Data centers are about to leave the atmosphere, and the pipes need to be wide enough for the exodus.
The Summary
- Observable Space closed a $90 million Series A to build laser-based satellite communication systems, targeting both traditional satellites and orbital data centers
- Laser comms offer 100x the bandwidth of radio frequency at a fraction of the power draw, critical for compute-heavy orbital infrastructure
- The real play: positioning infrastructure for the inevitable migration of AI training and inference workloads beyond Earth's regulatory and energy constraints
The Signal
Observable Space isn't solving today's satellite internet problem. They're building the backbone for something bigger: compute infrastructure that lives in orbit. The $90 million round signals investor belief that data centers will follow satellites into low Earth orbit, and when they do, laser communication systems will be the only viable way to move training data and model outputs at scale.
The physics tell the story. Radio frequency communication, the current standard, maxes out around 10 Gbps for satellite-to-ground links. Optical lasers can push 100+ Gbps with less power consumption and narrower beams that are harder to intercept or jam. For a traditional Earth observation satellite, that's nice to have. For an orbital GPU cluster running continuous training runs, it's table stakes.
"Laser comms offer 100x the bandwidth of radio frequency at a fraction of the power draw, critical for compute-heavy orbital infrastructure."
The orbital data center thesis has been theoretical until now, but Observable's fundraise suggests the timeline is compressing:
- Energy costs in space are near-zero once solar arrays are deployed
- Cooling is passive via radiative heat dissipation into the vacuum
- Regulatory arbitrage: no data residency laws, no local energy restrictions, no zoning battles
What's been missing is the pipe. You can't train frontier models in orbit if it takes 30 minutes to upload a dataset or download checkpoints. Observable is building that pipe, and the $90 million bet says the market believes orbital compute is coming faster than most realize.
The timing aligns with broader infrastructure plays. Starship's reusability is driving launch costs toward $10/kg. Solar panel efficiency keeps climbing. And AI companies are already hitting physical constraints on Earth, from power grid capacity to cooling water availability to the political friction of building new data centers near population centers.
The Implication
If you're building AI infrastructure companies, watch the orbital compute space. The first mover who can offer regulatory-free, energy-abundant training environments with acceptable latency will rewrite the economics of foundation model development. Observable just made that timeline more real.
For the rest of us: the Internet didn't stay terrestrial forever, and neither will the models. The question isn't if compute goes orbital, but when, and who controls the pipes when it does.