AI Weather Fundraising: GraphCast, Aurora, GenCast, Atmo

How Atmo, Salient Predictions, Excarta, Jua, Silurian, Brightband, Tomorrow.io, ClimateAI, Cervest, and AI-native weather.

Raising Capital for AI Weather, Climate, and Earth-Systems-Forecasting Startups

AI-native weather, earth-systems forecasting is the fastest-growing scientific-AI category since AlphaFold. Foundation models — Google DeepMind GraphCast (Science Nov-2023, 10-day global 0.25° in <60s on 1 TPU v4, outperformed ECMWF HRES on 90%+ of 1380 verification targets), Google DeepMind GenCast (Nature Dec-2024, first AI model to beat ECMWF ENS ensemble on 97.4% of 1320 targets, 8min for 15-day ensemble on 1 TPU v5), Microsoft Aurora (Nature May-2024 → production Jul-2025, 1.3B-parameter foundation model for atmosphere, ocean, air-quality, hurricane tracks, outperformed IFS-HRES on 91% of targets, Aurora-air-quality, Aurora-waves, Aurora-tropical-cyclones variants), Nvidia FourCastNet (Feb-2022, first AI-NWP at scale), FourCastNet v2, FourCastNet-SFNO, StormCast (regional km-scale Mar-2024), Earth-2 platform (Nov-2023, DGX Cloud, Omniverse, digital-twin climate simulation), Huawei Pangu-Weather (Nature Jul-2023, first AI to beat ECMWF HRES deterministic on all major variables), Fudan FengWu, FengWu-GHR, FengWu-4DVar (2023-2024, first AI-native data-assimilation), Tsinghua NowcastNet, Shanghai AI Lab Wan-Meteo, ECMWF AIFS (Jun-2024 operational, first NWP-center-native AI global model), UK Met Office, Alan Turing Institute FastNet (2024), Google MetNet-3 (Nov-2023, 24hr precip nowcasting outperformed HRRR, ERA5). Startups, commercial — Atmo ($60M+ total 2023-2024 Valor-Fusion Fund-Sam Altman-plus DoD SBIR, Philippines national-forecast contract Nov-2024, Cook Islands, Tuvalu sovereign contracts), Salient Predictions ($15M B Jun-2024 Emerson-First-Round-plus Union Square Ventures — sub-seasonal to seasonal S2S forecasts, MIT, Woods Hole spinout), Excarta ($4M seed Aug-2023 Kleiner-Wing, Y-Combinator — hyperlocal, energy-trader focus), Jua.ai ($16M seed May-2024 Green-Generation-Fund-plus 468 Capital — Swiss physics-informed foundation model, first European weather-AI unicorn candidate), Silurian AI (2024 stealth → $50M seed 2025 Radical-plus operator angels, ex-Google-DeepMind-GraphCast team), Brightband ($10M seed Feb-2024 Prelude-Radical — open-source AI-NWP, hurricane focus), WindBorne Systems ($15M A Oct-2024 Khosla-Footwork-Y-Combinator-plus Pear — long-endurance balloons, observation-data, AI-forecasts), Tomorrow.io ($400M+ total, incl. $87M E Apr-2022, Ontario-Teachers-JPM-Softbank — proprietary radar constellation, AI-NWP, enterprise forecasts), ClimateAI ($27M+ total, Blockchain-Capital-Tusk-plus Blue Bear — enterprise supply-chain, agriculture climate risk), Cervest ($30M A May-2021 Draper-Esprit-Astanor — asset-level climate risk), Sust Global ($9M seed Congruent-Yara), Ambiental (Royal Haskoning acq), Zesty.ai ($13M A Plug-Power-plus BitFinancial — wildfire, property risk), One Concern ($68M B May-2019 Softbank — earthquake, flood resilience), Kettle ($25M A Feb-2022 Acrew-Anthemis — wildfire reinsurance), Understory ($3.5M seed → 2024 growth — hail, parametric insurance), Reask ($10M A 2024 Aria-Group-Willis-Towers-Watson — tropical cyclone catastrophe modeling), Jupiter Intelligence ($97M D Aug-2022 MPower-Energy-Impact — physical-climate-risk for enterprise, insurance, wound down Dec-2024). Downstream, parametric-insurance, agri — Descartes Underwriting ($120M B Jan-2022 Highland-Europe-plus Serena — parametric climate), Arbol ($60M B Feb-2022 Opera-Tech-plus Mubadala — parametric crop-weather), Skyline Partners, Raincoat ($4.5M seed 2022 → 2024 growth — Caribbean, LATAM parametric), Floodbase ($12M A Mar-2024 Lowercarbon-Collab-plus Ubben — flood-parametric via SAR, AI), FloodMapp ($10M A Jun-2024 Down-Under-Ventures — real-time flood-mapping), Frontier Climate, Stripe Climate, Watershed, Sylvera, BeZero, Pachama, Persefoni (adjacent carbon, climate-risk data), Cervest, ClimateAI, Sust Global. Public sector, open data — NOAA (US $6B+ FY25 budget, $161M NCEP HPC, WoFS Warn-on-Forecast, HRRR, GFS, GEFS, AI-NWP research, NHC), ECMWF (Europe, ERA5 reanalysis, IFS, AIFS, ECMWF Open Data, Copernicus C3S), UK Met Office (£1B+ HPC contract with Microsoft Azure Apr-2024 for next-gen NWP, AI), JMA (Japan MSM, GSM), CMA (China, Pangu operational), BMKG (Indonesia), IMD (India), Météo-France (Arome, Arpege), DWD (Germany ICON), plus WMO Global Basic Observing Network (GBON), Systematic Observations Financing Facility (SOFF) $400M+ 2024-2027 for LDC observations.

