The Data Center That Doesn't Need Cooling, Power, or a Building — Explained

[00:00:00] Speaker 1: Picture the three biggest problems standing in the way of every AI company on earth right now. Data centers need staggering amounts of electricity, and the power grid cannot keep up. Data centers need constant, aggressive cooling, and that usually means millions of gallons of water. And data centers need physical land, permits, and local approval, and more and more communities are saying no. Now imagine a data center that sidesteps all three of those problems simultaneously, one that never asks the grid for a single watt, never touches a drop of water, and never needs a zoning permit from a single human being because it is not sitting in a field in Virginia or Texas. It is sitting in orbit, 350 miles above your head, circling the planet every 90 minutes. This is not science fiction anymore. Companies with real funding, real hardware, and real satellites already in space are betting that the future of AI computing does not belong on the ground at all. By the end of this video, you're going to understand exactly how a data center in space actually works, why some of the most serious investors in the world just poured hundreds of millions of dollars into the idea, and why other credible experts think the entire concept might not add up the way its boosters claim. Let's start with the crisis that makes an idea this radical even worth taking seriously in the first place. Because you cannot understand why anyone would try to launch a data center into space without first understanding just how badly the ground-based version of that same data center is currently straining under its own weight. The artificial intelligence boom of the past few years has created an appetite for computing power that is growing faster than almost anything the electricity grid has ever had to absorb. Training and running today's largest AI models requires enormous clusters of GPUs, and those clusters draw electricity on a scale that used to be reserved for entire cities, not single buildings. Modern AI server racks can now pull somewhere between 40 and well over 100 kilowatts each, compared to the 7 to 15 kilowatts that traditional data center racks were originally designed around. That is, not a small, incremental jump. It is a complete rewrite of what a data center's electrical and cooling systems need to handle, and most of the buildings racing to house this new generation of hardware were never designed for loads anywhere close to this intense. That electrical crunch has already started colliding with something even harder to solve than engineering, and that is local politics. In Dalhurst, just the first few months of this year alone, more than 75 separate data center construction projects, worth a combined $130 billion, were successfully blocked by local opposition across the United States, driven by residents and officials, worried about soaring electricity costs, strained water supplies, and grids that are already stretched. Then, interconnection queues, the waiting lists utilities used to hook new facilities up to the power grid, now stretch out past the year 2030 in some regions. Put simply, even if a company has the money and the hardware to build a massive new data center today, getting that data center actual electricity and actual community approval has become one of the single biggest bottlenecks in the entire AI industry. This is the exact pressure point that made a handful of engineers and executives start asking a genuinely strange question. What if you built the data center somewhere nobody could block it? Somewhere the sun never stopped shining, and somewhere there's no water table to worry about at all? Let's talk about the company that has pushed this idea further than almost anyone else. Because a Washington-based startup called StarCloud has gone from a rebranded idea to an actual operational satellite faster than most people in the space industry expected. Originally founded under the name LumenOrbit before rebranding in early 2025, StarCloud has already raised $200 million in total funding, including a $170 million round earlier this year that valued the company at $1.1 billion. In November of last year, the company launched its first satellite, StarCloud One, carrying an NVIDIA H100 GPU, the same class of chip that powers a huge portion of today's most advanced AI systems into low Earth orbit. This marked the first time a data center class GPU had ever operated in space. Weeks later, the team went even further, running Google's Gemini model directly on that orbital chip and separately training a small language model called NanoGPT, built by OpenAI co-founder Andrej Karpathy, using the complete works of Shakespeare as training data, all while the satellite was circling the Earth roughly 17 times a day. Now, let's actually unpack the specific claims sitting inside this video's title, because "no cooling, no power, and no building" sounds almost too clean to be real, and it deserves a genuinely careful, honest explanation rather than just repeating the marketing pitch. Let's take these one at a time, starting with power, because it is the most straightforward of the three. On the ground, a data center's electricity has to come from somewhere: a coal plant, a natural gas plant, a nuclear reactor, a solar farm limited by weather and nighttime, or some patchwork combination of all of them, delivered through a strained, aging grid. In orbit, especially in a specific kind of path called a sun-synchronous orbit, a satellite can be positioned so that it stays in continuous, uninterrupted sunlight for the vast majority of its journey around the planet, essentially never experiencing a real nighttime at all. Google explained this, advantaged directly in its own research blog, noting that "In the right orbit, a solar panel can be up to eight times more productive than the same panel sitting on the ground, and can produce power nearly continuously, dramatically, reducing