Topics Pioneers Insight Method Author
Back

Planet Labs CEO: Why AI Data Centers Will Move to Space

·Sourcery

Rendering the long image…

Topics Specialized Edge AI Reshaping Defense Tech Next-Gen Disruptive Computing Architectures AI Data Center Power Bottlenecks
Pioneers Will Marshall

Summary

  • Planet reported 58% year-on-year growth last quarter, and Marshall thinks defense supplied the largest share. The company has doubled manufacturing capacity in San Francisco as countries seek satellite data, AI analysis and their own spacecraft. Its daily scan supplies “peripheral vision” beyond intelligence targets governments already know to watch.
  • Sovereignty is turning imagery customers into buyers of nationally controlled satellite fleets. Marshall cites Japan, Germany and Sweden, with Sweden’s first launch four months after signing and a Germany-sponsored launch in two months, versus the industry’s usual five or 10 years. Customers’ demand was explicit: “We really want the satellites so that we have sovereign control of them.”
  • Marshall expects AI to expand the audience for Planet’s data while leaving its core business largely intact. O’Shea suggests another business model; Marshall counters, “we were mainly selling data anyway.” The beta app lets people ask questions in ordinary language, opening imagery to farmers, firefighters and local governments without specialist analysis teams.
  • Planet’s product stack combines daily coverage, targeted imaging and spectral detail. Doves scan about 200 million square kilometers of landmass at three-meter resolution, plus ocean territory; Pelican’s goal is a requested image returned and analyzed within 30 minutes anywhere on Earth; Tanager carries 420 spectral bands. Marshall’s methane example captures the value: “you can literally see the gas,” enabling an operator to stop a leak and save money.
  • Planet’s orbital computing work already includes a Google test project and NVIDIA chips aboard Pelican spacecraft. Marshall says the Suncatcher test project launched “last week”; the spacecraft is commissioned, with TPU testing still to come. Processing images in orbit and sending down answers aims to replace hours of delay with minutes, potentially improving wildfire response.
  • Marshall’s case for moving AI data centers into orbit rests on continuous solar power and simpler energy infrastructure. He claims “five times more power” per panel in a continuously sunlit orbit, removing the need for batteries or backup generation, with radiators cooling the chips. His economic prediction is categorical: “it’s all just going to be cheaper.”

Deep dive

1. Defense demand is scaling Planet’s information business

  • Marshall calls Planet “an information business”: automated manufacturing and shrinking satellites from “the size of a double-decker bus” enabled more launches at lower cost. Planet says its fleet was the first to image the entire Earth every day and grew beyond 100 satellites, though Starlink is now larger. Planet doubled San Francisco manufacturing capacity as demand rose. It reported 58% year-on-year growth last quarter; “I think the biggest part of that was the defense growth,” he says.
  • Ukraine illustrates the demand mechanism. Against Russia’s greater troop and equipment resources, Marshall says Ukraine has an “information advantage,” using satellites, drones and AI to monitor threats well before they reach its borders. Other countries see those applications and want similar capabilities. Planet supplies “peripheral vision”: finding emerging threats beyond facilities intelligence agencies already know to watch.
  • The US Navy uses Planet to monitor 13 million square kilometers of the South China Sea, sharing information with regional allies. Admiral Paparo also monitors China for threats, including industrial facilities that might supply it. The objective is a week or months of warning rather than a day, creating room for preparation, deterrence and diplomacy. Marshall’s stated mission is to reduce the likelihood of war.
  • Commercial customers include John Deere, Bayer and BASF improving crop yields, hedge funds seeking “alpha on stocks and commodities,” and hyperscalers monitoring both their own and competitors’ data center construction worldwide, including China. Civil uses include wildfire response, NASA science and Brazil’s deforestation monitoring. Those businesses are also growing, Marshall says, but defense is growing fastest.

