It includes specialized processors such as GPUs and TPUs, fast networks that link thousands of chips together, large storage systems for data, power and cooling equipment, and the software that schedules and manages the work. Training a large model and serving it to millions of users both depend on this foundation.
The cost is very large, and it is a major reason why technology companies report rising capital expenditure. Cloud providers rent this capacity to customers, and chip makers, server builders and data-center operators all earn revenue from it.
The main risks are supply, power and demand. Advanced chips and electricity can be scarce, and building capacity takes time. If spending runs ahead of customer demand, companies are left with expensive, underused assets.
When you assess a company that depends on AI infrastructure, look at how much it spends, who is using the capacity, whether it is contracted in advance and how quickly it turns into revenue.