Data‑center capex spike
- Data‑center construction and AI infrastructure spending have accelerated sharply as generative AI deployment scales. (x.com) - Construction hit about $2.5B per month, and hyperscalers plan roughly $700B of AI capex in 2026. (x.com) - Analysts estimate $930B in AI infra capex over six years, with power and grid capacity now the primary constraint. (x.com)
U.S. data-center construction has climbed to roughly $2.5 billion a month in inflation-adjusted terms, as the AI buildout shifts from software into concrete, steel and power hookups. (ourworldindata.org) Moody’s Ratings said in March that capital spending by six U.S. hyperscalers — Microsoft, Amazon Web Services, Meta, Alphabet, Oracle and CoreWeave — will likely reach about $700 billion in 2026, up nearly sixfold from 2022. (moodys.com) Alphabet told investors on February 4 that its 2026 capital expenditures are expected to run between $175 billion and $185 billion, driven by infrastructure demand tied to AI products and Google Cloud. (abc.xyz) Amazon has been adding project-by-project commitments on top of that broader spending wave, including up to $50 billion for artificial intelligence and supercomputing capacity for U.S. government customers and $25 billion for new Mississippi data centers announced in April. (aboutamazon.com, aboutamazon.com) Meta broke ground in February on a 1-gigawatt data-center campus in Lebanon, Indiana, said the site will cost more than $10 billion, and said it is pairing those campuses with long-term chip and power deals. (about.fb.com, about.fb.com) The spending boom reflects how generative AI works in practice: companies train models on huge clusters of chips, then keep paying to run those models every time users ask a question, generate an image or deploy an agent. JLL said AI could account for half of all data-center workloads by 2030. (jll.com) That is turning electricity into the main gating item. JLL said nearly 100 gigawatts of new data-center capacity will be added globally from 2026 to 2030, but warned that grid constraints will require new energy solutions. (jll.com) The Electric Power Research Institute said in March that U.S. data centers could consume 9% to 17% of the nation’s electricity by 2030, up from about 4% to 5% today, depending on how fast AI demand grows. (powering-intelligence.epri.com) Developers and operators are responding by moving toward power-rich markets, onsite generation and larger campuses. Bloom Energy said in its 2026 survey that power availability has become a “defining boundary” on data-center growth, while one in five campuses is expected to exceed 1 gigawatt by 2030. (bloomenergy.com) Moody’s said the buildout is accelerating revenue growth for big tech groups even as it erodes free cash flow and pushes some of them toward more borrowing. The next question is no longer whether companies want more computing capacity, but how fast utilities and contractors can deliver the power to run it. (moodys.com)
