Big Tech to spend $720B on AI in 2026
An analysis estimates Big Tech will spend $720 billion on AI in 2026, a capex surge that intensifies pressure on foundries, supply chains and GPU procurement timelines. That scale of spending reshapes vendor leverage and prioritization in hardware allocations. (fool.com)
The five largest cloud providers published 2026 capex ranges: Meta $115–135 billion, Amazon $200 billion, Microsoft ~$150 billion (annualized run rate), Alphabet $175–185 billion, and Oracle $50 billion. (fool.com: ) TSMC told investors it expects 2026 capital spending between US$52 billion and US$56 billion as it ramps advanced-node capacity. (investor.tsmc.com: ) The company’s board also approved a US$44.96 billion capital budget to expand advanced and specialty fabs, packaging and testing capacity amid rising AI demand. (focus.taiwan: ) Independent analysts estimate AI wafers will consume the bulk of leading-edge N3 capacity—projected near ~86% of N3 wafer demand by 2027—and warn HBM memory and CoWoS/advanced packaging are the choke points for accelerator production. (the-decoder.com: clarifai.com: ) Hyperscalers are diversifying silicon: AWS continues to expand its Trainium/Inferentia accelerator family and Google is rolling out its Ironwood TPU (a pod-scale inference design that can scale to 9,216 chips), both moves aimed at shifting inference volumes off general‑purpose GPUs. (aws.amazon.com: blog.google: ) NVIDIA has leaned into partnerships with specialized cloud providers to relieve supply pressure—announcing a $2 billion investment in CoreWeave and expanded collaboration to accelerate multi‑generation NVIDIA deployments and a planned 5‑gigawatt AI factory buildout by 2030. (nvidianews.nvidia.com: techcrunch.com: ) Spot market distortions and delivery lead times have widened this year, driving startups toward GPU‑first clouds and rental marketplaces (CoreWeave, Lambda, RunPod and others) as interim supply channels while advanced-node wafer and packaging capacity scale. (clarifai.com: gpu.fm: )
