Earth Day in the age of AI

Earth Day began in 1970, the United Nations later recognized April 22 as International Mother Earth Day, and the official Earth Day 2026 theme is “Our Power, Our Planet.” That theme lands differently in a year when a growing share of the world’s electricity, water, and industrial capacity is being pulled toward artificial intelligence. Earth Day was built around public pressure, civic action, and a demand that industry stop treating nature as an acceptable sacrifice. AI now belongs inside that argument, not outside it.

The easy version of this story is comforting and incomplete. AI is often presented as a climate tool that will sharpen forecasts, catch methane leaks, improve electric grids, and speed scientific discovery. Some of that is true. Another version, just as incomplete, says AI is a giant environmental mistake and little more than an energy-hungry status symbol. That misses something important too. AI is not one thing. It is a stack of models, chips, cooling systems, data centers, supply chains, procurement decisions, and public claims. Any serious Earth Day conversation has to look at the whole stack.

The climate crisis is already worsening with every increment of warming, and the broader ecological strain is not limited to carbon. The UN and UNEP frame the moment as a triple planetary crisis of climate change, biodiversity loss, and pollution and waste. That framing matters because AI touches all three at once. Its environmental footprint is not only about electricity. It also reaches water stress, mining, manufacturing, land use, and the waste stream created when hardware turns over faster than the planet can absorb it.

Earth Day meets the compute era

Earth Day has always been a fight over what counts as progress. In 1970, the immediate targets were visible and local: toxic air, dirty rivers, oil spills, pesticides, industrial waste. The modern version of that argument is harder to see because much of the damage is hidden behind interfaces. A chatbot reply feels weightless. A generated image feels instant. A cloud platform feels like abstraction, not industry. Yet AI is physical from end to end. It runs through transmission lines, transformers, chillers, water systems, chip fabrication plants, construction contracts, and logistics networks. That physical reality is where the Earth Day lens becomes useful again.

The theme “Our Power, Our Planet” is nominally about public agency. It is also an accidental description of the AI era. Power now has two meanings that collide. One is democratic power, the kind Earth Day has always tried to build from the ground up. The other is electrical power, the resource increasingly required to train, serve, and scale AI systems. The tension is not rhetorical. It is structural. A movement built around environmental accountability is now confronting a technology whose benefits are widely distributed in marketing language while its environmental burdens are often concentrated in specific places: near data centers, along transmission corridors, at water-stressed sites, and across mineral supply chains.

That is why the Earth Day and AI debate cannot be reduced to a moral gesture about whether people should use chatbots less. Individual restraint matters far less than infrastructure choices, disclosure standards, and workload design. One extra prompt is not the real story. The real story is the aggregate demand curve, the siting of facilities, the electricity mix feeding them, the frequency of hardware replacement, and whether the systems being built are actually justified by public value. Earth Day has always asked a blunt question: who bears the cost of convenience and growth? AI needs to be asked that question with the same seriousness once reserved for smokestacks and tailpipes.

The public mood around AI has also shifted Earth Day’s usual political geometry. Environmental debates once pitted fossil-heavy incumbents against clean alternatives. AI scrambles that picture. It is pushed by some of the same companies buying renewable energy at large scale, investing in cleaner data centers, and publishing sustainability reports. Those efforts are real. They also coexist with rising power demand, rising operational emissions in parts of the sector, and a rush to build more capacity before standards have caught up. The point is not hypocrisy. The point is that cleaner operations do not automatically cancel growth in absolute impact.

Earth Day has never worked well as a branding exercise. It works best when it turns sentiment into a test. Applied to AI, that test is straightforward: Does this system create enough public and ecological value to justify its material footprint, and is that footprint being measured honestly? If the answer is vague, the technology is not ready for applause. If the answer is measurable, then AI belongs in the environmental toolkit, but only under terms stricter than the industry usually prefers.

The electricity behind the magic

Electricity is where the environmental argument around AI becomes unavoidable. The International Energy Agency estimated that data centers accounted for about 1.5% of global electricity consumption in 2024, or around 415 TWh, and projects that global data center electricity consumption could rise to about 945 TWh by 2030 in its base case. The same IEA analysis says data center electricity demand has been growing much faster than overall electricity demand, with the United States taking the largest share of consumption in 2024, followed by China and Europe.

