Whether you're looking for an angel investor, a growth advisor, or just want to connect — I'm always open to great ideas.
Get in TouchGoogle deploys FireSat, a constellation of over 50 AI-powered satellites designed to detect wildfires within minutes of ignition — combining thermal imaging with machine learning to protect communities from increasingly devastating fire seasons.
AI, startups & growth insights. No spam.
When the Palisades Fire tore through Los Angeles in January 2026, it destroyed more than 12,000 structures and forced the evacuation of over 100,000 residents. By the time firefighting aircraft reached the scene in force, the fire had already consumed thousands of acres. The gap between ignition and response — often measured in hours — is not a failure of human courage or equipment. It is a failure of information. Firefighters cannot fight what they cannot find, and for decades, finding a new wildfire quickly enough to contain it has been nearly impossible.
Google's FireSat program is designed to close that gap. Announced as a fully operational constellation in March 2026, FireSat deploys more than 50 AI-powered satellites in low Earth orbit, each equipped with thermal infrared sensors and on-board machine learning processors capable of detecting a fire roughly the size of a classroom — approximately 20 square meters — within minutes of ignition. The system promises to transform wildfire response from a reactive scramble into something approaching real-time situational awareness for fire managers, utilities, and emergency services worldwide.
FireSat is not a single satellite. It is a coordinated constellation — a network of small satellites spaced across orbital planes so that any point on Earth's surface passes under at least one sensor every few minutes. Google developed the underlying AI detection algorithms and partnered with satellite operators, including Earth-observation startup Muon Space, which manufactures and operates the hardware. Muon Space's satellites are purpose-built for the mission: compact, low-cost spacecraft carrying custom thermal infrared cameras optimized for the wavelengths at which vegetation and ground fires emit heat.
The constellation architecture is what makes the detection speed possible. Traditional weather satellites like GOES-West can image the continental United States continuously, but their sensors resolve objects at roughly 2 kilometers per pixel — too coarse to catch a small ignition point. Dedicated Earth-observation satellites like those from Maxar or Planet Labs offer high resolution, but they revisit any specific location only once or twice per day. FireSat's design threads the needle: moderate resolution (precise enough to detect a 20-square-meter fire) combined with high revisit frequency (minutes rather than hours or days).
Google funded the initiative through its philanthropic arm and through the broader Google.org climate portfolio, reflecting the company's framing of FireSat as a public-good infrastructure project rather than a commercial product in the traditional sense. Data from the constellation is intended to flow to fire agencies, utilities managing transmission infrastructure, and eventually to a broader network of emergency response tools.
The physics of wildfire detection from space come down to heat and wavelength. A burning fire — even a small one — emits significant energy in the mid-wave infrared (MWIR) band, roughly 3 to 5 micrometers. This wavelength penetrates smoke more effectively than visible light and is distinct from the thermal signature of sun-warmed soil or pavement, which tends to emit at longer wavelengths (8 to 14 micrometers). FireSat's cameras are tuned to the MWIR band, allowing them to distinguish an active flame from the thermal noise of a hot summer day.
But capturing the right wavelength is only the first step. A raw thermal image from orbit contains thousands of pixels, each representing a patch of ground, and distinguishing a genuine fire signal from false positives — a sun-glint off a metal roof, a controlled agricultural burn, an industrial facility running at high temperature — requires intelligence that goes beyond simple threshold detection.
This is where the machine learning layer becomes essential. Google trained FireSat's detection models on years of historical satellite imagery, cross-referenced with ground-truth fire records, weather data, vegetation maps, and industrial facility locations. The model learns to weight contextual signals: a thermal spike in a remote forested canyon during a red-flag wind event looks very different from the same temperature reading over a registered agricultural burn zone. By processing this context on-board or near-real-time in ground processing pipelines, FireSat can issue alerts within minutes of a detection event rather than waiting for a human analyst to review flagged imagery.
The on-board ML component is particularly significant. Rather than transmitting raw image data to ground stations and processing it there — which introduces latency whenever a satellite is not in contact with a ground station — FireSat satellites can perform initial detection inference directly in orbit. Only a compressed alert packet, along with the relevant image tile, needs to be downlinked immediately. This architecture mirrors the distributed edge inference approaches that companies like NVIDIA have pioneered for terrestrial applications — bringing computation closer to the data source to eliminate round-trip latency. The parallels to NVIDIA's distributed edge inference frameworks are direct: the principle of minimizing data movement by doing inference at the source applies whether the source is a factory floor sensor or a satellite thermal camera 550 kilometers above the ground.
