Elon Musk is pushing the AI race beyond Earth's atmosphere. On Monday, the CEO of SpaceX (SPCX.US) announced on X that the company's first batch of Starmind AI satellites, equipped with NVIDIA (NVDA.US) chips, will launch in the fourth quarter of 2027, reaching "significant scale" by 2028. This timeline marks an acceleration from the 2028 plan previously outlined in the company's IPO prospectus.
On the same day, NVIDIA officially confirmed that SpaceX AI will deploy its next-generation Vera CPU to accelerate agentic AI workloads, with optimized Vera Rubin NVL72 rack-scale systems destined for orbit as the computational core of the first-generation Starmind satellites. This collaboration signals a critical transition for the world's first space-based AI data center project, moving from concept to engineering reality.
From Terrestrial AI Factories to Orbital Data Hubs: Vera Rubin's Space Debut
NVIDIA's Vera is the first CPU purpose-built for AI agents, featuring 88 NVIDIA-designed Olympus cores and LPDDR5X memory with up to 1.2TB/s of bandwidth. In workloads such as agentic AI, reinforcement learning, and data processing, it completes tasks up to 1.8 times faster than x86 CPUs. Ian Buck, Vice President of Hyperscale and HPC at NVIDIA, stated: "Agentic AI requires a new kind of computing system—one that doesn't just generate answers but takes action. Vera empowers AI agents to act in real time—executing code, processing data, and orchestrating complex tasks. SpaceX AI is extending this architecture from terrestrial AI factories to the next frontier of orbital computing."
Mike Nicolls, President of SpaceX AI, noted: "Vera provides the CPU performance and memory bandwidth we need to handle massive orchestration, coding, and data processing tasks, while letting GPUs focus on what they do best. This translates to higher-performing AI agents and more valuable work produced per watt of compute."
Starmind: A 'Space Supercomputer' of One Million Satellites
Starmind is SpaceX's planned orbital AI data center constellation. In January 2026, SpaceX filed an application with the U.S. Federal Communications Commission (FCC) to launch up to one million computing satellites, designed to expand AI compute capacity for ground users from low Earth orbit. Each Starmind satellite is essentially an optimized Vera Rubin NVL72 rack-scale system, powered by NVIDIA Rubin GPUs and Vera CPUs. According to earlier disclosures, each satellite delivers 120kW of compute power (150kW peak), with the hardware designated as AI1. Laser links enable high-speed data transmission between satellites, forming a distributed space-based AI supercomputer.
The first Starmind AI1 satellites will be based on the optimized Vera Rubin NVL72 rack-scale system, marking the first time NVIDIA's accelerated computing platform heads into space. Musk posted on X that these data centers in satellite form are "simpler, lower cost, higher density, and lighter than traditional racks." Compared to ground-based facilities, space deployment eliminates the need for vast land use and faces relatively lower cooling pressure.
Compute Scale: An Aggressive Trajectory from 2GW to 10GW and Beyond
Musk also outlined more ambitious compute expansion targets. During an earnings call in early August, he revealed that SpaceX plans to have over 2 gigawatts of AI compute by the end of 2026, and by the end of 2027, capacity would be "closer to 10GW than 5GW." SpaceX AI also intends to expand ground-based AI infrastructure in Tennessee and Mississippi. The company aims to accumulate compute through multiple projects, potentially reaching 20GW by the end of next year. Musk acknowledged that some projects may not be fully completed on schedule, but he estimates around 15GW of online compute capacity by the end of 2027. He also indicated that SpaceX expects to secure a "very large proportion" of NVIDIA GPU supply in 2027.
Trillion-Dollar Capex and the Debate Over a 'Non-Traditional' Path
The cost structure of space-based data centers is under intense scrutiny from Wall Street. Wood Mackenzie estimates that building one gigawatt of orbital data center capacity would cost approximately $170 billion—more than three times the cost of a traditional ground-based facility. Morgan Stanley analyst Adam Jonas projects SpaceX's AI-related capital expenditure will reach $53 billion in 2026, with an additional $130 billion in 2027. However, proponents argue that space-based data centers can bypass regulatory hurdles and community opposition faced by terrestrial facilities. SpaceX is leveraging the reusability of its Starship rocket to lower launch costs, improving the economics of orbital computing. Musk previously explained that after evaluating multiple chip options, the company concluded that NVIDIA systems were the optimal choice.
From Exclusive Partnership to Deep Integration: SpaceX Goes 'All In' on NVIDIA
This collaboration represents a comprehensive upgrade of the relationship between SpaceX and NVIDIA. On August 4, during SpaceX's first quarterly earnings call post-IPO, Musk announced the company had decided to exclusively adopt NVIDIA technology for its AI compute infrastructure. "We have a very close working relationship and partnership with NVIDIA at multiple levels, and we have chosen to use NVIDIA exclusively," Musk stated clearly. SpaceX believes NVIDIA's Vera Rubin architecture is "the best AI computer available today." The two companies announced on August 4 that they would jointly develop the compute payload for the Starmind AI1 satellites, with each satellite carrying NVIDIA Vera CPUs and Rubin GPUs. SpaceX has informed investors that its space server chips will be exclusively sourced from NVIDIA. A 13F filing by NVIDIA on August 14 revealed the company holds approximately $21 billion in SpaceX shares. This deep binding relationship is now extending from the capital level to the technology level.
NVIDIA's Upcoming Earnings: The Space Narrative as the Latest Catalyst
The announcement comes just ahead of NVIDIA's quarterly earnings report on Wednesday (August 26). The space-based AI data center narrative is adding a new dimension to NVIDIA's AI infrastructure story. NVIDIA states that the Vera Rubin platform integrates compute, networking, and software into a unified architecture designed to lower per-token costs. As SpaceX AI advances toward gigawatt-scale compute capacity, Vera Rubin provides a unified architectural foundation from ground-based AI factories to orbital computing. In the orbital environment, constraints around power, thermal management, bandwidth, reliability, and physical integration are far more demanding than in traditional data centers. The partnership between NVIDIA and SpaceX signifies that its AI accelerated computing ecosystem is extending from Earth to low Earth orbit.
Comments