial offerings.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 5.0/10
- Maintenance · recency, releases, issue load
- 6.0/10
- Openness · license + docs
- 3.0/10
- Manual review · sanity read + red-flag check
- 6.0/10
- · Berkeley AI-weather startup, co-founders Alex Levy (CEO) and Johan Mathe (CTO)
- · Trained deep-learning models on 30,000+ TB of global weather data
- · Worked with the US Air Force and governments; demoed in Singapore, Uganda, France, Germany
- · Acquired by Vaisala June 2026 for $70M + up to $60M earnout; 2025 net sales just $2M
- · '10,000x more efficient, same accuracy' claims are provider-stated with no independent benchmark
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.