the need for the heavy, expensive battery systems, ground-based solar farms depend on to get through the night." Starcloud's own pitch echoes the same. Math, claiming its satellites, can achieve energy costs roughly 10 times lower than an equivalent terrestrial data center, purely because of how much more efficient solar power becomes once you take Earth's atmosphere and rotation out of the equation entirely. Now, let's get to cooling, because this is the part of the story that genuinely surprises people, and it is also the part that deserves the most careful, precise explanation, because the popular version of this claim tends to oversimplify what is actually happening. It is tempting to imagine the vacuum of space as some kind of giant, automatic refrigerator instantly whisking heat away from anything sitting inside it. That is not quite how it works, and it's worth being precise here, rather than letting the flashy version of the claim go unchallenged. Space is actually a surprisingly difficult place to shed heat, because a vacuum contains no air and no water for heat to convect into the way it does inside an ordinary building here on Earth. The only way heat can escape a satellite is through radiation, infrared energy leaving the spacecraft's surface and traveling out into the emptiness of space. What space-based data center advocates are actually pointing to isn't some magical, effortless cooling effect. It's the fact that this radiative process eliminates the specific, brutal, terrestrial bottleneck that has become such a flashpoint in local politics: water. Ground-based AI data centers, especially ones using modern liquid cooling systems, can consume millions of gallons of water every single day to keep GPU clusters from overheating, and that water usage has become one of the single most common reasons local communities push back against new data center construction. A satellite, cooled by carefully engineered radiator panels facing away from the sun, needs zero water, zero cooling towers, and zero complex mechanical HVAC systems humming away inside a building. That is the real claim being made, and it is a genuinely accurate one, even though it requires more nuance than the headline version usually gets credit for. And finally, let's talk about the building itself, because this is where the claim becomes almost self-evidently true in a way that doesn't require much unpacking at all. A terrestrial data center is, at its core, a requiring land acquisition, zoning approval, local permitting, foundation work, structural engineering, and years of construction before a single server ever gets powered on. A satellite skips every single one of those steps entirely. There is no building to construct, because the spacecraft itself, its structural frame, its solar arrays, its radiator panels, and its onboard chips is the entire facility. There is no zoning board to convince, no local water utility to negotiate with, and no neighborhood association filing a formal objection at a city council meeting. The rocket that carries the satellite into orbit effectively replaces years of ground-based permitting and construction with a matter of minutes spent reaching low earth orbit. If you're the kind of person who wants to actually understand the mechanics behind a headline like this, instead of just the flashy version, go ahead and hit that like button right now, because breaking claims like this down carefully is exactly what keeps this channel worth coming back to. Now, let's talk about who else has jumped into this race, because StarCloud is far from alone in chasing this idea, and understanding the other major players helps explain why this concept has suddenly gone from a fringe idea to a genuinely serious industry conversation, happening at the highest levels of some of the world's biggest technology companies. In November of last year, Google announced its own moonshot effort called Project Suncatcher, aimed at eventually launching solar-powered satellite constellations carrying the company's own custom AI chips called Tensor Processing Units. Google's first Concrete Step is a demonstration mission planned for early 2027, launching two prototype satellites specifically designed to test whether its chips can actually survive the harsh radiation environment of space over an extended period. Google's own blog post laid out the company's long-term thinking plainly, stating that, in the future, space may be the best place to scale AI computing power. A remarkable statement coming from a company that already operates some of the largest terrestrial data center campuses on the entire planet. Then there's Elon Musk's SpaceX, which, true to form, is pursuing what might be the single most ambitious version of this idea anyone has proposed. The company has formally applied for approval to eventually launch a constellation of up to one million satellites specifically designed to deliver orbital compute power, each one contributing roughly 100 kilowatts of compute capacity per metric ton of satellite mass. SpaceX made its rationale explicit in its own regulatory filing, arguing that the constant uninterrupted power generation available in a sun's synchronous orbit would reduce the need to rely on batteries and power cycling, making these orbits ideal for the kind of energy, intensive AI workloads that demand steady, reliable compute capacity around the clock. No specific launch date for that plan has been disclosed yet, but the sheer scale of the proposal, a million satellites, gives you a sense of just how seriously at least one major player is taking this idea, even if it currently sounds closer to science fiction than a near-term deployment plan. It's not just the biggest names in tech chasing this idea either. And understanding the wider field of players makes the whole picture feel considerably less like a single company's marketing pitch, and more like a genuine emerging category. Axiom Space has already flown a data center