Those figures matter because AI is not just another workload dropped into a stable system. It changes the profile of demand. Training frontier models requires dense compute over long periods. Serving those models to millions of users adds a second wave of demand, one tied to constant inference rather than a single training run. Cooling then adds overhead on top of the IT load, and local grids have to absorb not just more consumption but different load shapes and sharper peaks. That is why the debate has moved so quickly from abstract worries about “AI energy use” to utility planning, grid interconnection queues, and arguments over who pays for new infrastructure.

In the United States, a Department of Energy-backed Lawrence Berkeley National Laboratory report found that data center load growth has tripled over the past decade and could double or triple again by 2028. DOE’s public release of that report framed the result in the language of urgent system planning, not speculative futurism. That is a useful correction. Too much public discussion treats AI power demand like a science-fiction number that might arrive someday. Parts of it are already embedded in utility forecasts, siting battles, and industrial strategy.

A compact map of AI’s direct environmental ledger

This is the basic ledger Earth Day readers should keep in mind. AI does not have one environmental number. It has a chain of interacting numbers, and the emissions result depends heavily on location, hardware efficiency, workload design, and grid mix rather than on model size alone.

The industry response is usually to point out, correctly, that efficiency is improving. Better chips do more work per watt. Some operators are signing clean energy deals. Some data centers are being designed with lower overhead. Google says its data center emissions fell in 2024 even as energy demand increased, and Microsoft continues to report work on lowering energy and water intensity across its data center footprint. Those claims matter, and they should not be dismissed. The harder question is whether efficiency is outrunning growth in demand. Right now, the broader evidence suggests that demand growth remains the dominant force.

That is why Earth Day rhetoric about “green AI” needs a stricter vocabulary. A model can be more efficient than its predecessor and still contribute to a system whose total environmental burden is rising. An operator can buy more clean electricity and still increase absolute resource use. The environmental movement learned this lesson long ago in other sectors. Cleaner unit economics are not the same thing as ecological sufficiency. Absolute impact still counts.

Water, metals, and short hardware cycles

Electricity dominates headlines because it is easier to quantify. Water, materials, and hardware turnover are less visible and often underreported, yet they are central to the AI footprint. UNEP’s 2024 issue note on AI’s end-to-end environmental impact argues that the full lifecycle has to be considered, from extraction and manufacturing through transport, use, and end-of-life. That lifecycle view is not academic. It changes the entire ethical picture. An AI system is not just software running on anonymous cloud infrastructure. It is a material system that starts long before inference and continues long after deployment.

Water sits near the middle of this problem. Data centers use water directly for cooling in many designs, and indirectly through electricity generation and upstream industrial processes. The difficulty is that water numbers are often either missing or presented in ways that do not reveal local stress. A liter consumed in a wet region is not ecologically identical to a liter used in a drought-prone basin. That is one reason the ITU’s work on AI environmental assessment keeps returning to the same point: water is one of the least transparent parts of the current reporting landscape.

Industry reporting illustrates both the progress and the limits. Microsoft’s 2025 environmental sustainability reporting highlights work on water-positive goals and on lowering water use effectiveness in data centers. Google says it replenished 64% of the freshwater it consumed across offices and data centers in 2024 and reported lower data center energy emissions that year. Those disclosures are better than silence, but they are not yet a substitute for comparable, independently legible standards across the sector. Company reports are useful, but they are not the same thing as a common public accounting system.

The hardware side is even more easily ignored. AI accelerators are manufactured through energy-intensive, resource-intensive supply chains that depend on minerals, complex fabrication, precision manufacturing, and frequent refresh cycles. A 2025 arXiv paper from Google researchers on the life-cycle emissions of AI hardware describes one of the first comprehensive cradle-to-grave assessments of AI accelerators and their host systems, including manufacturing emissions. That kind of work matters because it pushes the conversation beyond the comforting fiction that the footprint of AI begins only when a model is switched on.

Then there is the waste stream. The Global E-waste Monitor 2024 reported that the world generated 62 million tonnes of e-waste in 2022, with only 22% formally recycled. Not all of that is AI hardware, of course. Still, AI’s appetite for accelerated infrastructure investment lands inside a digital economy that already struggles to recover materials and manage obsolete equipment responsibly. Earth Day politics used to focus on plastic because plastic was visible. Electronics are harder: dense, global, toxic in parts of their lifecycle, and deeply entangled with geopolitics and industrial secrecy. That does not make them less important. It makes them more urgent.

