Topics Pioneers Insight Method Author
Back

Planet Labs Factory Tour

·Sourcery

Rendering the long image…

Topics Venture Capital Bets Big on Hard Tech Hard Tech R&D Discipline and Commercialization
Pioneers Will Marshall

Summary

  • Marshall’s pitch pairs a difficult satellite buildout with the margins of an information business. He says all 700 satellites built over 15 years cost less than one typical NASA satellite, representing about a 1,000X cost reduction. Planet incurs CapEx, then sells pictures and analytics at very high margins — “the best of both worlds.”
  • Daily global coverage is Planet’s central data advantage. About 200 satellites produce roughly 4 million 47-megapixel images daily across eight spectral bands; Marshall claims roughly 100 times the area collection of other Western commercial operators. The distinction is imaging “the whole Earth every day,” rather than tasking satellites to selected locations.
  • Planet’s production figures describe substantial capacity, with clear qualifications on utilization. Doves can reach 40 builds per week, and Marshall thinks each takes about eight hours. “We don’t do that every week,” he stresses; an announced Berlin facility adds capacity of up to 60 larger satellites annually.
  • Owl’s upgrade moves daily mapping from three-meter to one-meter resolution. That means almost 10 times more data per unit area, with optical links returning it 10 times faster, within a few minutes. Satellite count will remain about the same, if not decline slightly: “it’s not how many satellites you have, it’s what you do with them or something.”
  • Marshall considers Planet’s claimed 100% satellite utility a reason to take more technical risk. The claim applies when launch successfully delivered the spacecraft to orbit, and includes satellites with focus, solar-panel and reaction-wheel problems. His conclusion: “it means we’re not pushing the envelope enough.”
  • The AI interface promises broader access, while the live demo exposes unfinished execution. Marshall claims searches across the entire Earth in under a second, drawing on roughly 3,000 imagery layers spanning 10 years. The Area 51 demonstration did not trigger the intended search tool, however, and public access remains a beta — “just a testing version.”

Deep dive

1. Consumer electronics made a cheaper satellite business possible

  • Marshall puts factory output at about 700 satellites over 15 years, second only to SpaceX by his count. Planet launches four or five times annually and has used over 40 rockets. Earlier satellites could be bigger than a room and double-decker-bus-sized, costing at least $1 billion; some NASA missions cost up to $10 billion. His cost comparison is the provocation: all 700 cost less than a typical NASA satellite, representing about a 1,000X reduction. Smaller spacecraft let Planet launch enough units to attempt fundamentally different missions.
  • O’Shea’s question about putting a phone in space returns Marshall to a NASA colleague’s challenge: “Why are we spending a billion dollars on a satellite when a phone like this has most of what a satellite has, but it costs 500 bucks?” The suborbital test failed, yet the electronics survived and the data was recovered — evidence, for Marshall, of consumer hardware’s robustness.
  • The institutional change was to follow electronics investment rather than recreate it. NASA had developed components because suitable computers, batteries and solar panels were unavailable; now Apple, Google and others fund the dominant R&D. Planet adopted commercial parts, launched extras and accepted that “if one or two of the satellites breaks, no big deal.” Marshall compares the resulting expansion of applications to the transition from mainframes to desktops.
  • Commercial sourcing leaves substantial integration inside Planet: “the chips we buy, the systems we build.” Purchased telescope blanks become assembled optics; radio chips become custom circuits and antennas. Marshall calls Planet fundamentally a data business, selling pictures and analytics at very high margins. Satellites require CapEx and create a high barrier to entry: few companies launch them, and they are complicated. Sales margins “look like a software business.”