prototype aboard the International Space Station itself, successfully running cloud computing, AI, data fusion, and even cybersecurity workloads directly in orbit. And the company has since announced plans to deploy a dedicated orbital data center node to the space station by 2027, backed in part by funding from the Texas Space Commission. Meanwhile, a company called Lone Star Data Holdings is pursuing an even stranger variation on the same idea, targeting data center deployments on the surface of the moon itself, aimed less at everyday AI workloads, and more at ultra-secure off-planet disaster recovery storage, the kind of backup system designed to survive almost literally anything that could ever happen on Earth. Now let's talk about the actual economics of this idea, because this is exactly where the story stops being pure, uncomplicated excitement and starts requiring some genuinely serious scrutiny. And a responsible breakdown of this topic has to sit with that scrutiny just as carefully as it sits with the exciting engineering details. Andrew McCallop, an aerospace engineer and head of research and development at Varda Space Industries, ran the actual numbers on this question, and his analysis throws a considerable amount of cold water on the simplest version of the pitch. Andrew McCallop argued that the entire premise needs to be judged on a very specific, unforgiving standard, not just whether energy in space is theoretically cheaper, but whether the full, real-world cost per usable watt of orbital computing power can actually beat the equivalent cost on the ground. In his own words, orbit has to win on cost, or it has to admit it's doing something else entirely. Running the full math, accounting for everything upstream of the actual compute hardware itself, McCallop calculated that building one gigawatt of orbital solar-powered computing capacity would cost roughly $51.1 billion, compared to just $15.9 billion for the equivalent capacity built here on Earth. That is not a small gap. That is more than a three times cost disadvantage for the orbital approach, at least using today's launch costs and today's technology. Here's where the counter-argument comes in, though, and it's worth taking just as seriously as McCallop's skepticism, because this entire debate ultimately comes down to a bet about the future, rather than a simple, settled fact about the present. Google has openly acknowledged that current launch costs remain a genuinely significant obstacle to this whole idea working economically. But the company has also argued that if the cost of launching things into space continues falling at the same sustained rate, it has been falling for the past decade, driven largely by reusable rocket technology from companies like SpaceX, then the total cost of launching and operating a space-based data center, could eventually become roughly comparable to the reported energy costs of an equivalent data center here on Earth, measured on a per kilowatt per year basis. In other words, this entire idea is fundamentally a bet on a specific trendline continuing that rockets keep getting cheaper faster than power, and permits keep getting more expensive and more difficult on the ground. Nobody, including the companies pouring hundreds of millions of dollars into this idea, is claiming that crossover point has already arrived. They're betting on when it arrives, not whether the economics already work today. If this kind of clear-eyed and both sides breakdown of a genuinely ambitious idea is exactly what you come to this channel for. Go ahead and subscribe, because we're going to keep tracking every serious step in this space-based computing race as the actual missions launch, and the actual numbers start rolling in, not just when the press releases get flashy. McCallop's analysis surfaced one more insight that deserves real attention, because it explains exactly why the companies pursuing this idea most seriously are structured the way they are. He argued that the success of any orbital computing venture depends heavily on vertical integration, meaning a single company controlling as many pieces of the entire supply chain as possible, from the rockets themselves, to the satellite buses, to the power hardware, to the actual deployment process. In his own words, if you have to buy launch capacity from one company, buy satellite buses from another, buy power hardware from a third, and pay a profit margin at every single interface along the way, the accumulated costs and the extra weight required to accommodate all those separate systems will eat any potential savings alive. Vertical integration, he concluded, isn't simply a nice bonus feature for a company chasing this idea, it's the entire ball game. This is a big part of why SpaceX, a company that already owns and operates its own rockets, is taken so seriously as a potential leader in this exact race, and why smaller, less vertically integrated startups face a structurally steeper hill to climb toward genuine economic viability. Now let's talk honestly about the other serious engineering challenges standing between this idea and any kind of large-scale reality, because cost is far from the only obstacle these companies still have to solve. Radiation is a persistent, serious concern for any electronics operating in space, since satellites sit outside the protective blanket of Earth's atmosphere and magnetic field that normally shields ground-based hardware from a constant stream of high-energy particles. Commercial AI chips, like NVIDIA's H100, originally designed and tested for climate-controlled data centers here on Earth, were never built with cosmic radiation specifically in mind. And part of the entire purpose of missions, like Star Cloud One and Google's upcoming Suncatcher prototypes, is simply proving that these chips can survive and keep functioning reliably under that kind of sustained exposure over months and years, not just days. Then there's the growing genuinely serious problem of