AI budgets set to triple
Enterprises are planning to sharply increase AI infrastructure spending — Deloitte says budgets are likely to triple as demand for AI workloads grows. (ciodive.com) That spending is already showing up in chip makers: Taiwan Semiconductor reported first‑quarter revenue up 35.1% year on year, which analysts link to resilient AI demand. (benzinga.com)
Companies are no longer treating artificial intelligence like a software subscription. Deloitte says most enterprises expect their artificial intelligence infrastructure budgets to triple by 2028, which means more money for chips, servers, networking gear, and power instead of just chatbots on employee laptops. (deloitte.com, ciodive.com) That shift starts with where artificial intelligence runs. A large language model needs racks of specialized computers inside data centers, the same way a factory needs heavy machinery on a shop floor instead of just office desks. (deloitte.com) Deloitte says more than 70% of surveyed respondents expect to operate “artificial intelligence factories” at scale by 2028. In plain terms, that means companies expect artificial intelligence to become a permanent production system, not a side experiment in one department. (deloitte.com) The survey behind that forecast covered 515 United States enterprise leaders in December 2025. Almost half said they already have more than 30 artificial intelligence pilots in progress, which helps explain why spending is moving from trials to plumbing. (ciodive.com) Most of those companies do not want to bet on one setup. Deloitte says many are choosing a hybrid model, which means keeping some computing in their own buildings while renting more from cloud providers when demand spikes. (ciodive.com, deloitte.com) That choice gets expensive fast because artificial intelligence uses far more electricity than older corporate software. Deloitte estimates power demand from artificial intelligence data centers in the United States could rise from 4 gigawatts in 2024 to 123 gigawatts by 2035, which is more than a thirtyfold jump. (deloitte.com) One Deloitte example shows how steep the jump can be inside a single site. A five-acre data center that adds graphics processing units to central processing units can see power use rise from 5 megawatts to 50 megawatts. (deloitte.com) The electric grid is already a bottleneck. Deloitte says some data center projects face waits of up to seven years just to connect to the grid, which turns artificial intelligence expansion into a race for land, transformers, and utility approvals as much as a race for software talent. (deloitte.com) You can see the spending wave in Taiwan Semiconductor Manufacturing Company before many companies even report their own budgets. On April 10, 2026, Taiwan Semiconductor said March revenue rose 45.2% from a year earlier and first-quarter revenue reached NT$1.134 trillion, up 35.1% from the same period in 2025. (pr.tsmc.com, investor.tsmc.com) Taiwan Semiconductor makes chips for companies like Nvidia and Advanced Micro Devices, so its monthly sales work like a readout from the middle of the artificial intelligence supply chain. When its revenue jumps this hard, it usually means customers are still placing large orders for the processors that fill new data centers. (pr.tsmc.com, sec.gov) So the story is not just that companies want more artificial intelligence. The story is that they are starting to buy the industrial base behind it: chips, buildings, cooling systems, and electricity contracts that take years to line up. (deloitte.com, deloitte.com, pr.tsmc.com)