The environmental case for slower hardware turnover, longer equipment life, circular procurement, and better recycling is not glamorous, which is exactly why it matters. Environmental progress rarely comes from glamour. It comes from standards, maintenance, lifecycle thinking, and a willingness to treat waste as design failure rather than as somebody else’s downstream problem. AI will deserve “planet-friendly” marketing only when those habits become routine rather than exceptional.

Bad accounting makes bad debate

The public argument about AI’s environmental impact is often noisy for a simple reason: measurement is still weak. Some estimates focus on training runs and ignore the years of inference that follow. Others count operational electricity but miss cooling overhead, idle capacity, hardware manufacturing, or water use. Some turn a single prompt into a universal proxy for “AI impact,” even though different models, devices, contexts, and serving stacks vary enormously. ITU’s 2025 report Measuring What Matters describes the field as fragmented and inconsistent, with missing data, inconsistent boundaries, and poor comparability across methods.

That weak accounting produces two opposite distortions. One side uses large headline numbers to suggest AI is ecologically indefensible by default. The other side responds with selective efficiency figures that make the problem look trivial. Both moves can mislead. A single real-world example may be instructive without being representative, while a dramatic training estimate may be accurate for one model and useless for another. Bad accounting does not only create confusion. It creates room for marketing.

A good example of the nuance is Google’s 2025 paper on measuring AI delivery at production scale. The authors report that the median Gemini Apps text prompt consumed about 0.24 Wh of energy and around 0.26 mL of water in their measured environment, far below many public guesses. That is a valuable result because it replaces speculation with telemetry for a real system. It is also easy to misuse. The result does not mean all AI workloads are environmentally negligible. It means one specific serving stack, measured carefully, looked very different from loose public extrapolations. The real lesson is methodological: observed data is better than guesswork.

The same caution applies to training-versus-inference arguments. Training frontier models is extremely resource-intensive. But once a model is deployed widely, the cumulative burden of serving billions of requests becomes central to the ledger too. That is why lifecycle work and system-wide measurement matter so much. The environmental impact of AI is not a single event. It is a continuous operating condition attached to a product, a user base, and a hardware stack over time.

What honest AI environmental reporting would include

A reporting framework like this would not solve the footprint problem by itself. It would do something equally important: remove the fog that lets weak claims travel farther than solid evidence. That is why ITU’s recent guidance and UNESCO’s ethics framework both matter. They turn environmental accountability from a voluntary gesture into something closer to a governance expectation.

Once the measurement problem is taken seriously, the debate gets sharper. The question is no longer “Is AI green or dirty?” The real question is: which AI system, for what task, measured across which boundary, in which place, using which electricity, with which hardware, and to deliver what public value? Earth Day does not need broader slogans than that. It needs this level of specificity.

AI where the climate case is real

The strongest case for AI in environmental work is not that it is magical. It is that, in certain domains, it does something concrete better, faster, or at lower computational cost than older approaches. Weather forecasting is the clearest example. In late 2025, NOAA announced a new suite of operational AI-driven global weather prediction models, describing them as faster and more accurate guidance that uses a fraction of the computational resources required by traditional methods. The World Meteorological Organization has since endorsed actions to promote AI for forecasts and warnings while stressing that these systems should complement rather than replace traditional forecasting tools.

That is a serious environmental use case, not a decorative one. Better forecasting protects lives, reduces economic loss, supports disaster preparation, and improves the timing of interventions across energy, agriculture, and emergency management. When forecast skill rises and compute cost falls, the public case is stronger than it is for countless convenience applications. Google DeepMind’s GraphCast and GenCast work also points in this direction, showing strong forecast performance and much faster generation than conventional numerical pipelines for certain tasks. The broader pattern is clear: AI earns its best environmental argument when it improves decision quality in systems already under climate stress.

The energy system is another place where AI’s value can be real rather than speculative. The IEA’s Energy and AI work frames AI as both a driver of electricity demand and a tool for improving parts of the energy system, including grid operations, forecasting, maintenance, and integration of supply and demand. That dual role matters. It would be intellectually lazy to treat AI only as a load problem when it can also sharpen the management of a more electrified system. The catch is that enabling value has to be measured against direct footprint rather than asserted as a free offset.

