In the first half of 2025, climate-related disasters caused an estimated $100 billion in global economic losses, a significant increase from the previous year. The United States alone accounted for a substantial portion of this total, with events like the LA wildfires marking the costliest first half on record.

As extreme events grow in frequency and severity, a pressing question arises: are we investing enough in forecasting and risk assessment to keep pace with their escalating impacts? While traditional physics-based models are still considered the gold standard, artificial intelligence is fundamentally reshaping how we understand, predict, and prepare for climate risks. Over the past several years, dozens of AI-powered tools have emerged, each promising faster, more accurate insights into the threats communities face. But a troubling divide is forming: while some platforms democratize access to life-saving data, others lock essential climate intelligence behind expensive paywalls.

I work with a team of researchers and scientists trying to address precisely this challenge. Our project, ClimateIQ, is an AI-driven climate hazard assessment tool aimed at providing communities the information they need to plan and prepare effectively. But significant work remains, and considerable uncertainty surrounds how these new AI tools are being developed, including their methods, their approaches, and their accessibility.

In this article, I explore recent advances in using AI to improve weather forecasting and hazard assessment. Who is building these tools, what methods are they using and who can access them? Most importantly, I make the case for why democratizing climate intelligence matters: it saves lives, protects economies and livelihoods, and ultimately saves money. Understanding the landscape of AI climate tools, and the gaps that remain, is the first step toward ensuring everyone has access to the information they need to face our changing climate.

How did we get here?

The story of modern weather forecasting begins in the realm of physics and computation. In 1922, Lewis Fry Richardson imagined a “forecast factory” of human calculators solving the primitive equations of atmospheric motion. It would take until the 1950s, with the advent of early digital computers like ENIAC, for numerical weather prediction (NWP) to take its first credible steps.

“Weather Forecasting Factory” by Stephen Conlin, 1986.

NWP relies on solving a set of partial differential equations, momentum, thermodynamics, mass continuity, and state, that describe the fluid dynamics of the atmosphere. These equations are computationally expensive and sensitive to initial conditions, but they marked a profound epistemological shift: from observational generalization to simulation. Throughout the late 20th century, global modeling systems like the European Centre for Medium-Range Weather Forecasts (ECMWF) and the U.S. Global Forecast System (GFS) emerged as major players. These models improved over decades via better physics parameterizations, satellite data assimilation, and increased computational resolution, from ~200 km grids in the 1980s to ~10 km or finer today.

Newspaper article dated August 11, 1929, when the weather station was located in the Stahlman Building.

But even as model skill improved, key challenges persisted: limitations in initial observations (especially over oceans and the Global South), biases in parameterizations (like how clouds and land surface heat are represented), and the ever-present chaos of the atmosphere itself. Weather, after all, is not merely physics, it is fluid physics in an open system, riddled with feedback and uncertainties.

The New Weather Machines: AI and Machine Learning in Forecasting

The past five years have seen an inflection point toward higher-resolution simulations. Companies and research labs are increasingly building models that learn the weather, rather than simulate it from first principles.

Deep learning architectures like GraphCast (DeepMind) and FourCastNet (NVIDIA) are trained on vast atmospheric reanalysis datasets, decades of labeled weather data, and can now predict global atmospheric fields (temperature, pressure, wind) with accuracy rivaling some traditional NWP systems.

What’s new is their speed and cost: GraphCast can produce forecasts in seconds, where conventional models might require hours on supercomputers that are expensive to utilize. And this matters: for real-time decision-making in hazards, agriculture, energy, and emergency response, latency is often more critical than perfection.

While AI approaches have potential to improve short-term weather forecasts, it offers limited value for long-term resilience planning, which requires multi-hazard scenario assessments. What’s more, these AI models are data-intensive, sometimes brittle when pushed outside their training distributions, and often lack physical consistency. Their internal logic is also largely opaque, making it difficult to replicate information provided by the models.

The future in many ways is about hybridization–combining the generalizability of physics with the pattern-recognition power of machine learning. This is where tools like physics-informed neural networks and data assimilation via AI are beginning to change this for researchers. As a result the market for such tools is projected to keep expanding as organizations across sectors recognize that climate risk is financial risk. 

The Market Widens

One of the early innovators in this space was Fathom, a UK-based company founded by academic hydrologists to provide high-resolution flood risk mapping. By combining terrain and hydrodynamic models with global datasets, Fathom produced granular, building-level flood intelligence. Its tools quickly became useful for insurers and development banks seeking to assess water risk under current and future climate conditions. In 2023, Fathom was acquired by Swiss Re, one of the world’s largest reinsurers, an indication that flood risk modeling had moved from a niche academic pursuit into a critical tool for the insurance sector.