Wildfire behavior follows exponential dynamics. A fire that covers one acre after five minutes of burning can expand to 100 acres in the next 20 minutes under the right wind and fuel conditions. The Caldor Fire in California (2021) grew from a small ignition to a 5,000-acre blaze in just a few hours. The Camp Fire that destroyed the town of Paradise (2018) spread at roughly 80 football fields per minute at its peak.
The implication for response is stark: every minute between ignition and detection represents a geometric increase in the eventual containment problem. Fire suppression experts use a concept called the "initial attack window" — the period during which a small crew with hand tools can potentially stop a fire before it escapes to large-fire status. Historically, that window is 15 to 30 minutes for a fire burning in moderate conditions. Missing that window means the difference between a half-acre nuisance fire and a multi-thousand-acre incident requiring aircraft, thousands of personnel, and weeks of effort.
Current detection systems routinely miss that window. A fire that starts in a remote canyon at night may not be reported until someone drives past it hours later, or until it grows large enough to generate a smoke column visible on weather radar. Aerial detection patrols help in high-risk areas during peak season, but they cannot cover all terrain continuously. Ground-based camera networks (like those operated by ALERTCalifornia) provide real-time video detection but are limited to areas within line of sight of a camera tower, typically 20 to 40 miles.
FireSat's promise is to make the initial attack window reliably accessible for fires anywhere on Earth's landmass, not just in areas wealthy enough to deploy dense sensor networks. A rancher in a remote part of New South Wales and a fire agency in a rural Portuguese municipality would receive the same quality of detection coverage as a well-resourced California fire district.
To understand FireSat's significance, it helps to inventory the existing detection infrastructure and its limitations.
Geostationary satellites (GOES-East, GOES-West, Himawari): These satellites provide continuous coverage but at low spatial resolution (2 km/pixel). They detect large, established fires well but routinely miss small ignitions until they have burned for 30 minutes or more and grown to sufficient size.
Polar-orbiting weather satellites (MODIS, VIIRS on Suomi NPP and NOAA-20): These offer better resolution (375 meters per pixel for VIIRS active fire products) and are widely used by fire agencies. But polar orbiters revisit any given location only twice per day in each direction, creating large detection windows. A fire that ignites and dies before the satellite passes overhead goes completely undetected.
Commercial high-resolution satellites (Planet, Maxar, Airbus): Excellent resolution but not optimized for fire detection, and revisit rates remain inadequate for near-real-time applications except at premium cost for tasked collection.
Ground-based camera networks (ALERTCalifornia, PanTilt): Excellent for covered areas, real-time detection, but geographically limited and expensive to deploy at scale.
Aerial detection (fixed-wing patrols): Highly effective but expensive, weather-limited, and cannot cover all fire-prone terrain simultaneously.
FireSat's architecture is explicitly designed to fill the gap between the coarse-but-continuous geostationary sensors and the precise-but-infrequent polar orbiters. By achieving both reasonable resolution and high revisit frequency, it occupies a detection niche that no existing operational system fills.
FireSat does not exist in isolation. It is one component of a broader portfolio of climate-focused AI investments that Google has been assembling over the past several years, a portfolio that increasingly positions the company as an infrastructure provider for the climate response rather than merely a software company.
Google DeepMind's materials science work — including the Genesis AI system that identified 25 new magnetic materials from 380 million candidates — reflects a similar philosophy: use AI to dramatically accelerate discovery in domains where the rate-limiting factor is the ability to search vast possibility spaces. For wildfire detection, the vast possibility space is the surface of the Earth, sampled continuously and analyzed at scale. The machine learning infrastructure for identifying anomalies in high-dimensional sensor data is directly analogous to the infrastructure for identifying anomalous candidates in computational chemistry.
Google's investment in FireSat also reflects a strategic calculation about where AI creates defensible value. Fire detection algorithms require training on enormous, well-curated datasets combining satellite imagery, ground truth fire records, and contextual metadata. Google has the data infrastructure, the ML expertise, and the computational resources to build and maintain those models at a quality level that smaller organizations cannot easily replicate. The philanthropic framing of the project does not preclude the possibility that the underlying technology becomes a platform on which additional services — including commercial services — are built.
