Google is reportedly exploring a radical answer to the AI power crunch: put the data centres in orbit. It sounds like science fiction, but for most businesses the real story is not rockets. It is that AI compute is becoming an energy problem, and energy problems end up on your invoice.
What is the Concept
The idea, which Google has described as Project Suncatcher, is to fly solar-powered satellites carrying its AI chips, linked by laser connections, and run machine learning workloads in space. Sunlight in the right orbit is near-constant, so the satellites would need far less storage and no grid connection. Google has said early prototype launches are planned with a partner for around 2027, so this is research, not a product you can buy.
In plain terms: the largest AI providers are running out of easy places to plug in. Orbital compute is one bet on where the next gigawatts come from. Your task is to understand what that pressure does to the price and availability of the AI services you rely on.
Why It Matters Now (2025–2026 Context)
Hyperscalers are spending hundreds of billions of dollars a year on AI infrastructure, and power availability is now a common bottleneck for new capacity. When supply is constrained, providers ration it through pricing tiers, quotas and waiting lists.
A contrarian point: space will not make your AI cheaper in 2026 or 2027. The founder mistake is waiting for a miracle price drop instead of fixing wasteful usage today.
How AI Is Changing This
AI itself is the demand driver. Agents that call large models in loops can burn through tokens quickly, and each call has an energy cost somewhere. Model efficiency is improving fast, and smaller models now handle many business tasks well.
The non-obvious idea is that your best near-term lever is model routing: send easy tasks to small, cheap models and reserve frontier models for the hard ones.
Real-World Examples
Microsoft, Amazon and Google have all signed large power deals, including nuclear agreements, to feed their data centres. Those deals show how seriously the industry takes the energy constraint.
A realistic scenario: a 40-person SaaS firm adds an AI support agent and sees its monthly inference bill triple in a quarter. Caching, routing and shorter prompts bring it back down without hurting customer satisfaction.
Practical Insights / Actions
We call this the Compute Resilience Ladder: measure, route, cap, and diversify. Measure cost per outcome, route tasks to the cheapest capable model, cap runaway agent loops, and avoid single-provider lock-in.
Five actions for this quarter: track AI spend per workflow, set budget alerts, test a smaller model on your top three tasks, negotiate committed-use terms only after measuring, and ask vendors how they plan for power constraints.
Future Outlook
Orbital data centres face hard problems: launch cost, radiation, heat rejection, maintenance and latency. Even supporters treat them as long-term experiments. Earth-based capacity, from nuclear to better cooling, will carry most demand through this decade.
The hidden opportunity is for efficient builders. Companies that design lean AI workflows now will be the least exposed if compute stays expensive.
Conclusion
Google's off-planet plan is a signal, not a purchase decision. Treat it as proof that AI compute is energy-bound, and respond by tightening your own usage. If you want help auditing AI workloads and cutting waste, RP SoftTech can run a short compute cost review as a starting point.