Google announces its first Suncatcher prototype satellite to test its AI chips

On September 24, 2026, Google announced the first prototype satellite for Project Suncatcher, designed to test its artificial intelligence chips in orbit. Developed with Planet, the satellite is set to fly on SpaceX’s Transporter-18 the following week. Suncatcher is Google’s program dedicated to solar-powered space computing.

Suncatcher was introduced by Google on November 4, 2025, with a two-satellite mission planned by early 2027. The new development is this first additional demonstrator. The New York Times report republished by The Business Times gives October 1, 2026 as the target launch date, which is subject to change.

That report describes a prototype named MVP, equipped with four TPUs and supplying those chips with approximately 1 kW from its solar panels. TPUs are Google’s specialized processors for artificial intelligence computations. The mission involves testing experimental hardware before any deployment of an operational orbital data center.

Google is counting on solar electricity for its potential computing constellations. In a suitable orbit, the company estimates that a panel could generate up to eight times more electricity than on the ground, with almost continuous exposure reducing the need for batteries.

Radiators limit computing time

The chips dissipate some of the electricity they consume as heat. In the vacuum of space, no airflow carries it away. NASA’s technical documentation describes heat transfer through materials and heat transport devices, followed by its release through radiation. A heat pipe can carry heat to the radiator, which releases it into space. Cooling therefore depends on the satellite’s ability to transport and release that heat.

In the New York Times report republished by The Business Times, Travis Beals envisages approximately fifteen minutes of operation on short queries before a shutdown for cooling. This duration is anticipated, not measured in orbit. The length of the pauses remains unknown, making it impossible to infer the proportion of time actually devoted to computing.

Google says it has tested cooling solutions in a thermal vacuum chamber. The company also reports promising tests involving launch vibrations and radiation, including TPUs running AI tasks under a proton beam. The flight must now test the complete system and measure its temperatures, hardware errors and performance.

The economic scenario depends on launch prices

Google Research envisages launches costing less than 200 dollars/kg in the mid-2030s, assuming that prices continue to fall. This threshold is a research scenario, not a price that has been achieved.

In this projection, orbital launch and operating costs, expressed per kilowatt per year, could approach the energy costs alone of an equivalent terrestrial data center. The comparison concerns different components of the total cost. It demonstrates neither full economic parity nor any reduction in prices for cloud customers.

For local authorities hosting data centers and electricity grid operators, success could eventually influence choices between terrestrial energy infrastructure and space computing. The prototype does not establish any current relief for power grids. Its environmental assessment will also need to account for equipment manufacturing and replacement, as well as launches, operations and end of life, beyond the solar electricity used.

After launch, the decisive data will concern sustainable computing time and hardware reliability in flight. A separate two-satellite mission is set to test laser communications in 2027. These links are intended to enable multiple satellites to work together. Validation of an orbital computing network will therefore also depend on data exchanges between satellites.