space debris and orbital congestion, an issue that gets considerably more complicated the moment you start talking about deploying hundreds, thousands, or in SpaceX's case, potentially up to a million individual satellites into an already increasingly crowded band of low Earth orbit. And finally, there's the basic physical reality that satellites have finite operational lifespans. Star Cloud's own satellites are currently expected to operate for roughly five years, based largely on the expected lifespan of the NVIDIA chips they carry, meaning any full-scale orbital data center strategy has to account for a continuous, expensive cycle of replacement launches, not a single one-time investment that simply keeps running indefinitely once it's up there. It's worth pausing here to appreciate the genuine scale of ambition sitting underneath all of these individual technical details, because Star Cloud's own long-term roadmap is not modest by any reasonable definition of the word. The company has stated its ultimate goal is building a five-gigawatt orbital data center, a facility whose solar panel arrays would need to measure roughly four kilometers by four kilometers, an absolutely enormous structure by any spacecraft standard in human history, larger than most cities' downtown districts if you laid it flat across the ground. Achieving a facility at that scale would require not just one successful satellite launch, but an entire evolving family of satellites, each one representing roughly a hundred times leap in compute capacity over its predecessor. Star Cloud 2, the company's next major milestone, is scheduled to launch before the end of this year, and unlike the first satellite's relatively simple single GPU setup, it's expected to carry a complete GPU cluster along with persistent onboard storage and far more sophisticated thermal and power management systems. Essentially, the first real test of whether this entire concept can scale meaningfully beyond a single proof-of-concept chip. Let's also think about why some of the biggest names in cloud computing and AI chip manufacturing are already climbing aboard this specific mission, because it tells you something important about how seriously the broader industry is starting to take this idea, even amid all the legitimate skepticism. Amazon Web Services, Google Cloud, NVIDIA, and a company called Crusoe are all reportedly flying either hardware or software aboard Star Cloud 2 when it launches, each one using the mission as an opportunity to independently test how their own specific technology handles operating in an orbital computing environment. That kind of broad, cross-industry participation, spanning, multiple competing cloud providers and chip makers all willing to put real resources into the same experimental mission, suggests this is no longer being treated purely as one's startup's speculative moonshot. It's being treated as a genuinely open, still unresolved question that some of the most resourced companies on the planet consider worth, actively investigating for themselves, rather than simply waiting on the sidelines to see how it plays out. So where does that leave the actual claim sitting in this video's title? The honest, complete answer requires holding two things true at the same time, and a responsible breakdown of this story has to resist the urge to simplify it down into a single, clean verdict. In either direction, it is genuinely true that a data center built directly into a satellite, operating in the right orbit, requires no water-based cooling infrastructure, no grid electricity, and no physical building constructed on the ground. And those really are three of the biggest structural bottlenecks currently strangling the terrestrial AI data, center industry here on Earth. Real satellites carrying real AI chips are already up there right now, actually running real workloads, not just existing as a concept sketched out in a slide deck. But it is equally true that today's version of this idea remains expensive, genuinely unproven at any meaningful commercial scale, and dependent on a specific bet about future launch costs continuing to fall that has not yet actually happened. Nobody serious in this field, including the companies betting hundreds of millions of dollars on the idea, is claiming that orbital data centers are ready to meaningfully replace ground-based infrastructure today. What they are claiming is that the trajectory, cheaper rockets, more resilient chips, and increasingly desperate terrestrial power and water constraints is bending steadily in this idea's favor, in that whoever solves the remaining engineering and economic puzzles first stands to gain an enormous, durable advantage in the next phase of the global AI infrastructure race. If you want to keep tracking exactly how this race actually develops over the coming months, whether StarCloud 2's launch later this year proves the concept can scale, or whether McCallup's cost skepticism ends up being the more accurate long-term read, make sure you're subscribed, because this is exactly the kind of ambitious, still unresolved story that deserves patient, ongoing coverage rather than a single video, and then silence. Here's the detail worth sitting with as we wrap up. 350 miles above the ground right now, a satellite about the size of a refrigerator is quietly finishing another lap around the planet, carrying a chip originally, built to sit inside a climate-controlled data center in Virginia, running an AI model powered entirely by sunlight, cooled entirely by radiating heat, out into the black emptiness of space, with no building, no water pipe, and no power line anywhere in sight. Whether that becomes the dominant model for how the world builds AI infrastructure, or whether it remains a fascinating expensive proof of concept that never quite closes the cost gap back down to earth, is genuinely still an open question. But for the first time in the history of computing, it is no longer a hypothetical one.