The race in artificial intelligence is no longer solely decided in the laboratories where models are trained or in presentations where tech companies boast increasingly capable assistants. The battle is moving to a less glamorous, yet far more decisive, battleground: data centers. Those with greater computing capacity, available energy, more GPUs, and faster infrastructure deployment will have a hard-to-match advantage.
SpaceX has recently entered this league through an unexpected route. The company known for its rockets, Starlink, and space ambitions has signed a computing capacity agreement with Google that positions it squarely in the AI infrastructure business. The contract, disclosed in SEC filings, states that Google will pay $920 million per month from October 2026 to June 2029 for access to approximately 110,000 NVIDIA GPUs, in addition to CPUs, memory, and other related components. Capacity will be delivered progressively, and the agreement includes termination clauses if the committed terms are not met.
This move follows another large-scale agreement with Anthropic, the creator of Claude. According to xAI, Anthropic will have access to Colossus 1, an AI supercomputer located near Memphis, which the company describes as one of the largest and fastest deployment sites in the world. xAI reports more than 220,000 NVIDIA GPUs, including H100, H200, and GB200, dedicated to training, fine-tuning, inference, and high-performance computing tasks.
The implications are significant: Grok, Gemini, and Claude—three major players in generative AI—now depend, to varying degrees, on infrastructure linked to Elon Musk’s enterprise ecosystem. This doesn’t mean they share models, data, or internal technology. What matters more practically and commercially is that all require tremendous amounts of compute power, and the market lacks sufficient idle capacity to meet the demand at the pace that generative AI demands.
From Rocket Making to Renting Computing Power
The industrial logic behind this shift is clear. AI consumes infrastructure at a rate that’s stretching the entire tech sector. Google has its own cloud, TPUs, data centers, and one of the largest AI capacities worldwide. Still, the growth of its Gemini-based products and enterprise platforms has driven it to seek additional capacity.
In SEC filings, SpaceX explains that its agreement with Google covers around 110,000 NVIDIA GPUs and other infrastructure components. The contract stipulates a weekly payment of $920 million from October 2026 through June 2029, with a ramp-up period and clauses allowing Google to cancel or adjust the deal if capacity isn’t available on time.
For SpaceX, such contracts open a massive, relatively new revenue stream. The company isn’t selling a full cloud offering like AWS, Azure, or Google Cloud. It doesn’t yet provide managed databases, development platforms, analytics, security, observability, or enterprise services at that scale. Instead, what it offers is more straightforward: raw capacity to train and run AI models.
That distinction matters. SpaceX isn’t competing as a full cloud provider but as an provider of extreme infrastructure. In a market where NVIDIA GPUs remain among the most coveted resources, this position may be enough to secure billion-dollar contracts.
| Agreement | Indicated Capacity | Period | Reported Amount |
|---|---|---|---|
| Google / SpaceX | Approximately 110,000 NVIDIA GPUs, CPUs, memory, and related components | October 2026 to June 2029 | $920 million per month |
| Anthropic / xAI-SpaceX | Access to Colossus 1, with over 220,000 NVIDIA GPUs, per xAI | Until 2029, according to published contract info | $1.25 billion per month, per specialized media reports |
| Colossus 1 | AI cluster near Memphis | Initial build in 122 days, per xAI | xAI describes it as a gigafactory of computation |
The scale hints at the magnitude of the ongoing shift. Generative AI has elevated the cloud market to a point where having software, clients, and advanced models isn’t enough. Secure power supply, land, cooling, networking, servers, GPUs, and skilled personnel to quickly assemble huge clusters are now essential—without delays that could span months.
Colossus and the New Cloud Power Geography
Colossus has become a symbolic element of this new phase. xAI reports that the facility was built in just 122 days and later doubled its capacity to 200,000 H100 GPUs in a single interconnected cluster. The company talks about plans to expand toward one million GPUs, illustrating how scale has become central to the competitive message.
For Anthropic, extra capacity means more margin for Claude users, higher inference capabilities, and better conditions for training or fine-tuning models. For Google, the deal serves as a bridge to handle demand that has outpaced expectations. For SpaceX and xAI, these contracts transform an infrastructure built for their own models into a commercial asset.
This shift also redefines the role of data centers in tech. For years, cloud narratives focused on abstraction—no need to think about servers, disks, or networks, since everything was consumed as a service. Now, AI has brought hardware back into the spotlight. Companies are talking again about megawatts, GPU availability, liquid cooling, permits, electrical tension, latency, and energy contracts.
