The conversation around space datacenters has quickly moved from science fiction to serious industry discussion. Promises of abundant solar power, fewer permitting hurdles, and virtually unlimited room for expansion have fueled speculation that the future of AI infrastructure may ultimately extend beyond Earth.
But according to Dan Nishball, Ellie Holbrook, and Harley Blackard of SemiAnalysis, authors of one of the most talked-about market research reports in 2026, the more important question is not whether space datacenters are possible. It is whether they make economic sense.
The trio joined the Shared Everything podcast this week to talk through their findings and narrow in on what the future might look like for orbital compute.
As Nishball told us, the starting point is total cost of ownership.“Where we are right now is quite far from parity, around a four times, four to five times greater total cost of ownership in terms of a space DC versus a terrestrial data center.”
That cost gap stems from a combination of factors. Space deployments require expensive launch capacity, shorter assumed operating lifetimes, significant hardware redundancy, and solutions for reliability challenges that are largely solved in terrestrial environments. On Earth, failed hardware can be replaced. In orbit, operators must plan for failure before systems ever leave the ground.
Yet the conversation was not centered on why space is difficult. Instead, it focused on what conditions might eventually make it attractive.
One of the more interesting conclusions from the SemiAnalysis work is that cost parity alone is not enough.
Cost parity opens the door to going to space, but what really forces you to go to space is the availability of power. And it’s actually twofold. It’s number one, the availability of power, but it’s also the availability of compute and the demand.
Orbital compute only becomes compelling if terrestrial infrastructure begins running into limits that can’t easily be overcome.
To evaluate that possibility, the SemiAnalysis team developed a framework that examines four layers of terrestrial infrastructure expansion. The first layer relies on traditional grid-connected power. The second repurposes existing powered infrastructure such as former bitcoin mining sites. The third shifts toward behind-the-meter generation, increasingly powered by dedicated gas infrastructure. And the fourth layer explores emerging options including nuclear, geothermal, and other advanced energy sources. What’s interesting here is that each layer becomes progressively more expensive.
Importantly, Holbrook noted that the industry’s challenge is often misunderstood. The problem is not necessarily that power does not exist.“We wouldn’t say that the constraint is the power. I think it’s much more to do with the transmission and substations and the queues that these companies are waiting to join up and hook to the grid.”
That distinction matters because it changes the conversation from energy scarcity to infrastructure deployment.
The discussion also challenged several common assumptions surrounding orbital compute. While many proponents point to lower power costs in space, Blackard argued that the reality is more nuanced. Space changes power from an ongoing operating expense into a large upfront capital investment involving solar arrays, batteries, and supporting systems.
Cooling presents an even larger challenge, they all agree. “It is a major, probably if not the most major technological challenge to figure out between now and 2039.”
Likewise, launch costs, while significant today, may not ultimately be the dominant factor. As the team modeled future scenarios, they found that compute hardware itself remains the largest cost component. Even dramatic reductions in launch pricing have less impact than many assume because AI accelerators and supporting infrastructure account for the majority of system costs.
The conversation eventually turned to a less obvious constraint: people.
While power, semiconductors, and datacenter construction receive most of the attention, Holbrook argued that workforce limitations may become one of the most significant barriers to continued expansion. “One of the most serious binding constraints is actually manpower.”
Building hundreds of gigawatts of infrastructure requires engineers, construction workers, operators, and specialists. As facilities expand into more remote locations, finding and keeping qualified talent becomes increasingly difficult.
Viewed through that lens, the debate over orbital compute becomes less about rockets and more about economics. The real question is not whether AI infrastructure can be deployed in space. It is where the next incremental unit of infrastructure should be built.
As Blackard summarized, operators and investors will ultimately evaluate a simple question:
“What’s the most efficient use of your next megawatt and your next dollar?”
That question may determine the future of AI infrastructure far more than any discussion about life beyond Earth.