The company has also been investing in climate-relevant AI through its search and information products. The Google AI Mode rollout in the US has included integrations with authoritative climate and weather data sources, suggesting a longer-term vision of AI-augmented information infrastructure for climate resilience.
The FireSat constellation's orbital design is optimized for coverage of fire-prone landmasses. The highest wildfire risk zones on Earth — the western United States, southern Australia, Mediterranean Europe, the Amazon basin, sub-Saharan African savannas, and Southeast Asian peatlands — are concentrated between roughly 10 and 55 degrees latitude in both hemispheres. The constellation's orbital inclinations and phasing are tuned to maximize revisit frequency over these zones.
With 50+ satellites distributed across multiple orbital planes, FireSat achieves median revisit times of under 20 minutes over most fire-prone regions, with revisit times in the highest-risk zones (California, Mediterranean, southeastern Australia) approaching 10 minutes. As the constellation grows — Muon Space has indicated plans to expand beyond the initial deployment — revisit times will continue to decrease.
The ground processing infrastructure supporting FireSat is distributed across multiple ground stations to minimize downlink latency. Satellite contact windows with ground stations are a fundamental constraint for any low Earth orbit system: a satellite at 550 km altitude is in contact with any given ground station for only 5 to 10 minutes per orbital pass. By deploying a global network of ground stations, FireSat ensures that any given satellite's data reaches processing infrastructure within minutes of collection.
Processed fire alerts flow to a web-based platform and API designed for integration with existing fire agency software stacks. Rather than requiring agencies to adopt a new interface, the system is built to push alerts into tools like ArcGIS, CAD dispatch systems, and the National Interagency Fire Center's (NIFC) data infrastructure. This integration-first design philosophy reflects lessons learned from previous attempts to deploy space-based fire detection tools: the technical capability to detect a fire matters little if the alert cannot reach the right decision-maker in a usable form within the response window.
The downstream implications of reliable, near-real-time fire detection extend well beyond faster initial attack responses. Insurance companies managing wildfire exposure need early warning systems to trigger evacuation notifications and pre-position loss adjusters. Electric utilities — particularly in California, where transmission lines have caused multiple catastrophic fires — need to know immediately when a fire starts near their infrastructure so they can execute Public Safety Power Shutoffs or de-energize specific lines. Prescribed burn managers need confirmation that escaped smoke from planned burns has not ignited new spot fires.
FireSat's data also has value for post-fire analysis. The satellite record of a fire's early growth — where it started, how fast it spread, which direction it moved — is crucial for understanding fire behavior, validating fuel models, and designing better prevention strategies. Current fire origin investigations rely heavily on witness accounts and ground surveys, both of which are slow and imprecise. A satellite-captured thermal record from the first minutes of a fire's life provides forensic evidence of a quality that no other technology can match.
For the global insurance industry, which absorbed record wildfire losses in 2025 and 2026, reliable early detection could shift the actuarial calculus for wildfire exposure. Earlier containment of smaller fires means fewer large loss events. More precise data on fire behavior means better underwriting models. FireSat's data is likely to be incorporated into catastrophe models within years of its deployment, changing how the industry prices wildfire risk.
At the humanitarian level, the stakes are more direct. Faster detection means longer evacuation windows. In the Maui fires of 2023, victims had very little warning time before fires reached populated areas. In northern California's history of camp fire disasters, the difference between 30 minutes and 10 minutes of additional warning time is measured in lives. FireSat cannot guarantee that its detections will always translate into faster evacuations — that depends on the quality of emergency management systems, communication infrastructure, and community preparedness. But it can reliably provide the information that makes faster evacuations possible.
FireSat is not a complete solution to the wildfire detection problem, and Google has been transparent about its limitations.
Cloud cover: Thermal infrared sensors, like visible-light cameras, cannot see through thick cloud cover. During monsoon seasons or along coastal areas with persistent marine layer, detection performance degrades. FireSat is most effective in the hot, dry, low-humidity conditions that characterize the highest wildfire risk periods, which tend to also be the clearest atmospheric conditions — a useful correlation, but not a universal one.
