What is actually happening?
Google reportedly agreed to pay SpaceX about 920 million dollars a month, from late 2026 through 2029, for roughly 110,000 GPUs, a contract reported to total around 30 billion dollars. Anthropic reportedly pays about 1.25 billion a month for similar capacity. Together that makes SpaceX something new: an AI compute landlord with more than 2 billion dollars a month in run-rate. Google's stated reason was bridge capacity for surging demand, with its enterprise AI platform reportedly growing 300 percent quarter over quarter.
Why does renting beat buying at the top?
Because capacity, not money, is the binding constraint. Even the largest companies cannot build data centers fast enough to match demand, so they rent, and they will rent even from a rival when that is where the chips are. The reported detail that Google's compute sits in xAI's facility is the tell: the scarce asset is not the model or the capital, it is the compute itself, and whoever controls it sets the terms.
What does this mean for your company?
Your AI cost is an operating expense, not a one-time purchase, and it scales with how much you use. That is easy to miss when a pilot is cheap and painful to discover when usage grows. The companies that stay in control treat AI compute the way they treat any critical recurring infrastructure: a named owner, a cost model tied to usage, and a plan for what happens when the bill doubles. The giants are showing the shape of the cost. The discipline is the same at any size.
Read next: Is a Single Misconfigured File Your Biggest AI Risk? · What Is the Most Dangerous Line on Your Balance Sheet?