Infrastructure Winners Expand
Deals and capacity bets outside model-making are scaling up: MarketScreener reports CoreWeave struck a $21bn deal with Meta for AI‑cloud support, while Amazon’s chips business and AWS AI revenue are both cited at multi‑billion run‑rates. Those deals underscore that cloud, custom silicon and datacenter providers are capturing much of the economic upside from AI build‑outs. (marketscreener.com)
Meta just committed another $21 billion to rent artificial-intelligence computing from CoreWeave through December 2032, and that tells you where a lot of the money is going in this boom: not only to the companies building models, but to the companies renting out the machines. (usnews.com) (investors.coreweave.com) CoreWeave is not a household-name app company. It is a cloud provider that packs data centers with graphics processing units, which are the specialized chips used to train and run artificial-intelligence systems, and then leases that capacity to companies that need it fast. (coreweave.com) (usnews.com) Meta is buying that capacity because building enough of its own is slower than ordering more from a specialist. Reuters reported the expanded deal as Meta races to support more complex artificial-intelligence workloads after a disappointing model release last year. (wncy.com) (usnews.com) This is not Meta’s first CoreWeave order. CNBC reported the new $21 billion commitment comes on top of a prior $14.2 billion arrangement, which means the relationship now stretches far beyond a one-off capacity shortage. (cnbc.com) (bloomberg.com) At the same time, Amazon is showing that another layer of the stack is paying off too: the chips inside the cloud. In his shareholder letter published on April 9, Andy Jassy said Amazon’s chips business is now running at more than $20 billion a year, covering Graviton central processors, Trainium artificial-intelligence chips, and Nitro networking cards. (aboutamazon.com) (usnews.com) Amazon also said its Amazon Web Services artificial-intelligence revenue run rate topped $15 billion in the first quarter of 2026. That was the company’s first direct disclosure of how much annualized revenue its cloud artificial-intelligence services are producing. (aol.com) (finance.yahoo.com) Put those two numbers together and you get the shape of the market. One bucket is renting out finished artificial-intelligence services through Amazon Web Services, and the other bucket is selling the picks and shovels inside the cloud itself through Amazon-designed chips. (aboutamazon.com) (aol.com) That helps explain why investors keep hearing about data centers, power, and chip supply instead of only hearing about chatbots. Every new model needs warehouses full of servers, expensive networking gear, and enough electricity to keep those machines running around the clock. (investors.coreweave.com) (aboutamazon.com) CoreWeave’s announcement even said some of Meta’s new capacity will use Nvidia’s Vera Rubin platform, which shows how the money keeps flowing down the chain: from the model company, to the cloud provider, to the chip supplier. (coreweave.com) (investors.coreweave.com) The cleanest read on this week’s news is that artificial intelligence is turning into an infrastructure business as much as a software business. Meta is locking in outside capacity through 2032, and Amazon is already showing multi-billion-dollar annual revenue from the cloud and chip layers that sit underneath the models. (usnews.com) (aboutamazon.com)
