What is Earth observation, really — and why, after fifty years of satellite imagery, is it still not “mainstream”?

In this episode, I’m joined by Aravind Ravichandran, founder of TerraWatch, an independent research and advisory firm focused entirely on Earth observation. Aravind writes the TerraWatch newsletter, runs the EO Summit, and spends his time thinking about the strategy and economics of the industry more deeply than just about anyone.

We start with a deceptively simple question — is Earth observation even an industry? — and end up somewhere more interesting: Aravind’s argument that when the technology truly succeeds, it becomes invisible, quietly embedded in agriculture, insurance, energy, and defense the same way weather satellites already are.

Along the way, we get into:

• Why 60+ countries are now building their own satellite constellations, and whether they’ll still exist in five years
• What Planet restricting imagery access really means — and why Aravind thinks they were “punished for doing something progressive”
• The technology actually moving the needle: hyperspectral data going free, AI foundation models, edge computing on satellites, and inter-satellite laser links
• Which use cases are genuinely picking up (utilities, parametric insurance) — and which were always hype (counting cars in parking lots)
• The defense paradox: how the industry that built Earth observation may also be the biggest thing holding back its commercial future

Some open questions we sit with: If satellite data is critical infrastructure, what happens when someone turns it off?

Should high-resolution imagery of the whole world be open — and what are the privacy and security costs if it is? And can sixty countries ever pool their data, or will sovereignty always trump logic?

In Conversation

What Is Earth Observation — and Is It Even an Industry?

Daniel: Maybe we should have started here. What is Earth observation, for you?

Aravind: Earth observation is essentially synonymous with remote sensing — anything that senses the Earth remotely. That includes satellites, which are the biggest part of the market in terms of money going in, but also drones, helicopters, and aerial. And by the very definition of how remote sensing works, you cannot make satellite data work without ground measurements and validation, so you include that in the mix too. In terms of satellites, it’s remote-sensing satellites — weather satellites, imaging satellites — not communication, not navigation.

Daniel: I always think of imaging — passive and active sensors. I draw on my university days, when Earth observation was called remote sensing and it was looking at satellite images.

Aravind: Fair enough. My definition doesn’t go into geospatial, because geospatial is a wider domain — it’s a discipline all by itself. Remote sensing is just a horizontal technology; it’s not a discipline, and it will stay that way. That’s also why I wouldn’t really call Earth observation an industry. It’s a subsector within space. Geospatial is definitely an industry, and Earth observation is a small part of it. When it goes mainstream, it doesn’t become a huge industry by itself — it becomes invisible. Invisible in agriculture, in insurance, in energy.

The Mainstream Question

Daniel: So your argument is that it’s not mainstream yet. Can you point to a technology and say “that’s mainstream”?

Aravind: Even within Earth observation, the most mainstream part is weather. You look at the forecast on your phone, and 80 to 90% of that was supplied by satellite data. I’ve surveyed people and they can’t name a single weather satellite — which is great, because that’s how it’s supposed to be. It’s invisible infrastructure. Earth observation isn’t there yet, because we’re still filling gaps. Can you monitor all the farms in the world? In the West you can with 10-meter resolution, now open and free through Copernicus. But farms in Asia, Africa, and Latin America are very small, so you need high resolution — and high resolution is expensive. So it takes a long time to get mainstream.

Daniel: Arthur C. Clarke — the future is here, it’s just unevenly distributed. I’d argue it’s mainstream in some places and definitely not others.

Aravind: Exactly. And to be mainstream you also need to be operational and applied to the point that people forget it exists — that’s when it becomes mainstream.

Sovereignty and the Rush to Build Constellations

Aravind: Where we are in the world right now is countries — and in some cases regions — launching their own assets into space to have independence, especially because satellite data is foundational to national security. Defense and intelligence has always been the dominant industry in terms of revenue. There were two milestones: 2022 with the war, and 2025 when the new US administration came online. The first seven years I was in the industry were more about collaboration and sharing data; since then we’ve gone in the other direction. We track about 61 countries that are either planning or have already started deploying their own satellites.