2. Daily global imaging depends on throughput and ground infrastructure

  • About 200 satellites image the entire Earth every day, Marshall says, seeing individual trees and houses from roughly 500 kilometers away. He says it is the only system that images the whole Earth daily. Output is about 4 million 47-megapixel images, “on the order of 100 terapixels,” with eight spectral bands spanning optical and near-infrared wavelengths. Visual AI can recognize buildings, ships and planes, making imagery interpretable at a scale humans cannot manually review.
  • The first spacecraft downlinked around one megabit per second; current radios reach two gigabits per second over roughly 1,000 kilometers. Marshall describes roughly 1,000 times more data per satellite per unit time, enabling daily global imaging. GPS, star cameras and pointing metadata make the pictures locatable, so Planet can stitch individual observations onto the Earth’s surface.
  • That throughput required a ground network Planet also built: about 48 dish antennas across a few dozen locations, often with four or eight at one site. Alaska, northern Canada, northern Norway and southern New Zealand and Australia supply more contact opportunities for mainly polar orbits. The constraint is practical: go as far north or south as possible “with an internet connection.”
  • The telescope assembly is about 50 centimeters; telescope size and satellite altitude determine the smallest ground object visible. Pelican-10 is one of Planet’s highest-resolution satellites, and Pelicans fly low to improve resolution. Specialized, hyperspectral Tanagers are being developed with NASA’s JPL; Marshall says they help track ecosystems, and the second launched “last week.” Showing a methane plume, he explains the intervention: tell the gas company about its leaking pipeline. Turning off that one large leak would equal taking a couple hundred thousand cars off the road, he says.

3. The next fleet upgrade targets data density and delivery speed

  • O’Shea asks whether constellation growth is expansion or replenishment. “It is replenishing now,” Marshall answers; capability matters more than count. He claims the largest Earth-imaging fleet, roughly 10 times competitors’ size and about 100 times the area collection of other commercial operators, qualifying that to Western ones. Others task selected locations and cover less than 1%, or at most a couple percent, of Earth’s land daily.
  • More than 650 Doves have been built in the lab. Marshall puts capacity at up to 40 per week: “We don’t do that every week.” He thinks one takes about eight hours, versus his NASA comparison of “8 years, not 8 days.” Larger spacecraft take longer; an announced Berlin facility doubles manufacturing space again, with capacity of up to 60 larger satellites annually.
  • Owls take the daily scan from three-meter to one-meter resolution: roughly three-by-three, or almost 10 times more data per unit area. Adding optical links brings data back 10 times faster, within a few minutes. Satellite count will stay about the same, if not decline a little. Marshall’s summary: “it’s not how many satellites you have, it’s what you do with them or something.”

4. Marshall wants more experimentation despite a 100% utility record

  • Tests reproduce 15 Gs of launch vibration, roughly 200 Gs of separation shock and temperatures swinging from minus 40°C to plus 20°C every 45 minutes. Broken telescopes have forced redesigns: focus depends on alignment within a fraction of a wavelength, while ordinary aluminum expands and contracts too much. Special materials compensate. Shock, vibration and thermal vacuum are the main checks — or, less formally, “vibrate the shit out of it, and hopefully nothing falls apart.”
  • The reliability claim has an explicit boundary: if launch delivered the satellites to orbit, every one turned on and worked, with utility from 100%. That includes out-of-focus optics, undeployed solar panels and early reaction-wheel failures. O’Shea asks why he considers the record a problem. “It means we’re not pushing the envelope enough,” he answers; perhaps they should have tried a newer sensor that would improve capability.
  • O’Shea raises Starlink’s proposed sharing of collision-avoidance data; Marshall welcomes it and describes cooperation among operators. His larger concern is debris: of order 10,000 operating satellites against of order 100 million debris objects. Breakups, crashes and anti-satellite weapons can turn one spacecraft into tens of thousands of fragments.

5. The AI interface opens the archive while exposing early limitations

  • Planet’s interface sits over roughly 3,000 imagery layers spanning 10 years, almost daily. Marshall frames the combination as an LLM’s general knowledge connected to observations of the real world. A conventional chatbot can explain agricultural theory but “doesn’t know shit about your field”; a farmer wants information about that specific field and its crop yields. Journalists similarly need evidence about an actual flood or fire.
  • Marshall says the interface can search the entire Earth in under a second, compare changes and plot graphs. Asked earlier what changed in 24 hours, he says it found a Yosemite fire and flooding that washed material into a river. That result is on another computer, rather than shown live. His framing is a “daily planet report.”
  • The live Area 51 request compares 2021 with 2026. Marshall points to aircraft that disappeared between images and a structural change he cannot identify: “I actually don’t know what that is.” Zooming proves awkward, and afterward he acknowledges, “I don’t know why it wasn’t doing the search tool.” His assessment: “It doesn’t do a perfect job, but it’s getting better and better.”
  • The intended audience extends beyond NASA and the NRO to farmers, journalists and anyone who can ask a question in any language. A couple of engineers and a specialized AI team built the interface. Anyone can access the beta, Marshall says, but it remains “just a testing version”; he hopes to launch it soon.