🧠 Food for thought

Implications, context, and why it matters.

Capacity shortages are driving the spending

• In the fourth quarter of 2025, Microsoft, Meta, Alphabet, and Amazon hit limits on available AI computing capacity 1.
• Forecasts put 2026 capital expenditure for the four hyperscalers, very large cloud computing companies, near US$700 billion 2.
• Commentators have compared that figure to the entire economy of Sweden 3.
• Some investors doubt the pace, while a group of analysts calls it a sensible bet to cover AI-driven demand 4.

Big buildouts are pushing longer, stricter deals

• To lock in scarce supply, some cloud computing and AI infrastructure buyers are signing take-or-pay contracts, which require payment for a minimum amount of product such as semiconductor wafers, used or not 5.
• For chipmaking equipment and semiconductor suppliers, these terms can offer a clearer view of orders plus steadier cash flows, which lowers risk and supports capacity expansion 6.
• Some writers say the setup can cause friction when rigid agreements, such as five-year take-or-pay graphics processing unit (GPU) leases, run into customers who prefer usage-based pricing 7.