Aravind: I have strong opinions on whether they’ll exist in five years — how many will actually have the funds to keep deploying. It’s still a $100 million investment to have something good enough online. One thing is to put a flag on your satellite and say you have one; another is to actually have your Ministry of Agriculture or Defense use that data. And you have to ask whether it’s future-proof. Planet already has hundreds of 3-meter satellites — so if you’re a government launching a 3-meter constellation today, is that future-proof?

Daniel: The technology is almost becoming a commodity — you go and get one off the shelf and throw it into space.

Aravind: Take the Canary Islands — a small group of islands, part of Spain. They have plans to launch their own constellation. Technology has become so commoditized that it’s maybe a 30 or 40 million investment — enough for six to ten satellites covering your area with imagery every couple of days. Look at Iceye: founded around 2016 to monitor ice melting over the Arctic with SAR, now valued at two or three billion euros with over 250 million in revenue, three-quarters of it from defense. They have a good manufacturing plant, so a country can have SAR satellites launched in a year or two — or even just lease 30% of the capacity of something already in orbit. Five years ago you couldn’t do that.

The Planet Access Story

Daniel: Planet recently announced they’re restricting access to satellite data. Do you see this as another turning point? If you treat something as infrastructure and then someone turns it off, that’s a big deal.

Aravind: It’s only making the sovereignty discussions more validated. It’s not great — whether you’re a journalist, in the humanitarian sector, or a commercial company that had its access cut off. I’ve spoken to happy commercial customers who suddenly can’t see what’s happening around their asset. But the under-appreciated point is that Planet did something progressive for the industry. Before them, you didn’t have an API for satellite imagery — Planet launched one. They gave data to journalists, humanitarians, and the OSINT community for free. Now they’ve taken steps forward, but they’re being punished for going further. A couple of other American companies also don’t supply data over the US, but they were never easy to get data from anyway, so they didn’t have to put out a press release. Planet is being portrayed as a villain for trying to do a little thing better.

Daniel: I’m not here to jump on Planet either. But it’s another example of someone turning off what a lot of people assumed was critical infrastructure. If I were thinking about this, I’d want to future-proof myself — we need more options.

Aravind: It’s making the infrastructure point clearer for a lot of governments. If you had trouble getting budgetary approval for your own constellation before, it should be easier now.

Technology Trends: Sensors, Edge Computing, and AI

Daniel: What other technology trends are you seeing, hardware and software?

Aravind: On hardware, there’s the full spectrum of sensors — thermal, hyperspectral, multispectral, optical, SAR, LIDAR. And we shouldn’t forget the science missions; they’re the foundation behind everything companies do. ESA’s biomass satellite with P-band SAR, NASA’s NISAR with L-band SAR — amazing instruments that don’t exist in the commercial world. I’d also point to edge computing and onboard processing: processing the image on the satellite and just downlinking an alert or a location — say, spotting a ship and sending the location down. In parallel you have inter-satellite links and laser communication, which means you can beam an image down in a few minutes. So there’s a trade-off, but technologically both are becoming possible.

Aravind: On software, you can’t talk about 2026 without AI. Earth observation is one of the best use cases. ECMWF, the European weather agency, now runs an AI weather model alongside the physics-based one. On imagery, there’ve been exciting developments — Google’s Earth AI and their foundation model, the AI2 model, the NASA-IBM model, ESA’s work, and custom models fine-tuned for agriculture or flooding. My hope is that these make people aware of what satellite data can do. Before, to find out what you could get from a satellite, you’d spend a few thousand dollars and wait a couple of weeks for someone to hand you a dashboard. If you can get that down to a few minutes and zero cost, that’s a step change — because not knowing what to do with the data has been one of the biggest barriers in the industry.

Use Cases That Are Actually Picking Up

Daniel: What use cases are picking up — and are there winners and losers?

Aravind: If I’m cynical, the only industry really picking up is defense and intelligence — it’s 80% of revenue. For Planet, a public company, 80% of revenue is from government. But there are a couple of areas where Earth observation is almost non-substitutable. One is utilities. If you have a pipeline, power line, or railway running tens of thousands of kilometers, you can’t monitor all of it with drones or helicopters — satellites give you scale. Companies like PG&E in California use satellite data to monitor their assets, combined with drones where you need 7 or 10 cm resolution. The other is parametric insurance — automating payouts using proxies like soil moisture instead of sending someone to inspect a flooded farm. That’s especially important in non-Western markets, where farmers in India, Africa, and Southeast Asia can get paid out objectively, before any middleman interferes.