Agriculture provides another test of seriousness. The World Bank’s recent work on AI in agricultural transformation argues that AI can support climate resilience and productivity for smallholder farmers, but only where the basic infrastructure, governance, and local fit are in place. That caution is important. A lot of AI-for-climate talk assumes the technology is useful simply because the problem is hard. Real environmental value appears when tools are grounded in data quality, institutional capacity, and the needs of the people expected to use them. The Earth Day version of this point is simple: technology that ignores place usually ignores ecology too.

What separates these stronger use cases from weaker ones is not just nobility of intent. It is the ratio between footprint and benefit. A forecasting system that arrives faster, costs less computationally, and improves early warnings has a stronger public claim than an AI feature built to keep users inside a platform for a few extra minutes. Earth Day should force that distinction into the open. Not every AI application deserves the same moral status just because it shares a technical foundation.

Machine learning in the sky and at sea

The most convincing environmental uses of AI often begin with observation. Earth systems are vast, dynamic, and data-rich. Satellites, weather stations, ocean sensors, and industrial monitoring networks produce more information than humans can sort manually. AI is well-suited to that setting because it can classify patterns, flag anomalies, and speed interpretation across large volumes of data. ESA describes AI in Earth observation as “a force for good,” not as a slogan but as a practical response to a data challenge: turning enormous streams of satellite information into operational insight.

That observational power matters for climate and biodiversity work because the hardest environmental problems are often measurement problems first. You cannot cut methane leaks you cannot see. You cannot protect fisheries effectively if the activity at sea remains opaque. You cannot respond well to fast-moving wildfires if the best models arrive too late or at the wrong resolution. AI is not the whole answer in any of those domains, but it sharpens visibility, and visibility changes enforcement and response.

UNEP’s International Methane Emissions Observatory and its Methane Alert and Response System show the point well. UNEP says MARS is the first global satellite observation system that detects, analyzes, and communicates methane emission data, and in 2025 the organization reported that it had sent more than 3,500 alerts across 33 countries using satellite monitoring and AI-supported analysis. Methane is a short-lived but potent climate pollutant, so detection speed and transparency matter. This is not abstract “AI for good.” It is a direct use of machine-assisted analysis in a problem where faster notice can produce near-term climate gains.

Wildfire management offers another version of the same logic. NASA’s Wildfire Digital Twin project aims to produce much finer-resolution models of fire and smoke in minutes rather than hours, while newer NASA-supported sensing work uses AI in compact instruments to improve fire observation. In a world of longer fire seasons, heavier smoke impacts, and cascading effects on water supplies and public health, that kind of speed is not cosmetic. Time is environmental infrastructure too. Faster insight can change evacuation decisions, air-quality responses, and the protection of ecosystems already under stress.

At sea, Global Fishing Watch has used AI, satellite data, and open analysis to reveal patterns of human activity that were previously hard to map at scale, including potentially illegal fishing and industrial encroachment. That matters for biodiversity, food security, labor protection, and state capacity. Environmental politics often gets trapped in a false choice between high technology and grassroots accountability. These tools show a better model: technology in service of enforcement, public visibility, and shared evidence. Earth Day has always needed better facts. AI is most defensible when it produces them.

Still, none of this erases the footprint discussed earlier. It changes the balance sheet only when the use is meaningful, the outcome is measured, and the system is governed. A methane-monitoring model is not morally equivalent to a frivolous generative feature because they draw electricity from the same grid. Earth Day should be willing to rank AI uses rather than talk about the technology as a single block. That is a more honest environmental politics than the blanket enthusiasm that usually accompanies AI launch season.

Efficiency gains do not erase demand growth

One of the most persistent errors in digital sustainability debates is the belief that efficiency automatically solves scale. It rarely does. More efficient chips and cleaner data centers are real gains. They can lower the impact per unit of work, and sometimes they do so dramatically. But when cost drops, speed rises, and access expands, demand often grows as well. That pattern is older than AI. In this sector, it now shows up as the gap between per-task efficiency improvements and rising total infrastructure demand.

The ITU and World Benchmarking Alliance’s Greening Digital Companies 2025 report captured the tension cleanly. It found that AI-driven companies saw operational emissions rise sharply since 2020, even as transparency improved and some firms strengthened climate commitments. That does not prove that efficiency efforts are fake. It proves that efficiency and absolute growth can happen at the same time. The environmental movement should already be familiar with that outcome. Cars got cleaner per mile. Total transport emissions still became a massive problem. Appliances became more efficient. Household electricity use did not vanish. AI sits squarely inside that same pattern.

