Foreword:

During SpaceX's recent earnings call following its public listing, Elon Musk extensively discussed topics such as electricity, server racks, and a satellite named Starmind AI1.

All of SpaceX's future AI computing needs will be exclusively met by NVIDIA's systems, spanning from ground-based data centers in Memphis to installations in low-Earth orbit.

Following the announcement, NVIDIA's stock price surged by 3.43%, closing at $219.22, while SpaceX's stock dipped by 13.61%. This divergence in market reactions adds an intriguing layer to the partnership.

Author | Fang Wensan

Image Source | Internet

Exclusive Agreement: A Strategic 'Insurance' for Both Companies

At this year's GTC conference in March, when Jensen Huang unveiled the Space-1 Vera Rubin space computing module, the announced partners included five aerospace companies—Starcloud, Planet Labs, Kepler, etc.—but notably absent was SpaceX.

This omission was conspicuous: NVIDIA possesses cutting-edge chips but lacks rockets or its own satellite constellation. The concept of orbital computing had been circulating for months, yet it lacked a key player with genuine launch capabilities.

SpaceX stepped in to fill this void, and did so exclusively.

For NVIDIA, this partnership means securing a foothold in space, complementing its dominant 85% share of ground-based data center GPUs.

Competitors like AMD have yet to produce GPU products validated for orbital environments, making the exclusive agreement a significant barrier to entry for potential rivals.

SpaceX's 'insurance' is more pragmatic. Its current bottleneck lies not in chip selection but in ensuring a reliable supply and adequate power.

As of the end of Q2 2026, SpaceX's available AI computing power stood at 1.4 gigawatts, with ambitions to exceed 2 gigawatts by year-end and approach 10 gigawatts by the end of 2027. During the earnings call, when questioned by Morgan Stanley analysts, Musk adjusted the target to a cumulative 20 gigawatts, later clarifying that 15 gigawatts was a more realistic goal.

Such rapid expansion necessitates securing a top spot on NVIDIA's production capacity list.

The Vera Rubin NVL72 server racks are yet to be mass-produced, with cloud providers worldwide—from Memphis to Riyadh—competing for limited supplies. The exclusive commitment guarantees SpaceX priority in the supply chain.

Opting to abandon self-developed chips in favor of a deep partnership with a leading supplier is a calculated strategic move.

Starlink: Adding a Layer of 'Computing' to Its Arsenal

SpaceX's most notable achievement has been its ability to bridge gaps between different industrial segments, integrating them into a cohesive whole.

Falcon and Starship handle transportation, Starlink satellites form the network, and ground terminals connect users, transforming rocket launches into a continuous revenue-generating communication service.

The value of this model extends beyond cost savings on rockets.

A deeper transformation lies in SpaceX's capacity to redesign the entire system according to its own needs. Satellite weight, launch frequency, orbital deployment, communication links, and terminal capabilities are no longer optimized separately but adhere to a unified cost model.

Starmind continues this logic. With GPUs entering orbit, the Starlink satellite network is no longer just a conduit for data transmission. Some data can now be processed in space, determining what content needs to be sent back to Earth.

While this may seem like a minor addition, it fundamentally alters the functional boundaries of the satellite network.

Today, Starlink primarily addresses connectivity issues. In the future, orbital networks may also handle tasks such as identification, filtering, compression, inference, and scheduling.

A satellite thus evolves from a mere 'communication node' to a 'computing node,' a subtle yet significant shift in SpaceX's foray into space computing.

It's not about building a floating data center from scratch but enhancing an existing large-scale satellite network with new capabilities.

For the first time, transportation, communication, and computing are being integrated into a single engineering framework.

Once this path is established, SpaceX's competitive boundaries will also expand.

Cloud computing companies, AI infrastructure firms, and data center operators will increasingly find themselves competing in the same arena.

NVIDIA Takes Its AI Factory Beyond Earth

In recent years, NVIDIA has shifted focus away from individual GPUs.

