Satellite Imagery is Not Becoming a Commodity
Disclaimer: I work at Umbra, a satellite imagery company. I’m hopelessly biased on this topic. As Upton Sinclair famously quipped, “It is difficult to get a man to understand something, when his salary depends on his not understanding it.” And please don’t confuse the personal opinions expressed below with those of my employer!
P.S. I originally published this nine whole months ago on my newsletter. If you like this blog and want to hear my most recent thoughts you can sign up for that here).
Five years ago, a gaggle of VC-backed startup CEOs sat on conference stages in blazers and t-shirts smugly parroting each other with lines like, “Satellite imagery is quickly becoming commoditized. Analytics built on top of satellite imagery is the real opportunity.”
In the years since, a lot of those CEOs have lost their job or their company or both. Most of the high profile CEOs from the ~2014–15 golden age of geospatial analytics have either stepped down or been forced out.
To be fair, by the end of 2016, I was drinking the same Kool-Aid as the rest of ’em. We all thought we were on the cusp of a revolution in the commercial earth observation industry. 2017 was going to be the big breakout year when a flood of cheap, awesome, civilian satellites would finally hit the market. All the charts looked like hockey sticks back then, but I never stopped to wonder why the axes weren’t labeled…
☝️ Thank you, BlackSky, for the pity “like.” I owe you one.
Planet was on track to hit “Mission One” in 2017 (they did) and the second Sentinel-2 bird was planned for launch (it did). AstroDigital launched its first satellites. A record 104 satellites were launched in a single day that year (the majority of them remote sensing satellites) and Planet even acquired TerraBella out of the jaws of Google where it wasn’t doing anything for anybody.
Fast forward to today and we’re still “on the cusp of a revolution in the commercial earth observation industry.” Maxar and Airbus are launching big constellations this year, Planet continues to add to the largest constellation of earth observing satellites ever assembled, Satellogic and BlackSky are nipping at their heels with big globs of money to spend, and of course the SAR Wars are heating up.
As exciting as that all sounds, very little has *actually* changed for consumers of satellite imagery. It’s still ridiculously difficult and expensive to buy high resolution, tasked imagery. Investor expectations are at all-time highs as reflected in the valuations of growth stage companies going public via newfangled Special Purpose Acquisition Companies (SPACs) at generous multiples of projected EBITDA five years in the future.
It feels a little like 2016 all over again. If this time is going to turn out differently, it’s going to take more than just money (although the money will certainly help). It will require delayed gratification and very longterm thinking from the industry as a whole. We have to think of this as a positive-sum game or else we’ll all be chasing bigger and bigger slices of a pie that stubbornly refuses to grow.
And even when we’re successful at making high resolution, timely satellite imagery affordable and accessible to a broad audience…it won’t be a commodity product. At least, not in my lifetime.
If Satellite Imagery Were Becoming Commoditized, You Could Buy It
I was recently listening to Aravind Raves’ fantastic podcast that aims to “demystify space technology.” His guest took a brief respite from flogging his “space manufacturing” startup to make a lazy argument I’ve heard people espousing for the last five years: earth observation is a crowded, low-margin market of increasingly undifferentiated companies.
If it’s such an efficient market with limited prospects for disruption (and therefore outsized returns), tell me…how do I go about buying a satellite image today? How much does it cost? How long will it take for my order to be filled? How will the data be sent to me? Let me take a crack at it:
- Hopefully you have already established close relationships with sales people at one of the major satellite imagery suppliers. If not, you’re gonna have a bad time. Either way, get ready to send some emails.
- Price depends on the company, last quarter’s sales numbers, this quarter’s sales numbers, the clout of the sales person, and the astral alignment of your horoscope. You’re about to spend anywhere between a few hundred dollars for an archival image to several thousand for a tasked image. That is, if they let you buy anything at all.
- I would budget a few weeks to a few months for this adventure if I were you.
- Are you familiar with File Transfer Protocol?
The truth is that supply is still very constrained at the highest resolutions, and that’s mostly what people are actually willing to pay for. Despite the fact that lots more satellites are getting launched this year and next, it’s not at all clear to me that the cost basis of most of those constellations will allow them to monetize their imagery at a significant discount to what’s currently being offered (especially if international governmental demand for that imagery continues to outpace production).
Not to mention it takes a special kind of investor to listen to a CEO pitch them on the idea of leaving millions on the table in the short term in order to grow the market over the long term. To the companies going public and joining the quarterly earnings rat race: I salute you. Godspeed.