The IEA’s latest analysis adds another layer. The agency does not present one fixed future for AI-related electricity demand. It lays out scenarios and notes that infrastructure bottlenecks, policy, hardware availability, and electricity supply will shape outcomes. That uncertainty matters. It argues against theatrical certainty from both boosters and critics. Yet uncertainty is not a license for complacency. When the credible range is already large enough to reshape grid planning, the burden shifts from prediction to preparation.

The rebound problem is not only about electricity either. A company that lowers the cost of serving AI may choose to embed AI into more products, more workflows, more search queries, more documents, more customer support channels, and more operating systems. A cheaper unit of compute can multiply total usage. That is why arguments built around a single low per-prompt number, even a carefully measured one, need to be handled carefully. A low number multiplied across a huge and expanding user base is still an infrastructure story.

Earth Day should also resist a subtler version of the rebound story: moral rebound. Once a company markets AI as useful for climate science or grid management, it becomes easier to imply that the rest of its AI expansion belongs under the same halo. It does not. A beneficial use case is not a blanket pardon for a business model. Environmental accounting has to stay granular enough to separate the high-value applications from the attention-engine features that mainly increase compute demand because they increase engagement.

Standards, disclosure, and public procurement

The governance side of this subject is finally starting to catch up, though not fast enough. UNESCO’s Recommendation on the Ethics of Artificial Intelligence states that environmental and ecosystem flourishing should be recognized, protected, and promoted through the lifecycle of AI systems. That principle matters because it places environmental stewardship inside AI governance rather than outside it. It rejects the old habit of treating environmental damage as an unfortunate side issue to be handled later, if growth allows.

ITU’s 2025 and 2026 work goes further into measurement and comparability. The 2025 report on assessing AI’s environmental impact lays out the fragmentation in current methods. The 2026 guideline gives lifecycle-assessment-based guidance for comparing AI systems with non-AI alternatives or with other AI systems. That is the direction Earth Day politics should want: less vague virtue, more comparable evidence. You cannot regulate, procure, or choose responsibly when every environmental claim is built on a different boundary and a different set of omissions.

UNEP has also pushed the conversation from critique to institution-building. In 2025 it backed a coalition intended to put AI on a more sustainable path, while its broader AI and environment work keeps returning to the same priorities: lifecycle thinking, better measurement, and governance that recognizes environmental trade-offs early rather than after deployment. That is important because the sector’s habit is to move first, rationalize later, and standardize only when external pressure arrives. Earth Day exists to create that pressure.

Public procurement may end up doing more than public speeches. Governments, universities, and large enterprises buy enormous amounts of digital infrastructure and cloud services. They can require standardized disclosures on energy use, water use, carbon intensity, hardware lifetime, and end-of-life handling. They can ask whether an AI system performs a task materially better than a non-AI alternative. They can favor workloads that deliver public goods over ornamental automation. The smartest climate policy is not always a ban. Sometimes it is a purchasing rule with teeth.

The minimum disclosure Earth Day should demand from AI vendors

This is not an impossible standard. Parts of it already exist in company reports, research papers, ITU guidance, and environmental accounting practice. What is missing is consistency and public comparability. Earth Day should push AI from selective transparency into ordinary accountability.

Earth Day needs a tougher test for AI

Earth Day does not need to choose between romantic anti-tech purity and blind technological faith. That is a lazy binary. The harder and more honest position is better: use AI where it clearly improves environmental understanding, resilience, and enforcement; refuse the idea that usefulness in one domain excuses waste everywhere else; and demand measurement rigorous enough to tell the difference.

That tougher test would change the conversation quickly. It would ask whether a model is small enough for the task. It would ask whether a non-AI method already works. It would ask whether the environmental claim includes training, inference, cooling, hardware, and waste. It would ask whether the system is sited and powered responsibly. It would ask whether the public benefit is real, measurable, and larger than the ecological bill. Those are not anti-AI questions. They are the first adult questions the sector should have been asked at scale.

Earth Day has always been strongest when it refuses abstraction. Rivers are not abstractions. Lungs are not abstractions. Landfills are not abstractions. The AI debate becomes more serious when compute is treated the same way. A server farm is not a metaphor. Cooling water is not a metaphor. A transformer upgrade is not a metaphor. Neither is the mine, the fab, the shipping lane, or the discarded board that ends up in a waste stream with too little formal recovery. Once AI is seen as industry rather than magic, environmental politics can finally get a grip on it.