2. AI makes a layered imaging system usable by more people

  • Planet’s sensor stack separates the daily scan from customer-directed imaging. Doves cover all Earth’s landmass—about 200 million square kilometers—and some ocean territory, in eight spectral bands at three-meter resolution. Pelicans are its highest-resolution satellites, photographing requested locations. The fleet being built aims to accept a request and return an analyzed image anywhere on Earth within 30 minutes.
  • Tanager’s 420 spectral bands provide a “fingerprint” of what is on Earth, helping distinguish tree and crop species, identify camouflage and detect methane. Marshall’s concrete example is a gas leak: although methane is colorless, “in this spectral band it sticks out like a sore thumb.” Operators can stop the leak, cutting pollution and saving money.
  • The beta AI app addresses the gap between collecting imagery and using it. NASA has 1,000 people who know how to analyze satellite data; farmers and local journalists do not. Marshall’s analogy is a computer gaining an operating system: suddenly many more people can use it. Users can “ask any question in simple language,” and the tools “will have a go.”
  • O’Shea’s suggestion that AI adds another business model gets a qualified rebuttal: “it’s not changing the business model that much because we were mainly selling data anyway.” Planet is adding analytical layers it calls “solutions,” while the more flexible tools widen access. Compute runs mainly on Google Cloud, with some AWS.

3. Space computing starts with faster answers and continuous power

  • Asked whether orbital data centers are rational, Marshall says Planet launched Google’s first “Suncatcher” data center test project “last week,” after previously putting NVIDIA chips on Pelican spacecraft. The Google spacecraft is commissioned and doing well, he says. Testing how the TPU performs in space is still ahead; Planet will report more as that work proceeds.
  • The immediate reason for onboard computing is speed: process imagery, recognize ships or run queries in orbit, then send down answers. During the LA fires, Planet captured imagery within a couple of hours, downlinked it a couple of hours later and needed perhaps another hour for processing. Getting information to responders in minutes instead could potentially help save lives.
  • Marshall’s broader argument begins with data centers being “all about energy.” Ground solar offers the cheapest watts per dollar, but clouds and nighttime require other power sources; adding gas or nuclear brings permitting and delay. In a sun-synchronous orbit that always looks toward the sun, he says each panel gets “five times more power,” with no interruption from clouds or night and no need for batteries, gas or other power systems.
  • Cooling requires radiators facing the darkness of space; commands and answers travel by radio, with a little latency. Marshall concludes orbital computing “is all just going to be cheaper” and expects it to ease community objections over electricity costs and water use. He sees compute as the “by far easiest” first step, preceding more ambitious ideas such as generating power in space and beaming it down.

4. Sovereign control turns data demand into spacecraft orders

  • Marshall connects sovereignty demand to geopolitical fragmentation and US pressure for other countries to shoulder more defense responsibility. Their first need is understanding nearby threats: satellites are their “eyes and their ears.” He says satellites also underpin missile warning and civilian and economic systems. Planet initially offered countries data, but customers insisted on buying satellites for sovereign control. Japan is buying satellites, Germany is funding another fleet, and Planet is also supplying Sweden.
  • Planet builds and launches the satellites, then “hand[s] over the keys” so customers control their operation and pointing. Sweden’s first satellite launched four months after contract signing; a Germany-sponsored satellite launched in two months. Marshall contrasts that with the five or 10 years such projects would normally take in aerospace.
  • Access has limits. Marshall cites US-embargoed countries, equivalent EU lists and UN restrictions, and says Planet does not sell to terrorists or North Korea. An employee group also reviews deals for alignment with Planet’s values and likely positive outcomes. “We know our technology is very powerful,” he says, making stewardship part of the sales decision.

5. Protecting “spaceship Earth” anchors the long game

  • Marshall describes the universe as “eerily silent” despite more than 6,300 discovered planets in the local galactic neighborhood, with no obvious signs of life found yet. He predicted life beyond Earth within 10 years a few years ago; now it is “probably more like six or seven.” Atmospheric signatures are the mechanism: Earth’s 20% oxygen is his example. Such evidence could suggest life without certainty or proof of technological civilization.
  • The absence of technological signals raises one worrying possibility: civilizations might destroy themselves after developing powerful technologies. Marshall cites nuclear weapons and AI, including the possibility of misuse, hacking and unresolved control risks. That possibility makes protecting “spaceship Earth,” the only place where life is known to exist, “not just locally important, it’s cosmically important”—a “cosmic duty.”
  • Marshall credits “persistence and endurance” above timing or the best idea, while conceding those matter. At NASA, Pete Worden shielded the team’s experiment to “throw some phones into space and see if they worked,” which led to Planet. For evidence of commitment, Marshall cites retention: seven of the first 10 employees remained at the 10th anniversary; about half were still there at the 15th.