While Fathom specialized in hydrology, Jupiter Intelligence expanded the scope to additional hazards. Jupiter’s platform, ClimateScore™ Global, combines downscaled climate projections, socio-economic exposure data, and physical hazard modeling to assess future risks from extreme heat, flooding, wind, and sea level rise. Jupiter’s clients include municipalities, real estate developers, and financial institutions. Their platform transforms climate data into market insights, allowing portfolio managers to understand and mitigate long-term exposure to climate volatility. 

At the same time, other players like AlphaGeo and First Street Foundation focused on some accessibility (often limited by geography) and scale. AlphaGeo builds AI-driven risk maps and climate adaptation scores by fusing remote sensing, demographic indicators, and historical climate data. Their tools serve infrastructure planning, ESG investing, and public sector resilience work. Meanwhile, First Street Foundation has mapped flood, heat, fire, and wind risk for millions of U.S. properties and integrated that data into publicly accessible real estate platforms (although recently removed from Zillow). Their mission is to provide some information on climate risk (at a price), making it visible and understandable for everyday decision-makers, homeowners, local officials, and renters alike.

On the technical frontier, firms like Google DeepMind, NVIDIA, and Planette.ai are rethinking approaches to weather forecasting. DeepMind’s GraphCast and NVIDIA’s FourCastNet use deep learning architectures trained on decades of atmospheric reanalysis data to produce global weather forecasts that rival traditional models in accuracy, but generate results in seconds instead of hours. These models don’t simulate physics explicitly, they learn patterns from vast datasets and predict how the atmosphere will evolve, often with some great precision.

Planette.ai, a newer entrant, positions itself in the “long-range forecasting” space, aiming to predict weather from the next week to the next year using quantum-inspired AI. Their approach seeks to fill the gap between short-term forecasts and long-term climate projections, an area historically fraught with uncertainty and computational difficulty.

Throughout all this, public infrastructure remains vital. The WMO sets global standards for data sharing and model interoperability. NOAA and NASA continue to run physics-based weather and climate models, operate satellites, and provide foundational datasets that power both public and private forecasting systems. However, their platforms often lack the user-centric design and localization that newer companies prioritize. As a result, many commercial actors build directly atop public data, adding proprietary models, APIs, risk scoring tools, and visualization platforms.

These new approaches to climate intelligence raise important questions. Forecasts are increasingly delivered not as maps or time series, but as decisions: a risk score, a zoning recommendation, or a building code update. As private platforms become intermediaries between raw data and public understanding, issues of transparency, equity, and governance emerge. Who owns and has access to the model? Who understands its assumptions? What happens when risk models determine access to mortgages, insurance, or infrastructure funding? And ultimately who can afford to access this increasingly important information?

These are the questions that ClimateIQ is trying to answer.

The Accessibility Issue

While many of these tools show great promise, premium platforms often charge enterprise fees that may exclude those facing the greatest vulnerability. A coastal town in the Global South cannot afford $100,000 worth of subscriptions. Nonprofits lack budgets for commercial assessments, and many mid-sized companies struggle to justify costs.

This creates a climate intelligence gap. Well-resourced organizations make data-driven decisions about asset relocation and infrastructure investments. Under-resourced communities operate blind, unable to quantify risks or prioritize adaptation measures.

Without high resolution risk data, cities cannot target resilience investments effectively. Farmers cannot access fairly-priced crop insurance. Emergency managers cannot pre-position resources before disasters strike. The consequences have and will be dramatic for many of the under-resourced areas across the world.

Enter ClimateIQ: Bridging the Gap

Against this backdrop, we developed ClimateIQ with an ambitious mission: become the world's first truly open-access platform delivering climate hazard intelligence at scale to cities and communities where most people live.

Learning from the sophistication of commercial platforms and the accessibility of open-source tools, ClimateIQ aims to deliver robust science and AI-enhanced hazard prediction, validated against ground truth, through intuitive interfaces without cost barriers.

ClimateIQ recognizes that transformative climate adaptation requires democratizing access. When only wealthy institutions afford risk intelligence, we guarantee maladaptation: resources protect privileged assets while vulnerable populations remain exposed. Open access flips this equation.

By synthesizing methodologies from leaders in the field, while committing to open access models, ClimateIQ charts a different path. One where a mayor in Cape Coast, Ghana has the same climate intelligence as a Wall Street portfolio manager. Where community organizers quantify flood risk as precisely as insurance actuaries. Where adaptation investments are guided by data, not outdated information.

It’s important to remember that climate change knows no borders and will increasingly shape our lives in ways we cannot imagine today. Whether the most advanced tools serve only those who can afford them, or become true global public goods that support adaptation everywhere, depends on the choices we make today.

ClimateIQ’s answer is clear: climate intelligence must be a right, not a privilege. Only with equitable access can we build the resilient and prepared communities that our rising risks demand.