So, yes, the cost of launch is going down. And so is the cost of manufacturing. But the cost isn’t, like, zero. These are still big-ass hunks of metal that have to be flung into space on the nose of even bigger-ass hunks of metal and then kept there, against the pull of gravity, for years at a time. Space-based Earth observation is a capital-intensive business where even the very best inevitably lose contact with (or control of) their assets in space. Those costs all get factored into the price of an image.
I personally did not appreciate how difficult it is to build and launch satellites before joining Umbra. It’s one thing to talk about it, and a totally different thing to feel the pressure of delivering on it. Every time I see a successful launch, I don’t feel envious — I feel excited. With each successful mission, we take one baby step closer to a competitive, efficient market. We’re all walking a tightrope of technical risk where the line between financial reward and ruin can come down to a single roll of the dice.¹
Satellite Imagery Suppliers are Like Dairy Farms
I grew up a few miles from a small dairy farm in North Carolina, Maple View Farm. When I was a kid, my Mom would schlep me up the long gravel driveway to the Nutter family’s farm house where a refrigerator waited, sweating in the summer heat, full of the most incredible chocolate milk you can imagine. The stuff is so thick you’d break your neck diving into a pool of it.
The Nutters employed the honor code back then — you dropped your cash in a basket, took your glass jugs of milk out of the fridge, and then returned your empties the next time you visited.
Later on, Maple View opened an ice cream parlor at the top of the hill overlooking the farm. There is no better ice cream on the planet. You can sit on the front porch of the ice cream parlor in one of their rocking chairs and look out across cornfields and pastures, waffle cone in hand, ruminating on everything and nothing. Maple View does not just sell milk and ice cream; they also sell peace and stillness.
Anyway, you probably see where I’m going with this. Today’s satellite imagery companies are a lot more like Maple View Farm than they are like Dairy Farmers of America.²
A big part of the reason satellite imagery is not a commodity is that, unlike raw milk, it’s pretty damn hard to figure out how to produce. This is one of those rare cases where the phrase, “It’s not rocket science,” can’t be employed, because it literally is rocket science.
To extend the analogy, let’s start by imagining satellite imagery is hot milk straight from the teet. Analytics are refined dairy staples, like milk and butter. And software to interpret those analytics is ice cream. The best ice cream isn’t all about the dairy that goes into it; it’s at least as much about the other ingredients you mix in. But the quality of the dairy still determines the quality of the ice cream. Maple View makes the best ice cream on the planet in large part because they also make the best milk on the planet. That same dynamic is why satellite imagery providers who build out analytics teams will always have the upper hand on their customers and partners when competing head-to-head.
Like Maple View, satellite imagery providers are happy to sell their excess inventory to customers who subsequently create competing products. The world is a vast place and it’s not a zero sum game; there’s more demand for milk than Maple View can squeeze from its prodigious heifers. But if one of those customers were to start crowding into Maple View’s territory with their own ice cream parlor franchise…or if, suddenly, one of their customers launched a new product that became extremely popular…Maple View could always just cut them off or raise prices or undercut them with their own copycat product, or some combination of the three. After all, they own the cows. Go raise your own if you don’t like it.
Not All Satellite Imagery is Created Equal
Even if access to satellite imagery had gotten significantly easier over the last five years (it didn’t), the idea that satellite imagery was becoming commoditized would still have been wrong. I’ve heard people employ the phrase, “a pixel, is a pixel, is a pixel,” to color the idea of satellite-imagery-as-commodity. It sounds reasonable enough — it shouldn’t matter which company produces the pixels as long as they meet the spatial, spectral, and temporal requirements of the customer.
This past week on Twitter has been a great opportunity to compare apples to apples. Everybody and their cousin scrambled to get a shot of the Ever Given, the massive ship stuck sideways in the Suez Canal gumming up world commerce. Here’s a great tweet from my friend Ignacio (go follow him on Twitter) with a few such images conveniently placed side-by-side for easy comparison:
Look at how different they all look! Compare those shots to this one from Maxar:
Something you should know, too — anything that makes it to social media has been tortured to death to make it as high-contrast, low-blur, Instagram-worthy as possible. This is what everyone looks like at their best.
Among the providers referenced above, Satellogic wins “most improved” in my book. But Maxar and Airbus, the big incumbents, still have the goods.