Vera Rubin represents far more than just a more powerful chip. CPUs, GPUs, NVLink, high-speed networking, DPUs, storage, and software platforms are organized into a complete computing ecosystem.

NVIDIA refers to this as the AI Factory, where customers purchase an entire architecture that defines how AI data centers operate.

Now, NVIDIA is preparing to extend this concept into orbit. This year, NVIDIA announced its Space Computing platform, covering orbital data centers, geospatial intelligence, and autonomous space missions. The Space-1 Vera Rubin Module further concretizes this vision in hardware.

For NVIDIA, the location of servers—whether in the U.S., Asia, or hundreds of kilometers above Earth—becomes a secondary concern.

As long as the computing nodes utilize its defined CPU, GPU, interconnect, and software ecosystem, NVIDIA retains control over the computing architecture.

NVIDIA is attempting to extend its influence over AI data center architectures from the ground into space.

The Real Drivers of Space Computing Are Earthbound

Why send server racks into orbit? The public rationale is compelling: boundless solar energy, free vacuum-based cooling, and no land or water usage.

However, the true motivations are more practical. Ground-based data centers face three critical bottlenecks: grid access queues, community opposition, and cooling water quotas.

During the earnings call, Musk candidly stated that current resource investments focus on power and cooling equipment, with GPUs being less of a constraint. This reveals a stark reality of U.S. AI infrastructure: chips are readily available, but power is not.

Orbital scenarios offer an additional, often overlooked advantage: data locality. The Space-1 module is not designed to process data transmitted from Earth but rather to handle information collected by other orbital sensors and spacecraft on-site.

Onboard screening of remote sensing imagery before transmission saves bandwidth and latency. For Starlink, with 9,600 in-orbit satellites and millions of users across 164 countries, embedding inference capabilities into the constellation upgrades each satellite from a signal relay to an edge computing node.

This is why SpaceX is better suited for this endeavor than any cloud provider—its data already resides in space.

The True Protagonist After Earnings: A New Growth Narrative

Returning to the earnings call itself, SpaceX's Q2 revenue grew approximately 90% year-over-year, with capital expenditures surging 550% to $18.4 billion. AI business revenue reached $2.6 billion, up 247% year-over-year.

The numbers appear impressive, but a closer look reveals that growth is highly dependent on two long-term contracts: Anthropic leases the entire capacity of Colossus 1 for $1.25 billion per month until May 2029, and Google leases approximately 110,000 GPUs for $920 million per month.

These contracts contribute the vast majority of predictable computing revenue, yet they share a common vulnerability: both clients are building their own computing capacity. Anthropic is developing its own chips, and Google's agreement includes an early exit clause.

A newly public company like SpaceX, needing to demonstrate growth to open markets, cannot rely indefinitely on subleasing GPUs as its core narrative.

Space computing offers a second narrative, rebranding SpaceX from a landlord of ground-based data centers to the sole builder of orbital infrastructure—a position competitors cannot replicate in the short term.

The market reacted to the $18.4 billion quarterly capital expenditure with a 13.61% stock price decline, but Musk responded with ambitious 10-gigawatt and 20-gigawatt targets and the Starmind satellite. The narrative's potential must outpace spending.

Simultaneously escalating paper losses and expansion speed indicate that space computing serves two purposes: seeking the next revenue curve and providing a longer valuation horizon for massive AI investments.

Conclusion:

When ground-based computing expansion hits power and community ceilings, the tech industry's solution is not to halt but to relocate to a neighbor-free zone for continued growth.

Whether space computing is cost-effective remains undecided, but it has already redefined the dimensions of data center competition: from competing for electricity and chips to competing for orbits and rockets.

NVIDIA secures new territory, SpaceX gains a new growth narrative, and the bill awaits in the flames of the next Starship test flight.

Partial References: MarketWatch: 'SpaceX is exclusive to Nvidia, Musk says,' Reuters: 'SpaceX slides as AI spending worries overshadow early returns,' NVIDIA News: 'NVIDIA Introduces Space Computing Platform,' arXiv: 'Orbital Data Center Economic Viability Analysis'