Key Takeaways

  • Within a decade, most global compute will move off-planet, becoming a multi-trillion dollar industry larger than any other space business today, predicts Planet Labs CEO Will Marshall.
  • This shift is driven by economics: Starship will drop launch costs to an unprecedented $200-$300 per kilogram, making space data centers financially viable.
  • Solar panels in sun-synchronous dawn/dusk orbits harvest five times more energy than on Earth, requiring no batteries and dramatically cutting power costs.
  • Despite challenges in inter-cluster communication, early tests with tech giants like Google and Nvidia are paving the way for this extraterrestrial computing revolution.
  • Marshall envisions "planetary intelligence" fueled by "large earth models" – AI trained on vast, real-world satellite data to solve global problems with "gazillions of applications."

The Trillion-Dollar Bet: Compute's Exodus to Orbit

Will Marshall, the CEO of Planet Labs, made an audacious prediction: “I think no question within 10 years most compute will be putting in space.” This isn't just a bold statement, it's a direct challenge to the multi-trillion-dollar terrestrial data center industry. Marshall says this future space compute sector will become “bigger than any of the other space businesses today.” His reasoning is surprisingly simple: economics.

Marshall points to declining launch costs as the first domino. SpaceX’s Starship, he explains, is set to slash the price of putting objects into orbit to an unprecedented “$200 to $300 a kilogram.” At that price point, the math fundamentally changes. “It would be cheaper, just simply cheaper to put the data centers in space,” Marshall stated.

But it’s not just launch. Once in orbit, solar power becomes a game-changer. “In space you can put a solar panel in a suns synchronous uh dawn dusk orbit where you're 24/7 looking at the sun,” Marshall explains. This setup means a single solar panel can collect “five times more energy per solar panel than on the ground and you don't have to have batteries or anything else.” Imagine massive energy savings, zero real estate costs, and no cooling towers needed for your core compute infrastructure.

From Earth Models to Planetary Intelligence

The vision isn't just about cheaper compute; it’s about a new class of AI. Marshall calls them "large earth models," a direct analog to large language models, but fed with real-world satellite data instead of text. "If you give them real world data, then they can answer real world problems," he says. This, he believes, will "open up gazillions of applications for these uh AI models."

He even rebrands AI itself for this new era: “instead of AI, planetary intelligence.” This isn't just wordplay; it points to a future where advanced AI, fueled by continuous, global data from space, understands and predicts Earth-scale phenomena with unprecedented accuracy. Think climate modeling, resource management, and disaster prediction – all at a planetary scale.

Of course, challenges remain. Andrew Feldman from Cerebras highlighted one critical hurdle: “We we're not super good yet at building the clusters in space necessary for the communication between Exactly.” Marshall acknowledges this, but points to ongoing work and partnerships with tech giants like Google and Nvidia for early tests, suggesting solutions are actively being developed to overcome these technical barriers.

What to Do With This

  • Track your critical cost curves: What seemingly fixed cost in your industry is poised for a 10x or 100x reduction in the next 5-10 years? Identify the "Starship moment" that will make your core input radically cheaper, then brainstorm the new business models and applications that become viable only at that future price point. Don't wait for it to happen; anticipate it and position your venture for the shift.
  • Build “Large [Your Niche] Models”: Start experimenting with novel, real-world data sources specific to your domain. If Marshall sees "large earth models" unlocking "planetary intelligence," what unique, previously unfeasible dataset can you collect or synthesize to train a "large [your customer/product/industry] model" that delivers unprecedented insight and "intelligence" for your users or operations? The compute may not be in space yet, but the data opportunities are already here.