Project Suncatcher name and sun-shaped symbol on a yellow, orange and blue background
Project Suncatcher name and sun-shaped symbol on a yellow, orange and blue background
+ AI News

Google's space AI plan assumes around 1,800 Starship launches to cut launch costs

Google's first TPU satellite is in orbit. Its launch-cost projection assumes a decade of heavy Starship traffic, while power, cooling and laser links still need testing.

Google has sent its first advanced AI chip into orbit, but its plan for large space data centers depends on a much larger change in launch economics. TechCrunch reports that Google’s updated research projects launch prices near $200 per kilogram by 2035, assuming Starship carries enough payload to follow a cost-reduction curve similar to SpaceX’s earlier rockets.

That assumption implies roughly 1,800 launches over ten years, each carrying 200 metric tons. Google’s researchers caution that their analysis is not an economic feasibility study. The estimate describes a route to cheaper launches, rather than a date when orbital AI computing becomes commercially viable.

A chip test before a connected cluster

The Planet Labs-built prototype launched from California on October 1. Once commissioned, it will run Google’s Tensor Processing Unit in 15-minute bursts to limit demands on the satellite’s power and cooling systems, according to TechCrunch.

Google describes the mission as a test of hardware survival and thermal management. Space has no airflow to carry heat away. The team is testing heat pipes and radiators to remove heat from the chips.

Two satellites planned for 2027 will test laser links between them. Those links must provide enough bandwidth for chips on separate spacecraft to work together. TechCrunch reports that Google’s longer-term design envisions 81 satellites flying close together and processing workloads in parallel.

Radiation tests offer some encouragement for inference, the work of running trained models. TechCrunch reports that Google repeated tests after finding that the earlier chip configuration supplied too much shielding. The revised tests produced more logic errors, but Google still expects its chips to handle large inference workloads over a five-year satellite lifespan. Beals, who leads the project, said the errors remain a problem for training runs that use thousands of chips for months.

The launch assumption does much of the economic work

The revised paper, which TechCrunch says will appear in Joule, uses SpaceX’s historical cost reductions to project future launch prices. Its estimate requires Starship to deliver 370,000 metric tons to orbit. TechCrunch contrasts the resulting annual launch target with Starship’s record of at most five flights in a year.

TechCrunch corrected its headline from 1,600 launches to 1,800. The older number remains in the page’s URL.

Sources

The AI Feed Desk

The AI Feed Desk

Editorial desk

The AI Feed Desk tracks AI provider updates, model releases, agent tooling, and enterprise adoption, turning fast-moving announcements into source-linked context for builders and operators.

Noticed a typo, incorrect information, or translation error?

Tell us so we can fix it.

Help Improve This Article

Related Articles

Matt Garman wearing a blue jacket with his arms crossed against a dark background

Amazon says it no longer uses NDAs for data centers as backlash grows

AWS says it has stopped using NDAs with government agencies. Its new commitments leave questions about existing agreements and access to local project data.

The AI Feed Desk

By The AI Feed Desk

A controlled workflow graph routes tasks through two glowing AI reasoning nodes

Google ADK 2.0 puts workflows around agents

Google's ADK 2.0 framing separates deterministic workflow routing from open-ended model reasoning, giving production agents a stricter execution boundary.

The AI Feed Desk

By The AI Feed Desk

A coding agent workflow loops through traces, grading, failure clusters, and approved fixes

Google gives coding agents an eval flywheel instead of another prompt tweak

Google's new quality-flywheel skill lets coding agents run structured agent evaluations with independent grading and production-trace loops.

The AI Feed Desk

By The AI Feed Desk

Google's Search launch graphic shows a mobile search page with green answer-loading bars

Judge dismisses Chegg and Penske antitrust suits over Google AI Overviews

A judge rejected Chegg and Penske's antitrust claims over AI Overviews. His reasoning separates publishers' expectation of search traffic from an agreement with Google.

The AI Feed Desk

By The AI Feed Desk

A modern data-center campus connected to power lines and abstract cloud compute blocks in a green landscape

Google puts $1.5B into its Alabama data-center campus

Google says it will invest $1.5B across 2026 and 2027 to expand its Jackson County, Alabama data-center campus while covering its power and infrastructure costs.

The AI Feed Desk

By The AI Feed Desk