Amazon raised its 2026 capital expenditure forecast to $220 billion from $200 billion, citing soaring memory chip prices as the primary driver. CEO Andy Jassy emphasized that even at $220B, Amazon will lack sufficient capacity to meet customer demand in 2026 and 2027. AWS revenue jumped 37% to $42.2 billion in Q2, marking its fastest growth in 18 quarters, with AI and chip businesses each exceeding $25 billion in annualized run rates. AWS backlog surged to $496 billion.
Combined, Big Tech's four largest hyperscalers—Amazon ($220B), Google Alphabet ($195–205B), Meta ($125–145B), and Microsoft (tracking ~$190B)—will spend roughly $725 billion on AI infrastructure in 2026, up 77% from ~$410 billion in 2025. Memory-constrained supply chains are the new capex accelerant: Apple's Tim Cook warned of a "hundred year flood" in memory chip pricing; Tesla's Elon Musk called costs "insane." High-bandwidth memory feeds accelerators during training and inference, while conventional server memory supports the wider compute fabric. Amazon's Q2 free cash flow swung to negative $7.6 billion from positive $18.2 billion a year earlier, mirroring weakness across the sector.
For architects, the memory pinch is structural, not cyclical. Data center construction timelines stretch 2–3 years, while servers typically recoup costs in under 3 years. Jassy disclosed Amazon breaks even on infrastructure investments via 5–6 year server lifecycles and multi-year customer contracts. The capex arms race remains on pace to exceed $1 trillion in 2027. Memory-heavy AI clusters, high-bandwidth memory for inference, and custom silicon are now table-stakes. Builders should assume memory scarcity and pricing pressure persist through 2027.