AI Capex

AI Capex, short for Artificial Intelligence capital expenditure, refers to investments in long-lived assets supporting Generative AI and other large-scale Machine Learning workloads, notably by Hyperscalers such as Microsoft, Alphabet, and Meta. It commonly includes Servers, AI Accelerators—including GPUs supplied by NVIDIA and custom-designed chips—land, Data Centers, power systems, Networking, and cooling infrastructure. The composition varies by organization and workload; GPUs are not universally the principal expense, and Liquid Cooling is one cooling option rather than a standard requirement. Qualifying purchases of property, plant, and equipment are generally capitalized, with depreciable assets depreciated over their estimated useful lives under applicable accounting rules; land is generally not depreciated. Electricity, personnel, Cloud Services consumption, maintenance, and some software or research costs are generally expensed rather than capitalized, although treatment depends on the circumstances and accounting rules. The infrastructure supports Model Training and Inference, including for Large Language Models, but its components have materially different useful lives: buildings generally last longer than servers and accelerators, whose useful lives are reassessed as technological obsolescence can occur relatively quickly. AI-related capital expenditure has surged among major hyperscalers seeking additional Compute capacity and is closely monitored by investors, but it is not inherently a reliable leading indicator of future revenue. Assessments also consider demonstrated demand, utilization, backlog, margins, depreciation, financing, and returns on invested capital; spending can create revenue shortfalls or profitability risks if anticipated demand does not materialize.