Money Is Moving to AI Infrastructure
Investors are rotating away from seat‑based software and toward semiconductors, GPUs and data‑centre infrastructure—the businesses selling compute to AI builders. ( ) That shift showed up this week in outsized infrastructure deals and funding, and it’s repricing companies tied to chips, power and cloud operations rather than traditional software licensing models. (reuters.com)
A software company used to win by selling more seats, like adding more office chairs. This week, one of the hottest deals in tech was Meta agreeing to buy about $21 billion of cloud capacity from CoreWeave through December 2032, which is more like renting an entire power plant than adding a few desks. (reuters.com) CoreWeave is not selling chatbots to office workers. It is selling access to giant clusters of Nvidia graphics processing units, the specialized chips that train and run artificial intelligence models inside data centers. (coreweave.com) Meta did not start from zero here. The new $21 billion commitment sits on top of a prior arrangement worth $14.2 billion, taking the relationship to roughly $35 billion as Meta races to secure enough computing capacity for its artificial intelligence systems. (cnbc.com) That is why investors are revaluing the supply chain under artificial intelligence rather than just the apps on top of it. If a model maker cannot get chips, electricity, networking gear, and data-center space, it cannot ship new features no matter how good its software team is. (reuters.com) The same turn showed up in OpenAI’s funding round on March 31, 2026. OpenAI said it closed $122 billion in committed capital at an $852 billion post-money valuation and said the money would go to next-generation compute and global infrastructure, not just research hiring. (openai.com) Once that much money is aimed at compute, the winners change. Chip makers, cloud lessors, fiber-network builders, cooling-system vendors, and power suppliers start to look less like background utilities and more like toll roads collecting fees from every model that wants to scale. (openai.com) CoreWeave’s own contract shows what buyers are paying for now. The company said the Meta capacity will be spread across multiple locations and will include some of the first deployments of Nvidia’s Vera Rubin platform, which means customers are paying in advance to lock up the next wave of scarce hardware. (coreweave.com) The scramble is spreading past chips into the plumbing between them. Aria Networks said on April 7 that it raised $125 million in its first funding round to build networking infrastructure for artificial intelligence, a sign that even the cables and switches connecting machines are now attracting growth capital. (reuters.com) Venture data is starting to reflect the same preference. Crunchbase said global startup funding hit $300 billion in the first quarter of 2026, with the surge driven by spending on artificial intelligence compute and frontier labs rather than ordinary software startups. (crunchbase.com) That does not mean software stops mattering. It means the market is acting like the scarce thing in 2026 is no longer another subscription seat but the physical capacity to train, serve, and power artificial intelligence at industrial scale. (reuters.com)
