Big Tech planning massive AI spend

Amazon told investors it expects roughly $200 billion in capital expenditures for 2026, with management saying the bulk will fund AWS capacity and chip projects like Project Rainier. (cnbc.com) Alphabet projected 2026 capex between $175 billion and $185 billion as it expands data‑center footprint and TPU capacity for model training. (cnbc.com) Meta guided full‑year 2026 capex of $115 billion to $135 billion to fund two massive "Titan‑class" compute projects and a stepped‑up AI buildout. (investor.atmeta.com) Microsoft has sharply increased quarterly capex — reporting $34.9 billion in one recent quarter with about $11.1 billion allocated to leasing data‑center space and the rest to GPUs, CPUs and long‑lived infrastructure. (datacenterdynamics.com) Hyperscalers are combining third‑party GPUs with custom silicon: AWS has activated Project Rainier with nearly 500,000 Trainium2 chips for Anthropic, while Anthropic separately expanded use of Google Cloud TPUs for up to roughly one million TPU units. (aboutamazon.com) Vendors and system integrators are responding with validated, pre‑tested stacks — examples include NetApp and Cisco shipping validated designs for NVIDIA DGX SuperPODs and enterprise reference architectures to speed on‑prem AI factory deployments. (netapp.com) Tightness in accelerator supply and data‑center capacity is already visible: analyst reports show past H100 shortages and multi‑quarter lead times, commercial GPU clouds reporting broadly sold‑out 2026 capacity, and forecasts that AI will sharply raise data‑center power and occupancy into late 2026. (tomshardware.com)