Cloud is becoming more tangible. This is precisely where SpaceX feels at home. Its corporate culture emphasizes vertical integration, rapid construction, supply chain control, and risk-taking—traits that give it an edge in a market where six months of delay can mean missing a generation of models.
But it also comes with costs.
Energy, Permits, and Environmental Concerns
The rapid deployment of AI infrastructure is sparking a broader debate beyond technology. Colossus and related xAI facilities have faced criticism over energy consumption, use of gas turbines, and impact on nearby communities. In 2026, lawsuits and complaints related to noise, emissions, and environmental permits have been filed in Tennessee and Mississippi.
This isn’t a minor issue. AI data centers aren’t just buildings with servers—they’re major energy and water consumers. Their presence can strain local electrical grids, increase pressure on urban infrastructure, and provoke conflicts with residents if deployment is perceived as opaque or unbalanced. The economic promise of new jobs and tax revenue coexists with uncomfortable questions about who bears the environmental and health costs.
SpaceX also aims to extend part of this infrastructure to space. According to plans, the company is exploring AI computation demonstrations in orbit from 2027 and has proposed satellite architectures capable of acting as space data centers. The goal is to leverage solar energy, launch experience, and Starlink networks to build computing capacity outside Earth’s surface.
This is an ambitious proposal but still filled with uncertainties. Launching servers into space could alleviate some terrestrial energy issues but introduces other challenges: maintenance, radiation, thermal dissipation, hardware lifespan, communications, launch costs, orbital debris, and security. For now, orbital data centers remain a strategic bet rather than an immediate alternative to traditional cloud regions.
What matters is that SpaceX approaches AI not merely as a digital product but as an infrastructure race. Its rockets, satellites, factories, funding capabilities, and now data centers all fit into this perspective.
A Warning for Hyperscalers
AWS, Microsoft Azure, Google Cloud, Oracle Cloud, IBM Cloud, and Alibaba Cloud will not disappear because SpaceX enters the compute market. Their strength lies in the upper layers: managed services, developer ecosystems, security, enterprise integration, compliance, databases, analytics, and business relationships with thousands of companies.
SpaceX is playing a different game. It doesn’t need to persuade banks or governments to migrate their entire operations to a new cloud. It just needs to sell capacity to existing clients, models, and demand that require more GPUs than they can activate alone within the needed timeframe.
This can shift market dynamics. If major AI labs begin leasing capacity to non-hyperscale businesses, the value chain changes. The compute provider becomes a strategic figure—even without control over the final product. Practically, this can influence costs, availability, time-to-market, and competitive capacity.
From a financial perspective, the contracts with Google and Anthropic turn AI compute into a recurring revenue source for SpaceX just as the company seeks to strengthen its investor narrative. Rather than being just a space company, SpaceX can position itself as an infrastructure platform—covering connectivity, launches, satellites, data, and AI.
AI is sold as software, but profits come from infrastructure. That’s the core idea behind these deals. Models, scientific teams, and products matter. But without energy, chips, and data centers, everything else is just a demo.
SpaceX understands this part of the equation and has decided to take the less glamorous yet more profitable role: the landlord. Google and Anthropic can continue competing with Gemini and Claude. xAI can keep pushing Grok. But all are now operating in a market where the ground, electricity, and GPUs almost weigh as much as the research talent.
Frequently Asked Questions
What has Google signed with SpaceX?
Google has signed a compute capacity agreement paying $920 million per month from October 2026 to June 2029. The deal includes access to around 110,000 NVIDIA GPUs, CPUs, memory, and related infrastructure.
What is Anthropic’s relationship with Colossus?
Anthropic has agreed to use capacity on Colossus 1, the supercomputer built by xAI near Memphis. xAI states the installation has over 220,000 NVIDIA GPUs for training, inference, and high-performance computing tasks.
Is SpaceX already competing with AWS, Azure, or Google Cloud?
Not at the same level. SpaceX mainly acts as a provider of raw AI capacity. Hyperscalers offer a much broader range of managed cloud services, enterprise software, and developer tools.
Why are GPUs so critical for AI?
GPUs enable processing massive amounts of operations in parallel, which is essential for training and running generative AI models. The scarcity of GPUs and energy constraints have turned physical infrastructure into a competitive advantage.
