Capital expenditure (capex) is money a company spends on long-term physical assets — in the AI context, the data centers, power infrastructure, networking gear, and AI accelerators (chiefly GPUs) that hyperscalers and AI labs build out to train and serve models.
How it works
Unlike operating expenses (electricity, salaries, software licenses), capex purchases are capitalized on the balance sheet and depreciated over years. Hyperscaler quarterly results disclose capex guidance, which analysts then use as a leading indicator of compute supply. Goldman Sachs and BloombergNEF both expect combined hyperscaler capex to exceed $600 billion in 2026 — about a 36% increase over 2025.
Why it matters
Capex is the most widely-watched real-economy signal of AI investment. IDC isolates the AI-specific slice and forecasts $487 billion in AI infrastructure spending in 2026 (~53% YoY), exceeding $1 trillion by 2029. For the full breakdown by hyperscaler, see AI Investment & Funding 2026.
Related terms: Hyperscaler · GPU · AI Inference · All glossary entries