SpaceX Orbital Compute Could Reach 5 GW by 2028
GF Securities analyst Jeff Pu projects the program's capacity target, with the first Kessel racks slated for production completion in December 2027.
Updated on Nov. 11, 2026 in Data Centers
GF Securities director Jeff Pu projected that SpaceX could reach 5 gigawatts of orbital computing capacity by 2028, with the first 6,000 Kessel racks providing 1.05 gigawatts. SpaceX and NVIDIA have not publicly disclosed a capacity target for the program.
Why it matters
If realized, the plan would place large-scale AI computing in orbit, where heat must be shed through radiation rather than air convection. The projection rests on an analyst forecast rather than an official company target, so the scale remains unconfirmed.
Each Kessel rack contains 72 Rubin GPUs, and the midplane is estimated to cost more than $20,000 according to Pu's projection. The estimates come from a single analyst, and SpaceX has not published component specifications or cost data for the system.
The players
SpaceX
SpaceX is a private American aerospace company founded by Elon Musk that develops the Starship launch system and the Starlink satellite network.
NVIDIA
NVIDIA is an American semiconductor company whose GPUs and NVL72 rack architecture underpin most large-scale AI data center hardware.
GF Securities
GF Securities is a Chinese brokerage and investment bank whose research division publishes equity analysis on technology supply chains.
The details
The Kessel design is a compact, wide-format system based on NVIDIA's NVL72 architecture and uses a single-loop liquid-cooling system that moves heat to external panels. Pu named Hon Hai (2317) as the sole assembler, with suppliers including Auras Technology (3017) for cold plates and Taiwan Hon Chuan (8039) and Taiwan Union Technology (2383) for PTFE and M9 laminate.
Timeline
May 2027: Projected new-product introduction of the first 6,000 Kessel racks.
December 2027: Projected production completion for the first 6,000 racks.
2028: Projected year for the 5 GW orbital computing capacity target.
The Tech Race
The Kessel design extends NVIDIA's NVL72 rack architecture into a compact, wide-format system, positioning the GPU maker against rivals pursuing AI infrastructure. The plan reflects a broader race to expand AI computing capacity beyond terrestrial data centers.
The projection points to a future where AI services could draw on orbital capacity, though no consumer-facing timeline or pricing has been announced. For now, the practical effect is limited to the hardware and supply-chain signals that analysts are tracking.
The takeaway
Forecasts of orbital computing capacity depend heavily on unconfirmed supplier and launch timelines, so readers should treat the 5 GW figure as an analyst estimate until SpaceX or NVIDIA publishes official targets.
Further reading
For related context, explore Data Centers coverage of how AI computing capacity is being built and cooled.
