The narrative being sold to the public is that Google and SpaceX are building data centres in space because it's exciting, futuristic, and good for innovation. The real reason is more urgent and less glamorous: Earth is running out of the water and energy required to sustain AI's growth trajectory, and orbital infrastructure may be the only viable exit from a resource crisis that's already constraining expansion plans across three continents. What's framed as a moonshot is actually a supply chain emergency response.
The Earth-Bound Crisis That Is Forcing the Space Decision
Start with the numbers. SpaceX's $75 billion IPO filing explicitly listed water access as a critical operational risk — on par with chip shortages and power grid constraints. A large-scale AI data centre consumes up to 5 million gallons of water daily for cooling. The Lincoln Institute of Land Policy projects that data centres in Texas alone could consume 399 billion gallons annually by decade's end. In Arizona — home to the second-largest US data centre cluster — local governments have begun restricting new data centre permits due to groundwater depletion concerns.
Energy is equally constrained. Industry analyses indicate that 30-50% of approximately 140 planned US data centres targeting 16 GW of capacity may miss 2026 timelines or face cancellation — primarily because power grid interconnection queues are running 4-7 years in most US markets. Europe faces the same problem: Germany, Ireland, and the Netherlands have all imposed moratoriums or severe restrictions on new data centre power connections in major markets. The AI computing demand curve shows no sign of levelling. Training a single frontier-class AI model now consumes an estimated 700,000 litres of water and tens of millions of dollars in electricity. With hundreds of models being trained globally each month, the arithmetic of resource consumption is forcing infrastructure planners to look beyond terrestrial solutions. We covered this in detail in our analysis of SpaceX's IPO water risk disclosure — orbital infrastructure is the logical next step.
How Orbital Data Centres Actually Work — and Why Now Is the Moment
Google's internal project, reportedly codenamed Project Suncatcher and cited by Tech Startups in May 2026, targets prototype satellite deployments as early as 2027. SpaceX plans dedicated AI compute satellites for 2028. The engineering premise is straightforward but previously uneconomical: satellites in sun-synchronous orbit receive continuous solar radiation, generating power with 5x the energy density of ground-based solar panels. Cooling is achieved through passive radiation — the temperature differential between a satellite's sun-facing and shadow-facing sides is approximately 270 degrees Celsius, providing massive heat dissipation capacity without any water.
Three things changed simultaneously to make orbital AI infrastructure economically viable: Starlink's reusable launch cost reductions (dropping payload costs to orbit from $10,000/kg in 2010 to approximately $1,200/kg today), the miniaturisation of GPU clusters (next-generation chips pack 5x more compute into half the physical volume of H100s), and the emergence of low-Earth orbit latency levels below 40 milliseconds — sufficient for most AI inference workloads. The combination of cheap launch, smaller hardware, and acceptable latency crosses the economic viability threshold for orbital computing for the first time. Before this convergence, orbital data centres were a thought experiment. After it, they're an infrastructure roadmap item with specific 2027-2028 deployment targets from the two most capable space companies on Earth.
What This Means for India, Asia, and the Global South
For India specifically, orbital AI infrastructure could be transformative. India's data centre buildout — including AirTrunk's announced Rs 3 lakh crore investment — faces the same power reliability and water stress challenges as US markets. If orbital AI inference becomes available via Starlink-class satellite downlinks by 2029-2030, Indian AI companies could access low-latency, cost-effective compute without needing to wait for domestic data centre infrastructure to mature fully.
Globally, orbital AI creates a compute equity opportunity: regions that have been disadvantaged by geography in the terrestrial data centre buildout — sub-Saharan Africa, Southeast Asia, Central America — gain access to the same AI compute resources as New York or London, connected via satellite. According to GSMA's 2025 Mobile Economy report, Starlink and LEO satellite connectivity has already reduced the mobile internet access gap by 18 percentage points in underserved markets. Orbital compute could replicate that pattern specifically for AI access, reducing the global AI capability divide that today heavily advantages US and Chinese tech companies over the rest of the world.
The Risks — Technical, Geopolitical, and Economic
Orbital data centres introduce new failure modes. Satellites in LEO experience orbital decay, solar weather events, and debris collision risks that ground-based infrastructure does not face. The regulatory environment for orbital computing is entirely undefined: which jurisdiction's laws govern data processed on a satellite over international waters? Data sovereignty questions that took a decade to resolve for cloud computing will need to be addressed before enterprise adoption becomes mainstream for orbital infrastructure.
Economically, the first-mover advantage for Google and SpaceX is enormous — but so is the capital risk. A satellite compute cluster that fails or underperforms represents billions in unrecoverable expenditure. The conservative estimate for Google's Project Suncatcher initial deployment is $3-5 billion in capital costs before a single inference is served commercially. Alphabet's announced $80 billion raise to fund AI infrastructure suggests the financial firepower exists — but converting capital into reliable orbital compute at enterprise scale is an engineering challenge that has never been attempted.
What This Means for You
For enterprise AI buyers globally: plan your 2028-2030 AI infrastructure strategy to include evaluation of satellite-based AI compute as a third option alongside on-premises and cloud — the latency and cost curves will be competitive with cloud by then for inference workloads. For developers and startups in underserved markets, including India's Tier 3 cities and Southeast Asia: orbital compute access could arrive before terrestrial infrastructure does in your region. Building for satellite-latency assumptions now is forward-compatible with that future.
Frequently Asked Questions (FAQs)
Q: Are Google and SpaceX really building AI data centres in space?
A: Yes. Google is developing Project Suncatcher with prototype orbital satellites targeted for 2027. SpaceX is planning AI compute satellites for 2028. Both use space-based solar power and passive radiative cooling to eliminate the water and energy constraints binding terrestrial data centres.
Q: Why are AI data centres being moved to space?
A: Terrestrial AI data centres face binding constraints on water (up to 5M gallons/day for cooling) and power (4-7 year grid interconnection queues in most US markets). Space eliminates both constraints using continuous solar power and radiative cooling — no water required at all.
Q: What is Google Project Suncatcher?
A: Google's reported internal project for orbital AI computing infrastructure. Prototype satellites are targeted for 2027 deployment, with the goal of providing AI compute capacity via space-based solar-powered servers connected through Starlink-class satellite internet at sub-40ms latency.
Q: How does orbital AI infrastructure affect India?
A: India's domestic data centre buildout faces power and water challenges similar to US markets. Orbital AI compute delivered via LEO satellites could give Indian AI companies access to competitive, low-latency compute capacity independently of domestic infrastructure timelines — potentially as early as 2029-2030.
Q: Will orbital data centres reduce the AI divide between developed and developing countries?
A: Potentially yes. Regions historically disadvantaged by terrestrial data centre geography — including sub-Saharan Africa, Southeast Asia, and South America — could gain access to the same AI compute resources as US and European markets via satellite connectivity, reducing the global AI capability gap that currently disadvantages developing-country startups and researchers.
Google and SpaceX aren't building data centres in space because they love space. They're building them because the alternative — watching AI's growth stall against water permits and power queues — is worse for their business than the cost and complexity of going orbital. This is what resource-constrained innovation looks like in 2026. The space race just became an infrastructure race, and the finish line is a teraflop served from orbit. The 2027-2028 deployment window will be the moment of truth for whether orbital AI moves from ambitious plan to operational reality.