Why 2026 is different

Four unlocks: (1) The GraphCast (Nov-2023), Pangu-Weather (Jul-2023), FourCastNet (Feb-2022), Aurora (May-2024 → Jul-2025 production), GenCast (Dec-2024), AIFS (Jun-2024 ECMWF operational) sequence proved that AI-NWP now beats ECMWF-HRES, IFS-ENS on 90-97% of verification targets at 1000-10000x lower inference cost — the entire $10B+ commercial weather industry now has to rebuild on AI foundations. (2) Sovereign, national-met procurement opened — Atmo's Philippines PAGASA, Cook Islands, Tuvalu contracts (Nov-2024), UK Met Office £1B+ Microsoft Azure Apr-2024 for next-gen AI-NWP, ECMWF AIFS operational Jun-2024, Anemoi open-source framework, Destination Earth €300M+, US NOAA AI-NWP research, DoD Atmo contracts, plus WMO SOFF, GBON $400M+ 2024-2027 for LDC observations set the government-procurement floor. (3) Parametric insurance, reinsurance crossed adoption — Descartes Underwriting, Arbol, Raincoat, Skyline, Understory, Floodbase, Reask, Kettle, Munich Re, Swiss Re, Hannover Re, Lloyd's parametric-triggers grew 50%+ YoY 2023-2024 as insured-loss gaps from Ian, Ida, Harvey, European floods, wildfires, LA fires Jan-2025 forced primary, reinsurance restructuring. (4) Observation, sensing consolidated — WindBorne balloons, Tomorrow.io radar constellation, Spire GNSS-RO, PlanetiQ, Muon Space, GHGSat, MethaneSAT, Carbon Mapper, Umbra, Capella, ICEYE SAR gave AI-NWP proprietary training, assimilation data beyond ERA5, GBON.

Realistic capital stack

Seed $3-15M for team, first published AI-NWP benchmark, first design-partner (Excarta $4M, Brightband $10M, Jua $16M, Salient $15M-then-B, Silurian $50M seed 2025 pattern). Series A $15-50M for first government, reinsurance, energy-trader contracts, expanded observation-partnerships (Cervest $30M A May-2021, Kettle $25M A Feb-2022, ClimateAI $27M cumulative, Floodbase $12M A Mar-2024, FloodMapp $10M A Jun-2024, WindBorne $15M A Oct-2024, Reask $10M A pattern). Series B $30-150M for scaled government, reinsurance, international, platform (Salient $15M B Jun-2024, Descartes $120M B Jan-2022, Arbol $60M B Feb-2022, Tomorrow.io $87M E Apr-2022, One Concern $68M B May-2019, Jupiter $97M D Aug-2022-then-wound-down). Series C+ $100M-$500M for global, sovereign, platform, M&A (Tomorrow.io $400M+ cumulative, Descartes, Arbol, One Concern late-stage). Non-dilutive: NOAA SBIR, NCEP HPC access, DoD SBIR, DIU, AFWERX (Atmo Philippines pattern), DARPA AI-Next, weather-modeling, NASA Earth Science, CSDA, Earth System Digital Twin, NSF NCAR, LEAP, AI2ES, ECMWF Destination Earth, Copernicus C3S/CAMS/CEMS, ESA Φ-lab, Space for Green Future, UK Met Office, Alan Turing, Innovate UK, Météo-France, CNRS, DFG, BMBF (Germany), JAXA, JMA (Japan), CSIRO, BOM (Australia), WMO SOFF, GBON $400M+ 2024-2027, plus Munich Re, Swiss Re, Hannover Re, Lloyd's Lab, AXA XL corporate-venture programs.

