AI spending concentrates on infrastructure
Major firms are committing big budgets to chips and cloud capacity while existing data‑centre builds are being delayed or cancelled. (stocktitan.net) One report links a long‑term Meta deal with CoreWeave worth roughly $21 billion through 2032 to that same trend. (timothysykes.com) At the same time, industry coverage finds more than half of planned U.S. data‑centre projects have been delayed or cancelled, tightening available capacity. (fool.com)
Big artificial intelligence spending is moving toward the pipes and power behind the software, with Meta locking in long-term chip and cloud capacity as new United States data-center projects slip behind schedule. (investors.broadcom.com, investors.coreweave.com, techspot.com) On April 14, Broadcom and Meta said they extended their custom chip partnership through 2029 to support Meta Training and Inference Accelerator processors, with an initial commitment of more than 1 gigawatt and plans for a multi-gigawatt rollout. (investors.broadcom.com, cnbc.com) Five days earlier, CoreWeave said Meta expanded its cloud agreement by about $21 billion through December 2032, bringing the two companies’ total contracted relationship to about $35 billion. (investors.coreweave.com, cnbc.com) Those deals cover two different bottlenecks. Custom chips set the pace for training and inference inside Meta’s own systems, while rented cloud capacity gives Meta more computing power without waiting for every new building to be finished. (investors.broadcom.com, investors.coreweave.com) The construction side is tightening at the same time. Sightline Climate estimated that 30% to 50% of large United States data centers planned for 2026 will be delayed or canceled, and only about 5 gigawatts of the 12 to 16 gigawatts planned for this year are under construction. (techspot.com, webhosting.today) Meta has already told investors how large the bill could get. In its January 29 results, the company said it expects 2026 capital expenditures of $115 billion to $135 billion, up from $72.22 billion in 2025, with growth driven by infrastructure and cloud spending. (investor.atmeta.com, cnbc.com) That spending pattern is spreading across the market for artificial intelligence hardware. Companies that already control chips, networking gear, and operating data centers can sell capacity now, while developers waiting on new sites face transformer shortages, grid constraints, and supply-chain delays. (investors.broadcom.com, techspot.com, bloomberg.com) CoreWeave has been one of the clearest beneficiaries. Its April 9 announcement said the expanded Meta contract will support inference workloads across multiple locations, including some early deployments of Nvidia’s Vera Rubin platform. (investors.coreweave.com, ciodive.com) Broadcom is making a similar case on the chip side. Its release said the Meta program uses Broadcom technology as the backbone for new artificial intelligence data centers and includes what the companies called the first 2-nanometer artificial intelligence compute accelerator. (investors.broadcom.com, markets.businessinsider.com) The result is a narrower market than the artificial intelligence boom can make it appear: more money is being committed, but more of it is being tied up in chips, cloud leases, and existing facilities that can deliver capacity before delayed projects do. (investor.atmeta.com, investors.coreweave.com, techspot.com)
