Why Google's SpaceX deal signals the rise of the AI compute landlord

Google's $920-million-a-month SpaceX agreement shows how AI compute is turning into a scarce, rent-generating infrastructure asset

Every AI model needs enormous computing power to answer questions, write code, generate images, or analyse data. That computing power now needs so many specialised chips, data centres, cooling systems and electricity that even the world’s biggest technology companies are looking for outside capacity.

 

Google’s recent agreement to pay SpaceX $920 million a month for AI compute capacity is a sign of that shift: AI compute is becoming something companies can rent, price and sell like infrastructure. It shows how the AI race is moving from software and models to the hard infrastructure that powers them.

 

The deal covers roughly 110,000 Nvidia graphics processing units (GPUs), central processing units (CPUs), memory, and related components, with payments scheduled from October 2026 through June 2029, subject to ramp-up and termination terms, according to SpaceX's US Securities and Exchange Commission (SEC) filing. For Google, the arrangement appears to offer bridge capacity for rising AI demand. For SpaceX, which handles the business contracts and operations for renting out Nvidia GPU clusters, it turns compute infrastructure into recurring rental income.

 

 

Is AI compute now becoming a new corporate asset class?

In a standard cloud usage model, companies buy access to software services, storage, or general computing. In this case, Google is securing access to GPU-backed capacity, the physical backbone needed to run heavy AI workloads. A GPU, unlike a CPU, can process thousands of tasks at the same time, which makes it central to training and running modern AI models, which depend on huge volumes of repeated mathematical calculations.

 

The SpaceX deal suggests that compute is no longer just a service sold by traditional hyperscalers - the world’s largest cloud computing and data center operators - such as Google Cloud Platform (GCP). It is becoming an industrial asset that can be owned, leased, and priced. SpaceX’s earlier compute agreement with Anthropic, reported at up to $1.25 billion a month, points to the same pattern of large AI companies contracting capacity outside the usual cloud routes. 

Created using AI

Why AI needs more than just chips

According to an April 2025 report by McKinsey, the bigger challenge is building the industrial system around AI data centres that need far more than specialised chips. They require high-voltage power, land, grid connections, liquid cooling, high-bandwidth memory, networking gear, and large capital outlays to run heavy AI workloads. McKinsey estimates that global data centres may need $6.7 trillion in capital outlays by 2030, with AI workloads accounting for $5.2 trillion of that amount. In its report titled How can we meet AI’s insatiable demand for compute power, Bain estimates that global AI compute requirements could reach 200 gigawatts by 2030.

 

Energy is emerging as one of the sharpest constraints. The International Energy Agency (IEA), in its report Energy supply for AI, projects electricity supply for data centres to rise from 460 terawatt hours in 2024 to more than 1,000 terawatt hours in 2030 in its base case. Jones Lang LaSalle Incorporated's (JLL's) 2025 global data-centre outlook also flags demand growth despite supply and power constraints. This is why companies that are able to assemble power, cooling, chips, and construction capacity quickly can command strategic value.

The corporate model: AI compute as rental income

The Google-SpaceX agreement gives shape to a new corporate model: the AI compute landlord. A company that has secured GPU-heavy infrastructure can use it internally, lease it to outside customers, or do both. That turns expensive capital equipment into a potential revenue-generating asset.

 

This model resembles older infrastructure plays, for example, telecom tower companies lease tower space, warehouse owners rent logistics capacity, and power producers sell contracted electricity. In the AI economy, GPU clusters, powered shells, cooling systems, and data-centre campuses could perform a similar role.

Why Big Tech may still rent despite owning data centres

What stands out in this deal is that the buyer is Google itself, a hyperscaler that already runs one of the world’s biggest cloud and AI businesses. Building a large data centre can take years because companies need land, power approvals, cooling systems, and specialised equipment.

 

Public disclosures show how large this investment cycle has become. Microsoft, in its annual report for 2025-26, says it will continue investing in capital expenditure to support cloud growth and AI infrastructure. Amazon’s 2025 annual filing says cash capital expenditure reached $128.3 billion, primarily reflecting technology infrastructure and AWS growth. Meta reported $72.22 billion in capital expenditure, including finance leases, for full-year 2025.

 

Therefore, renting compute does not mean Big Tech has stopped building its own data centres. It suggests that these companies may need extra capacity while their own projects catch up with AI demand, which they are fulfilling through renting.

Can India’s data centres move up the AI value chain?

India’s data-centre market is expanding fast. According to JLL's India Data Centre Market Dynamics Report 2025, India’s data-centre inventory stood at 1,123 MW of IT load capacity as of H1 2025, with net take-up growing 48 per cent year-on-year. Colliers estimates India’s data-centre capacity at 1,263 MW as of April 2025 and expects it to cross 4,500 MW by 2030, attracting $20-25 billion in investments. CBRE's November 2025 report 'India’s Data Centre Market in a New Era' says India’s operational data-centre stock reached about 1,530 MW as of January-September 2025, driven by digitisation, cloud adoption and AI workloads.

 

The question for Indian operators now is whether they remain providers of land, power, and colocation, or move into higher-value GPU-backed AI compute services. The Google-SpaceX deal shows that in the AI economy, the winners may not only be companies that build models but also those who control the infrastructure needed to run them.