SpaceX announced it has acquired xAI, Musk's AI venture valued at roughly $200B in September 2025, with plans to deploy orbital data centers that run on solar power via Starlink satellites.
The Pitch
Musk's argument: terrestrial AI data centers hit hard limits on power and cooling. His solution is launching compute to space where solar power is constant and cooling is free. The company claims launching one million tons of satellites annually - each generating 100kW - would add 100GW of AI compute capacity with "no ongoing operational or maintenance needs."
Starship, SpaceX's heavy-lift vehicle, would launch these satellites at a rate of 200 tons per flight. The company estimates space-based compute becomes cost-competitive with terrestrial options within 2-3 years.
Reality Check
Multiple issues require scrutiny. First, no merger has actually closed - reports from late January indicate SpaceX and xAI are in exploratory talks ahead of a potential SpaceX IPO mid-2026. The Nevada entity filings (K2 Merger Sub Inc.) confirm structure planning, not completion.
Second, the "no maintenance" claim ignores satellite replacement cycles, orbital debris management, and the latency physics that make space-based compute problematic for many enterprise workloads. Current Starlink latency runs 25-60ms - acceptable for broadband but challenging for real-time trading, gaming, or synchronous AI inference.
Third, Starlink's current enterprise offerings don't yet match traditional connectivity SLAs. The Business Priority plan costs $500/month versus $120 Residential, with better performance but no published uptime guarantees for critical operations. The High Performance enterprise kit runs $2,500 upfront plus $500/month.
What This Means
For enterprise IT leaders evaluating Starlink as backup connectivity, the space-based AI narrative is separate from near-term utility. Starlink works for remote sites where fiber isn't an option. Using it as primary connectivity for latency-sensitive workloads remains risky.
The orbital data center concept may eventually address training workloads where latency matters less than throughput. But inference - the use case most enterprises care about - requires low latency that physics makes difficult from orbit.
xAI currently holds a DoD contract worth up to $200M and develops AI for Starlink and Starshield (SpaceX's government-focused constellation). Whether orbital compute serves military applications better than commercial ones is an open question.
The Bigger Pattern
This follows Musk's playbook: Tesla acquired SolarCity in 2016; xAI acquired Twitter (now X) last year at a $33B valuation that has since declined. Tesla invested $2B in xAI this week; SpaceX invested $2B last year. Shareholder lawsuits have alleged Musk uses one company's resources to prop up others.
SpaceX's $800B late-2024 valuation could reach $1-1.5T at IPO. Consolidating xAI before going public packages the AI narrative with proven space hardware.
We'll see if the power math works. History suggests treating Musk timelines as aspirational rather than contractual.