SpaceX + xAI Sparks Orbital Data Centers: The Cost Curve Investors Can Trade

SpaceX said it is combining with xAI at a reported $1.25 trillion valuation, pulling “AI + rockets + satellite network” into one balance sheet.

In the merger statement and related posts, Elon Musk reiterated a core claim: global AI power demand will hit a wall if it relies only on terrestrial data centers, and that space-based compute is the only long-term scaling path.

That narrative would still be abstract if not for an important proof point: startup Starcloud says it has already launched Starcloud-1 as a demo satellite carrying an NVIDIA H100, and has run AI workloads in orbit.

For public-market investors, the significance is not “another Musk mega-deal.” It is that orbital data centers now have something markets can price: an end-to-end industrial loop where compute demand (xAI), launch cadence and cost curve (SpaceX), and low-earth-orbit connectivity (Starlink) can be optimized as one system.

This theme will not be won by hype. It will be won at the intersection of two cost curves:

1) The ground-side constraint curve is rising.
Grid interconnect timelines, local permitting, water and cooling limits, and “time-to-power” are increasingly becoming the binding constraint for incremental compute.

2) The space-side cost curve is falling.
Reusability has already compressed launch costs dramatically, but the investable question is the next leg: whether heavy-lift, high-frequency systems can push the all-in cost per kilogram down far enough for orbital infrastructure to compete economically.

Orbital data centers are not “magically cheaper” because space is cold. The advantage comes from changing the boundary conditions of power and thermal management.

In orbit, solar input is essentially continuous on the right trajectory, without atmospheric attenuation. A widely used reference for the solar constant is ~1,361 W/m², materially above ground-level irradiance after atmospheric losses.

Deep space has an extremely low background temperature (~2.7 K). But the practical mechanism is radiative cooling, meaning you need large radiators that can shed heat as infrared radiation. This is an engineering trade between surface area, mass, deployment reliability, and survivability.

A growing share of orbital assets generate data in place (imaging, sensing, communications). Processing in orbit can reduce downlink burden and latency for certain workloads, especially when “good enough now” inference is more valuable than perfect later.

Radiation and reliability
Higher orbits can mean harsher radiation environments. Shielding, redundancy, fault-tolerant design, and software autonomy are not optional. One major solar event can become a P&L line item if the architecture is not resilient.

Cooling is not “space is cold, problem solved”
No atmosphere means no convection. Thermal design becomes a system-level problem, and maintenance is mostly “design it out” rather than “fix it later.”

Launch cost remains a gating factor
Even after big improvements, a “few-thousand dollars per kg” world is still very different from a sub-$1,000/kg world. The step-change likely requires new heavy-lift economics and sustained high cadence.

GPU refresh mismatch
GPU generations iterate fast. Orbital infrastructure tends to be long-lived. The winners will be the architectures that make upgrades modular, serviceable, or economically replaceable, rather than stranded.

Beyond the biggest cost buckets in terrestrial data centers such as compute, storage and networking, the more incremental upside of space-based data centers is concentrated in the aerospace industrial base: launch, satellite platforms, space-grade solar power, thermal control, radiation-hardened electronics, optical communications, and space-to-ground backhaul.

Against that backdrop, Morgan Stanley’s January space sector report is a useful starting point to map the relevant companies, especially since the note broadly raised price targets across the space complex.

What to track: reusable progress, launch tempo, contract backlog, and any credible signals that the industry is moving down the next cost step.

What to track: NTN/direct-to-device momentum, spectrum/regulatory shifts, defense wins, and any move toward optical/laser-linked architectures that increase secure throughput.

What to track: delivery performance, capacity expansion, and whether defense/commercial customers start ordering “compute-capable” satellite buses as a standard option.