False positive management: Even with sophisticated ML filtering, any system designed to maximize detection sensitivity will generate false alarms. A false alarm that triggers a full emergency response is costly and erodes trust in the system over time. Google and Muon Space have invested heavily in false positive suppression, but calibrating the detection threshold involves a fundamental tradeoff between sensitivity (catching real fires) and specificity (avoiding false alarms).
Integration friction: Fire agencies vary enormously in their data infrastructure, technical capacity, and operational culture. A sophisticated API is useful only to organizations with the technical staff to integrate it. Reaching volunteer fire departments in rural Montana or community fire brigades in rural Portugal requires a different kind of interface investment.
Latency under load: During a major fire weather event — when dozens of fires may ignite simultaneously across a region — the system's processing pipelines and alert distribution infrastructure face peak load at exactly the moment they are most needed. Designing for graceful performance degradation under peak load is an ongoing engineering challenge.
How is FireSat different from existing fire detection satellites like VIIRS?
VIIRS (Visible Infrared Imaging Radiometer Suite) on NOAA-20 and Suomi NPP provides excellent fire detection data and is widely used by agencies globally, but it revisits any given location only twice per day. FireSat's constellation architecture allows revisit times of under 20 minutes. FireSat also uses on-board ML to filter detections in near-real-time, whereas VIIRS data typically requires post-processing before actionable alerts are generated.
Can FireSat detect underground peat fires or smoldering fires without open flame?
Smoldering peat fires emit significantly less thermal energy than open flame combustion, making them harder to detect at satellite resolution. FireSat's current detection threshold of approximately 20 square meters assumes active flaming combustion. Smoldering detection is technically possible but requires longer observation periods to build statistical confidence from smaller temperature anomalies. This is an active area of ongoing algorithm development.
Who has access to FireSat data?
Google has indicated that data will be made available to fire management agencies, emergency services, and research institutions. The exact access model — whether data is open, available through government partnerships, or requires a subscription for commercial users like utilities — has been detailed on a case-by-case basis during the initial deployment phase. Google.org-funded access is prioritized for public safety agencies in high-risk regions.
How does FireSat integrate with existing fire agency software?
FireSat's ground processing infrastructure outputs alerts through a REST API and standard geospatial data formats (GeoJSON, KML). Integration partners have included ESRI (ArcGIS), NIFC, and several state fire agencies during the testing phase. Commercial CAD system integrations are in development. The design philosophy is to push data into existing agency workflows rather than require adoption of a new platform.
What is the long-term plan for the FireSat constellation?
Muon Space and Google have indicated plans to expand the constellation beyond the initial 50+ satellites as launch economics allow. SpaceX's Falcon 9 rideshare program has dramatically reduced per-satellite launch costs, making constellation expansion significantly more affordable than it was five years ago. The longer-term roadmap includes higher-resolution sensors, additional spectral bands for vegetation stress mapping (a leading indicator of fire risk), and tighter integration with weather modeling systems for dynamic fire risk assessment.
The problem FireSat addresses is not new. The satellite imagery to detect wildfires has existed for decades, and the machine learning techniques to analyze it have matured rapidly over the past five years. What FireSat provides is the specific combination of spatial resolution, revisit frequency, and processing speed that converts a theoretical capability into an operational one. The technology is ready. The fires are not waiting.
As the 2026 fire season approaches across the northern hemisphere, FireSat's constellation will face its most rigorous operational test. If it performs as designed — delivering actionable alerts within minutes of ignition, across all fire-prone terrain, regardless of time of day — it will represent one of the most significant advances in wildfire management infrastructure since the introduction of aerial tankers in the 1950s. And unlike those tankers, it will scale to every fire-prone landscape on Earth, not just the ones wealthy enough to afford the aircraft.
Google launches Gemini 3.1 Flash-Lite at $0.25 per million input tokens — 2.5x faster than Flash with frontier-class performance, making production AI affordable for startups and enterprises running high-volume workloads.
Google Research unveils AI Co-Scientist, a multi-agent system built on Gemini 2.0 that autonomously generates, evaluates, and refines biomedical hypotheses — promising to compress years of research into weeks.
Apple is launching a completely redesigned Siri powered by Google's Gemini model with Private Cloud Compute, positioning 1 billion iPhones as the world's largest consumer AI platform.