Google has secured a rocket ride for its effort to run more AI on sunlight in space.
Jeff Bezos’ Blue Origin announced a launch agreement on Nov. 19 for Project Suncatcher, the search company’s long-term research into AI infrastructure in orbit. Success could open another route to expanding AI capacity as access to electricity constrains data center growth.
Upcoming missions will show whether individual spacecraft can work together as a larger computing system.
Each satellite would offer roughly a server rack’s computing power
A New Glenn rocket would deploy the spacecraft into sun-synchronous orbit on a flight targeted for the first quarter of 2029. Each would house Tensor Processing Units (TPUs), Google’s specialized processors for AI workloads.
Blue Origin describes the mission as the project’s first dedicated heavy-lift launch. “Its heavy-lift capacity and large payload volume give customers the flexibility to bring new capabilities to orbit,” said Jordan Charles, the company’s senior vice president of New Glenn.
Google says the agreement builds on its ongoing work with other satellite and launch partners. A satellite count for the flight was not disclosed.
Solar panels in the right orbit could be up to eight times as productive as those on Earth, according to Google’s estimates. Its proposed system favors near-continuous sunlight to power AI computing.
Laser links will face an orbital test in 2027
An earlier prototype, developed with satellite operator Planet, launched on Oct. 1 aboard a SpaceX rideshare flight. Engineers powered up its host computer and TPU hardware over the following weeks and completed initial machine-learning tests. During a ground-station pass later that month, the onboard Gemma 4 model returned an AI-generated haiku. Follow-up prompts reported the technical status of the payload’s systems.
A two-satellite mission planned for 2027 will test laser connections designed to move large amounts of data quickly between nearby spacecraft. AI jobs split across satellites would depend on those links, which must stay precisely aligned as both ends move.
Engineers have also tested cooling designs that combine heat pipes and radiators in a chamber simulating conditions in space. Without airflow, processors need a different way to shed heat. Further orbital tests will assess those designs and the chips’ response to radiation.
Neither company’s launch announcement details customer pricing or a partner program for commercial access.
Customer data should guide decisions about orbital AI
Systems integrators evaluating future orbital services should start with the business data an application needs to access. Google treats connections to Earth separately from links between satellites, noting that the proposed orbit increases communication delays to some ground locations.
An application’s dependence on live records would therefore influence how effectively it could use processing capacity in space.
| Potential application | How it would use data | Connection requirement |
| Scheduled analysis of a prepared dataset | Inputs could be transferred before a run, with results returned afterward | Enough transfer capacity to load the dataset and meet the completion deadline |
| Customer-service agent using current records | One interaction could trigger repeated requests to order and support systems | Reliable connections throughout the interaction, with minimal waiting between requests |
An integrator’s recommendation about where a workload should run should account for the cost of transferring data and keeping any copies current. Performance testing also needs to measure what happens when a connection slows or fails.
Recommend a move only when the full workflow meets the customer’s agreed service targets. A job that finishes faster on the processor can still take longer for the user if its data spends more time in transit.
More AI news: Google’s latest AI acquisition could create new opportunities for Workspace partners helping businesses automate administrative workflows.


