There's something almost perverse about the timing. Six straight years of insured natural disaster losses topping $100 billion—2025 alone clocked in at $108 billion, according to Munich Re—and what does the reinsurance market do at the January 2026 renewals? It drops rates by low-teens percentages. Capital poured in, catastrophe bonds hit record issuance of over $25.6 billion, and suddenly buyers had the upper hand.
Yet scratch beneath the surface and you find a more uncomfortable truth: many of the models underpinning those pricing decisions run on architectures built decades ago, frameworks increasingly ill-suited to the frequency and ferocity of today's climate-driven losses. The math that was supposed to tame uncertainty is struggling to keep up with reality.
Which is why a new generation of AI-powered platforms is making its move now, into a market worth roughly $7 billion. They promise property-level granularity, continuously updated data feeds, synthetic catastrophe simulations that adapt as climate patterns shift. And the timing, perhaps more than the founders expected, matters. Regulators are circling: the National Association of Insurance Commissioners debated a third-party model vendor registration framework in March 2026, the EU AI Act kicked into general application in August 2026, California's SB 253 emissions disclosure deadline just passed on August 10. The industry's growing demand for transparency and the technology to deliver it are colliding at exactly the moment legacy vendors are racing to modernize—and challengers sense blood in the water.
When the Old Math Stops Working
For decades, a handful of firms have owned catastrophe modeling. Verisk and Moody's RMS command the lion's share of a market valued at approximately $7.2 billion in 2025, projected to reach $7.9 billion this year, according to commercial market research. These platforms—probabilistic engines that simulate thousands of potential hurricanes, earthquakes, wildfires, floods—power the pricing decisions of insurers, reinsurers, and increasingly financial institutions trying to manage climate-exposed portfolios.
Then came the losses. Munich Re pegged 2025 economic losses at $224 billion, with insured losses at $108 billion. But here's the kicker: severe convective storms alone accounted for approximately $56 billion in insured U.S. losses. Hail, tornadoes, straight-line winds—events that strike with little warning and alarming frequency, the kind of thing legacy models historically relegated to "other perils." Wildfire, too, dominated headlines throughout 2025. As Munich Re executive board member Thomas Grimm put it in a January 2026 interview, the industry now faces a reality where "non-peak perils are not secondary anymore."
The market's response defied conventional wisdom. Instead of panic, capital flooded in. Catastrophe bond issuance for the full year 2025 exceeded $25.6 billion, a record. By mid-2026, another $16.1 billion had already been issued, setting up what looks like another historic year. When roughly half of global property catastrophe reinsurance renewed on January 1, 2026, prices didn't harden—they softened, in some cases dramatically. Gallagher Re noted U.S. property-catastrophe price declines greater than expected; Aon reported stable attachment points and ample capacity.
Emboldened buyers didn't just pocket the savings. They started asking for something more: granular, transparent, climate-aware risk views that legacy vendors, constrained by their own architectures, found difficult to deliver on demand.
Three Forces Reshaping the Market
Technology, for one, has finally caught up to the problem. Machine learning and artificial intelligence are no longer experimental toys. NOAA's Earth Prediction Innovation Center released its GraphCast-based nested-EAGLE operational ensemble in May 2026, delivering high-resolution machine learning weather forecasts at spatial scales down to 3–6 kilometers for parts of the continental U.S. ICEYE, the synthetic aperture radar satellite operator, launched natural catastrophe monitoring solutions for the global banking sector that same month—building-level flood depth measurements, updated every six hours, plus event footprints for hurricanes, wildfires, earthquakes. A decade ago this was science fiction. Today it's table stakes.
The dominance of secondary perils represents a second shift. Severe convective storms lack the telegraphed intensity of a named hurricane making landfall; they just appear and wreak havoc. Wildfire, once an afterthought in catastrophe model documentation, now torches communities with a frequency and severity that decades-old event sets struggle to capture. The vendors are scrambling to catch up. Moody's RMS upgraded its severe convective storm models in 2025 and 2026. Verisk reengineered its U.S. Tropical Cyclone model and delivered it on its new Synergy Studio platform in June. JBA Risk Management launched an enhanced Global Flood Model with climate-conditioned event sets in March. Each announcement signals the same uncomfortable reality: the old catalogs no longer fit the moment.
Then there are the regulators. The NAIC's Third-Party Data and Models Working Group debated a vendor registration framework at the Spring 2026 National Meeting, raising thorny questions about scope, whether registration should be mandatory or voluntary, how much proprietary modeling logic insurers must document. The EU's AI Act, which entered general application on August 2, 2026, requires risk management systems, transparency, and logging for high-risk AI applications—implementation timelines for embedded AI systems in insurance stretch to December 2027 and August 2028. In the U.S., federal banking regulators released revised Model Risk Management guidance on April 17, 2026, emphasizing risk-based governance for AI and machine learning models at large institutions. And California's SB 253, which just hit its Scope 1 and 2 emissions reporting deadline on August 10, is pushing companies to quantify physical climate risk in ways that pull catastrophe analytics out of underwriting cubicles and into finance and disclosure workflows.