For this type of use case (is the big ship still stuck in the tiny canal?) a pixel is a pixel, sure. But if subsequent questions come up like, “how many excavators do they have on site?” or, “Which class of tugboat are they trying to pry it loose with?” suddenly the imagery doesn’t seem so fungible.
In my opinion, the uses for satellite imagery with the most untapped potential tend to be about monitoring change. Any time you’re dealing with a time series of satellite images, you wind up facing down some of the more esoteric characteristics of satellite imagery, like absolute positional accuracy, co-registration, sensor calibration, and incidence angles.
In fact, there’s a whole subdomain of remote sensing research dedicated to the pursuit of making satellite imagery “analysis ready,” and it’s a concept that prominent geospatial analytics firms like Descartes Labs have more or less staked their brand on.
Just like store-bought milk has to be pasteurized and packaged and stored at a steady temperature below 40ºF, satellite imagery must be processed before it’s ok to ingest. And even at that point it’s pretty boring…like a glass of milk. Plus, just like no amount of processing will turn goat’s milk into cow’s milk, no amount of processing will turn low resolution data into high resolution data or one phenomenology into another.³
Less of an Oddity, but Still Not a Commodity
There’s a saying in finance that goes something like, “Being right about something but getting the timing wrong still just means you are wrong.”
I personally believe that high resolution, low latency, low cost satellite imagery is finally going to become available to civilians in the immediate future. I am working every single day to help make that happen, alongside a cadre of brilliant, highly motivated colleagues (not to mention a growing list impressive partners, some of whom are also launching satellites).
Truthfully, I can’t help but harbor more than a little fear that I am falling victim to the same delusions that blinded me in 2016. A handful of people much smarter than me have told me point-blank that the approach we’re taking at Umbra is naive and misguided. I’m not worried about being embarrassed if it doesn’t work out, but I don’t want to feel like a fraud, either. I am determined to prove them wrong.
Luckily, I’m not in this thing to get rich quick or help flip a company in a few years. For one thing, I am 100% convinced that satellite imagery is an irreplaceable tool in the fight against climate change, which is the central issue that will define my generation’s paragraph in the annals of history.
How else, besides using satellite imagery, can you:
- Verify human activity and natural capital on a global scale, even at low fidelity?
- Characterize the extent and severity of natural disasters as they unfold in real time anywhere on Earth?
- Monitor conditions on the ground in places made inaccessible due to geography or political regime?
- Inventory land use and track its changes over time?
How can we make planetary-scale decisions unless we have planetary-scale data to inform them? I don’t think satellite imagery is the answer to the climate crisis, but I think it’s going to be a necessary ingredient of any answers that emerge over the coming decades.
Even when we make it up onto the mountain top where high resolution, timely satellite imagery grows on trees…that doesn’t mean it’s going to turn into a commodity business. Some systems will be more suited to monitoring and others to foundational mapping; some to long-term change detection and others to instant verification; some to classifying stuff and others to counting things.
There will be visible light sensors, and infrared sensors, and ultraviolet sensors, and hyperspectral sensors, and radar sensors, and on and on. None of them will be “analysis ready” without a tremendous amount of specialized work, and the synthesis of these data into novel information will be an active field of research for as long as I’m living and long after I’m gone. Newer, better, smaller, cheaper, tougher, smarter systems will continue to be developed ad infinitum, and someone will always have a technology edge, however briefly.
So next time you hear someone say that satellite imagery is becoming commoditized, try asking them, “Oh yeah? What about Maple View Farm?”
¹ I got the chance earlier this year to interview Walter Scott, the founder and CTO of Maxar. Refining that conversation into a blog post will take a while; I didn’t anticipate starting a new job right around the same time I interviewed him. But one of the big takeaways from that opportunity was just how close Maxar came to insolvency. Their first two satellites failed. By the time they were on the platform counting down to launch for their third bird, ten years had already passed and $42M had gone up in smoke. They were hundreds of millions of dollars in debt. If QuickBird had blown up in the sky in October of 2001, the largest and most successful American satellite imagery company today would likely not exist. But it didn’t blow up!
² The largest dairy producer in the United States is a farming co-operative comprised of thousands of independent producers. I like my milk like I like my Scotch: blended.
³ Unless you are just having fun, stop trying to use style transfer to make Planetscope look like SkySat or make SAR images look like optical images. It’s a fun party trick, but there are so many more interesting applications of machine learning that feel under-explored, like replacing traditional interferometry methods with deep learning or using GANs to make seamless mosaics out of patchwork to bring the cost of producing basemaps down.