Several companies and state-backed programs are racing to put AI compute in orbit. Today the hardware in space is still small: individual GPU/TPU demonstrators and pathfinder constellations in the kilowatt range. This market asks who will be the first to cross the megawatt threshold.

Resolves to the first listed entity (or “Other”) to have an orbital data center that meets all of the following:

1. Power: at least 1 MW of continuous electrical power delivered to compute hardware (IT load: processors, memory, storage, networking).

   • Solar array peak output, power to the rest of the spacecraft, and power used only for propulsion, attitude control or thermal control do not count.

   • If an operator discloses only total spacecraft or system power, I may still count it when computing is clearly the system’s primary function and the figure leaves ≥1 MW for compute with reasonable margin. I will explain my reasoning in the comments.

   • Peak, burst or theoretical design figures do not count. The 1 MW must be sustainable in normal operation; duty-cycled systems count only at their average compute power.

2. Single system: the 1 MW must sit within one spacecraft or one group of spacecraft. A group counts only if it is linked by direct inter-satellite links (optical or RF) and operated as a single compute resource by the same operator. Separate satellites that only talk through ground stations do not add up.

3. Operational: the system must be in orbit (any Earth orbit, cislunar space or beyond) and must have run real computing workloads at or above the threshold. A launch alone, a deployment announcement or a capability on paper is not enough.

4. Evidence: credible public reporting confirms both the power figure and the operational status. This can be an official statement, a regulatory filing, an investor document or reputable press coverage.

Attribution:

• The answer goes to the entity that owns and operates the orbital system, not to its launch provider, chip supplier, cloud customer or hosted payload partner.

  • Example: Google workloads running on a Starcloud satellite count for Starcloud.

  • Example: a Starcloud satellite launched on a Falcon 9 does not count for SpaceX.

• SpaceX includes xAI and any SpaceX/xAI merged entity or subsidiary.

• “China” covers any entity based in the People’s Republic of China, including ADA Space, Zhejiang Lab (Three-Body Computing Constellation), Beijing Orbital Twilight Technology and BUPT/Tiansuan.

• For a genuine joint venture with shared ownership and operation, I may split resolution between the partners.

• Any qualifying entity not listed resolves to Other.

If no system qualifies by December 31, 2031, the market resolves to No one by the close date. If two systems seem to qualify in the same period, the earliest documented date of operation at ≥1 MW wins.

Context (Nov 2026):

• China’s Three-Body Computing Constellation has had computing satellites in orbit since May 2025.

• Starcloud-1 flew an NVIDIA H100 in November 2025.

• Axiom Space launched two orbital data center nodes in January 2026.

• Google launched its first Project Suncatcher prototype on October 1, 2026, with about 1 kW of power.

• SpaceX has filed for a constellation of up to 1 million data center satellites, and its AI1 satellite design is reportedly rated up to 150 kW.

• Starcloud plans 200 kW per Starcloud-3 spacecraft, launching on Starship around 2028–29.