The Incumbents Strike Back
Legacy vendors aren't standing still. Verisk's June 2026 launch of Synergy Studio represents a calculated bet that cloud-based, modular platforms can match the flexibility younger competitors claim as their advantage. Simultaneously, the company expanded its Model Exchange to include more than 20 third-party providers and over 400 peril models—an acknowledgment, however reluctant, that the future might be vendor-neutral. In May, Verisk announced Model Context Protocol connectors to Anthropic's Claude, embedding its analytics directly into AI-assisted underwriter and actuary workflows.
Moody's RMS took a different route, integrating Oasis SaaS—an industry-owned, not-for-profit catastrophe modeling platform—into its Intelligent Risk Platform in February 2026. Oasis LMF, an open-source loss modeling framework, released version 2.5.0 and launched Oasis SaaS that same month, providing multi-vendor model hosting and fostering portability. It's a clever play: position yourself as an enabler of open modeling while preserving your proprietary model revenue stream. The firm also introduced SlipStream, an agentic AI tool for policy slip interpretation, in January 2026 and announced plans to integrate CAPE Analytics' property AI insights following Moody's acquisition of CAPE back in January 2025.
Specialized players, meanwhile, are carving out defensible niches. ZestyAI, which markets an AI-powered wildfire model, has claimed its Z-FIRE model serves insurers covering a substantial portion of the California homeowners market, though explicit regulatory acceptance dates in public records appear to be from 2024. Floodbase partnered with Liberty Mutual in March 2026 to launch instant parametric flood quoting for U.S. commercial risks—a distribution model that sidesteps traditional loss modeling entirely by triggering payouts based on observed flood depth rather than claimed losses. ICEYE's expansion into banking in May reflects a broader recognition: catastrophe risk has escaped the underwriting silo. Mortgage servicers and commercial real estate lenders need the same property-level intelligence.
Into this increasingly crowded landscape comes Risklytics, a two-person startup founded by Samuel Gold (currently on leave from Harvard) and Alexander Risio. Admitted to Y Combinator's Summer 2026 batch, the San Francisco-based company describes itself as an "AI-powered catastrophe risk modeling" platform combining continuously updated data, property-level analysis, and synthetic catastrophe simulations. The company's public footprint remains minimal—its Y Combinator profile is the primary source of information as of early July 2026, with no accessible company website or independent press coverage.
What Risklytics shares with other emerging challengers is a core thesis: that legacy models, however mathematically sophisticated, cannot keep pace with the velocity of climate-driven change and the granularity buyers now demand. Whether two founders and a YC badge can crack a market dominated by entrenched vendors with decades of regulatory relationships and institutional trust? That's the $7 billion question.
The Next 18 Months
Several catalysts are about to collide, and the result will likely determine whether this market can accommodate both incumbents and insurgents—or whether regulatory and technological pressure forces consolidation.
The NAIC's third-party model vendor framework, if adopted in 2026 or early 2027, could impose registration, documentation, and change management requirements that favor established vendors with existing compliance infrastructure. Or it could level the playing field by forcing all vendors, legacy and startup alike, to meet identical transparency standards. The EU AI Act's high-risk AI timelines, stretching through December 2027 and August 2028 for embedded systems, will require conformity assessments, technical documentation, human oversight. Vendors selling into European markets or to multinational institutions will bear those costs regardless of size.
Climate disclosure timelines add urgency. California's SB 253 Scope 1 and 2 reporting, which just passed its August 10, 2026 deadline, will push thousands of companies to quantify physical climate exposure—potentially pulling catastrophe analytics into CFO and investor relations workflows that have little patience for opaque models. The SEC's climate disclosure rule remains voluntarily stayed pending litigation, but the direction of travel is clear. Physical risk quantification is migrating from underwriting desks to boardrooms.
Market dynamics favor innovation, but only to a point. That $16.1 billion in catastrophe bonds issued in the first half of 2026 signals robust investor appetite for transparent, rules-based risk transfer. Parametric structures, which pay based on index triggers rather than modeled losses, are gaining traction precisely because they bypass model opacity. Insurers and reinsurers seeking differentiation will experiment with new platforms—but operational risk limits how quickly you can replace systems processing billions of dollars in annual premiums.
For founders and investors, the opportunity is real but narrow. The catastrophe modeling market isn't winner-take-all; buyers increasingly run multiple models in parallel to cross-validate results, which creates openings for newcomers. Buyers also prize speed: Verisk's six-hour flood depth updates, ICEYE's near-real-time SAR imagery, instant parametric quoting—all reflect an industry growing impatient with monthly model updates and quarterly event set refreshes. But the technical bar is punishingly high. Property-level accuracy, climate conditioning, regulatory defensibility. And the sales cycle is long. Insurance CIOs don't swap out catastrophe models lightly.
What seems certain is this: the $7 billion catastrophe modeling market of 2025 will look markedly different by decade's end. Whether that difference comes from incumbents successfully modernizing their platforms, from startups capturing meaningful share with fundamentally new architectures, or from regulatory mandates forcing convergence on open standards—the pressure is unmistakable. Six years of $100 billion-plus losses have a way of focusing attention. So do regulators demanding to see inside the black box. And so, perhaps most of all, does the creeping realization that the climate is changing faster than the models built to predict it.