Common failure modes

Building an AI-NWP without published skill benchmarks vs. ECMWF-HRES, IFS-ENS, NOAA-GFS — government, reinsurance procurement will not evaluate without WMO-standard verification (RMSE, ACC, CRPS on Z500, T850, T2m, U/V10, precip at 1-10 day, plus per-basin hurricane skill). Ignoring the ECMWF Anemoi open-source framework (Jun-2024) — Anemoi, Destination Earth, AIFS created the reference stack, and vendors that don't interoperate get boxed out of European government procurement. Building without proprietary observation-data or without ERA5, IFS, GFS, regional-obs fine-tuning — foundation-model quality caps at the training-data ceiling, and winners either partner with WindBorne, Tomorrow.io, Spire, Planet, Umbra, Capella or ingest full ERA5, IFS, GFS. Depending on hyperscaler-only compute without a cost-optimized inference path — GraphCast, GenCast, Aurora inference on H100/H200/GB200/TPU v5/v6 must be paired with FP8/INT8 quantization, Cerebras/Groq/SambaNova offload, distillation for enterprise unit-economics. Ignoring aviation, FAA, EASA, ICAO safety-critical certification for aviation-weather (SIGMET, AIRMET, TAF, METAR) — commercial aviation is a $2B+ TAM but requires DO-178C-adjacent, safety-case documentation. Ignoring the EU AI Act Art. 6 high-risk classification for safety-critical forecasts (aviation, disaster, energy-grid), Art. 50 transparency for public-facing forecasts. Building parametric-insurance without reinsurance backing — Munich Re, Swiss Re, Hannover Re, SCOR, Lloyd's syndicate capacity is the gating factor, not primary demand. Ignoring the Jupiter Intelligence wind-down (Dec-2024) — climate-risk-data-only businesses without a reinsurance, trading, government anchor lack durable ACV.

Frequently asked questions

Is there room for another AI-NWP foundation-model startup vs. GraphCast, GenCast, Aurora, AIFS?
Narrowing on horizontal, global, expanding on regional, specialized. GraphCast, GenCast (Google DeepMind), Aurora (Microsoft), AIFS (ECMWF operational), Pangu (Huawei), FengWu (Fudan, Shanghai AI Lab) now cover 0.25° global 1-15 day deterministic, probabilistic. Defensible wedges are (a) km-scale regional, convection-permitting (Nvidia StormCast, Brightband hurricane, Google MetNet-3 nowcasting, UK Met FastNet), (b) sub-seasonal to seasonal S2S (Salient Predictions, ClimateAI), (c) sovereign, national-met turnkey (Atmo Philippines, Cook Islands, Tuvalu pattern for LDCs, SIDS), (d) verticalized enterprise (Excarta energy-trader, Jua energy, agri, ClimateAI supply-chain, Cervest asset-level), (e) proprietary observation, assimilation (WindBorne balloons, Tomorrow.io radar constellation, Spire GNSS-RO, PlanetiQ, Muon Space), (f) safety-critical aviation, disaster, energy-grid with DO-178C-adjacent certification, (g) parametric, reinsurance, cat-modeling (Descartes, Arbol, Raincoat, Understory, Floodbase, Reask, Kettle).
How exposed is a weather-AI startup to hyperscaler, national-met in-house AI?
Meaningfully at the horizontal foundation-model layer (GraphCast, GenCast, Aurora, AIFS all published openly or productized by Google, Microsoft, ECMWF). Defensible wedges are (a) sovereign, national-met turnkey deployment for governments without in-house AI, HPC (Atmo Philippines, PAGASA, Cook Islands, Tuvalu is the reference), (b) reinsurance, parametric-insurance, cat-modeling with Munich Re, Swiss Re, Lloyd's capacity, (c) enterprise, energy-trader, agri, supply-chain forecasts with proprietary observations (WindBorne, Tomorrow.io, Spire, PlanetiQ), (d) km-scale, convection-permitting, hurricane, wildfire regional models, (e) safety-critical aviation, disaster, energy-grid certification.
Realistic exit?
Strategic acquisition by weather, geospatial, climate-data incumbents (DTN, AccuWeather, Baron Weather, Vaisala, Pelmorex, Weather Group, Kayrros, Enverus, ICE, Refinitiv-LSEG, Bloomberg), by hyperscalers for weather-AI talent, platform (Microsoft-Nuance-legacy pattern, Google DeepMind-in-house, Amazon-AWS-Earth-adjacent), by reinsurance, insurance incumbents (Munich Re, Swiss Re, Hannover Re, SCOR, Berkshire, Lloyd's, AXA, Allianz, Zurich, Chubb, Sompo, MS&AD, Tokio Marine potential parametric-platform acquirers), or by defense, intel primes (Palantir, Booz Allen, Leidos, SAIC, L3Harris, Northrop, Raytheon, Lockheed, Anduril for defense-weather, intel-weather). IPO reserved for two or three category leaders at $200M+ ARR — Tomorrow.io, Atmo, Silurian, Jua, and Descartes Underwriting are current candidates on a 3-5 year horizon, alongside geospatial adjacents Planet Labs $PL, BlackSky $BKSY, Spire $SPIR, Satellogic $SATL.

Related fundraising verticals (40)

Investor directory · Fundraising library · Articles A–Z · Company funding database