Daniel: When I started this podcast, the classic example was counting cars in a parking lot to figure out the value of a stock. That didn’t really pan out, did it?

Aravind: The billion-dollar parking-lot market was hype. But the technology behind it — the computer-vision model — is foundational, especially in commodity trading: counting containers in ports, seeing what was shipped and what wasn’t. That was very useful in the early weeks of the war for tracking how many barrels of oil were being shipped from where. So there’s real value, just not in that specific headline use case.

The Defense Paradox

Daniel: We’ve got more satellites coming, AI, cheaper compute and storage, cloud-native formats making access easier. So why isn’t it mainstream? Why isn’t it bleeding into more industries?

Aravind: I call it the defense paradox. The industry that contributed the most to the growth of Earth observation is defense — but it’s also the biggest hindrance to the commercial market. Defense has a lot of say in whether you can supply data and what the price is. The price of your data is X because the government pays X, and you can’t price it 10 times lower for a commercial market. Technologically, the utility sector might want high-resolution data over a wide swath — but constellations are being built with what the defense industry wants in mind: which sensor, which orbit, which configuration. What gets pushed into the commercial world is “try to make it work,” shoehorning the pieces together. Government wants as much as possible and will pay for everything; commercial has an ROI principle and only pays for what adds value. So there’s a fundamental business-model mismatch. The Earth observation industry is mostly a government contractor.

Daniel: I never thought about them as the client — everything is built for them, and the rest of us shoehorn whatever we can get into our own workflows.

Aravind: Exactly. My thesis is that maybe in 10 to 15 years it can become mainstream. AI will be a big part of it, and I hope open data continues. People say free data stops people paying, but I’ve seen enough cases — like utilities — where people will pay when they need to see the asset at 10-meter resolution that free data can’t give them. The demand from insurance, utilities, agriculture, and finance isn’t going away; we just need a model that works for them.

A Utopian Vision — and What He’s Excited About

Aravind: My maybe-naive utopian vision is that the current moment — every country launching their own thing, not much open — is a blip, maybe three to five years. Then we start pooling resources. You can’t monitor wildfires, droughts, and floods properly as one country; you need real-time monitoring at a scale no single country can manage. We did exactly this for weather 40 years ago — countries agreed to split the monitoring. We have five or six GNSS systems even though we didn’t strictly need all of them, so maybe we’ll have 60 or 70 Earth observation constellations because sovereignty trumps logic. But if their data isn’t interoperable, you just get a fragmented system of systems. Bringing some governments together, like Copernicus did, would be great.

Daniel: I wonder if models and embeddings are part of the answer — access to the data through the model, rather than the raw pixels.

Aravind: I can definitely see that — you give the embedding out instead of the raw imagery. But we need to discuss whether embeddings are open or closed. Sentinel and Landsat are so widely used because they’re reliable, trustable, and open. If you’re going to trust a science or disaster-management decision, can you base it on a closed model? That’s an interesting discussion.

Daniel: What are you most excited about for the next 10 years?

Aravind: I’d be disappointed if the future of Earth observation is just being a government operator. What excites me is the newer missions — NISAR, the new Copernicus missions. Hyperspectral data was never free before; now Copernicus’s CHIME mission will provide it for free and unlock a new market. And on adoption: Earth observation is a technology where you can’t imagine all the end-user use cases yourself. The crowd mind is much better. Now someone can spend a weekend and build a farm-monitoring application without knowing remote sensing — the data is there, the model is there. It reminds me of 2008 when the app stores opened up and a flood of apps came in. We never quite got there with Earth observation because processing it was so hard. Now you can spin up a notebook, write your code, and get an application live. Maybe you’ll be able to check the wildfire or flood risk of a house before you buy it. That’s exciting — people will unlock applications that I, and a lot of satellite companies, never thought of.

Daniel: Aravind, thank you so much for coming on. I really enjoyed the conversation, and I hope we can get you back on at some stage.

Aravind: This was a great chat — maybe when the industry radically changes in the next five years. If it’s the same in five years, there won’t be enough to talk about. So let’s hope for radical change. Thank you, Daniel — it’s my pleasure.