The official Earth Day 2026 theme says “Our Power, Our Planet.” The best reading of that line in the AI era is not celebratory. It is demanding. Power, in every sense, has to answer to the planet. AI deserves support where it sharpens forecasts, spots methane, maps illegal extraction, improves wildfire response, and gives policymakers better sight. It deserves resistance where hype outruns value, where disclosure is evasive, and where “efficiency” is used to hide rising total impact. Earth Day is not asking whether AI is exciting. It is asking whether AI is accountable to the world it depends on.

FAQ

Author:
Jan Bielik
CEO & Founder of Webiano Digital & Marketing Agency

This article is an original analysis supported by the sources cited below

Earth Day 2026 | Theme, Activities, Events & Resources
Official Earth Day 2026 page with the current global theme and campaign framing.

The History of Earth Day
EarthDay.org’s overview of the movement’s origins and historical development.

International Mother Earth Day
United Nations background on April 22 as an official observance.

AR6 Synthesis Report Climate Change 2023
IPCC synthesis of the latest assessment cycle on climate risks, impacts, and mitigation.

Energy and AI
IEA’s flagship analysis of how AI is affecting energy demand and supply.

Energy demand from AI
IEA section detailing projected growth in electricity demand from data centers and AI workloads.

DOE releases new report evaluating increase in electricity demand from data centers
U.S. Department of Energy summary of Lawrence Berkeley Lab’s latest data center load findings.

2024 LBNL data center energy usage report
Lawrence Berkeley National Laboratory publication page for the updated U.S. data center energy report.

The environmental impact of the full AI lifecycle needs to be considered
UNEP issue note on AI’s end-to-end environmental footprint.

The Global E-waste Monitor 2024
ITU’s global reference on e-waste volumes, recycling rates, and policy progress.

Greening Digital Companies 2025 Monitoring emissions and climate commitments
ITU and World Benchmarking Alliance assessment of emissions, energy use, and climate commitments across major digital firms.

How to assess AI’s environmental impact
ITU report on current measurement approaches, reporting gaps, and methodological weaknesses.

Guidelines for assessing the environmental impact of artificial intelligence systems
ITU guidance that applies lifecycle assessment principles to AI systems and comparisons.

Recommendation on the Ethics of Artificial Intelligence
UNESCO’s global ethics framework, including environmental and ecosystem protections.

Life-Cycle Emissions of AI Hardware A Cradle-To-Grave Approach and Generational Trends
Research paper on cradle-to-grave emissions from AI hardware and host systems.

Measuring the environmental impact of delivering AI at Google Scale
Production-scale measurement study of AI serving energy, carbon, and water use.

Our 2025 Environmental Sustainability Report
Microsoft’s sustainability reporting on carbon, water, waste, and data center efficiency.

Sustainable innovation and technology
Google’s public sustainability reporting, including recent data center emissions and water disclosures.

NOAA deploys new generation of AI-driven global weather models
Official NOAA announcement on operational AI weather prediction systems.

World Meteorological Congress endorses actions to promote AI forecasts and warnings
WMO statement on the institutional push to use AI in forecasting and early warnings.

GraphCast AI model for faster and more accurate global weather forecasting
Google DeepMind overview of its AI weather forecasting model and performance claims.

AI in Earth observation a force for good
ESA discussion of AI applications in satellite-based Earth observation.

International Methane Emissions Observatory
UNEP overview of its methane monitoring initiative.

Methane Alert Response System MARS
UNEP’s explanation of the satellite-based methane alert system.

Better data driving action on methane emissions, but more work needed
UNEP update on methane alerts, response rates, and AI-supported monitoring.

NASA Wildfire Digital Twin pioneers new AI models and streaming data techniques for forecasting fire and smoke
NASA article on AI-supported wildfire and smoke forecasting.

Novel NASA sensor incorporates AI into lightweight camera for observing fires
NASA technology piece on AI-assisted fire sensing.

New research harnesses AI and satellite imagery to reveal the expanding footprint of human activity at sea
Global Fishing Watch research summary on AI and satellite monitoring of maritime activity.

Our technology
Global Fishing Watch explanation of how machine learning and satellite data support ocean monitoring.

Harnessing artificial intelligence for agricultural transformation
World Bank analysis of AI’s role in agriculture, resilience, and governance conditions.