AI spending concentrates

Five companies are on track to pour about $725 billion into artificial intelligence infrastructure in 2026, up from roughly $450 billion in 2025, according to a forecast highlighted after Jamie Dimon’s April 7 shareholder letter. The group is Microsoft, Amazon, Google, Meta, and Apple, and the jump is more than 60% in a single year. (nationaltoday.com) (fool.com) This is not spending on chatbots alone. Capital spending means the physical layer: data centers, specialized chips, power equipment, cooling systems, and the networking gear that moves data between machines. (goldmansachs.com) (cnbc.com) The money is concentrating because the biggest cloud companies already own the land, the servers, the customer relationships, and the balance sheets. If training a frontier model is like building a new airport, only a handful of companies can afford the runways, terminals, and power lines. (goldmansachs.com) (rbcwealthmanagement.com) You can see the scale in company guidance. Alphabet said on its February 4 earnings call that 2026 capital spending would be $175 billion to $185 billion, and Amazon’s 2026 plan reached about $200 billion, with most of it aimed at data centers. (abc.xyz) (cnbc.com) Microsoft is spending at a pace that already looks industrial rather than experimental. In its January 28 fiscal second-quarter call, the company reported $37.5 billion of capital expenditure in just one quarter as it expanded artificial intelligence infrastructure. (finance.yahoo.com) (fool.com) Wall Street had already been moving in this direction before Dimon’s letter. Goldman Sachs Research said consensus estimates for hyperscaler capital spending in 2026 had risen to $527 billion, while Bloomberg-reported company plans for four firms alone pointed to roughly $650 billion this year. (goldmansachs.com) (finance.yahoo.com) The winners from a buildout like this are usually the companies selling picks and shovels. Nvidia supplies the graphics processing units used for training, Broadcom sells networking and custom chip technology, and data-center operators benefit when tenants need more power and floor space. (goldmansachs.com) (rbcwealthmanagement.com) The pressure lands on smaller software companies from the other direction. If a company rents artificial intelligence from Microsoft Azure or Amazon Web Services instead of owning the infrastructure, its costs rise with every heavy query, so it has to prove that each new feature brings in real revenue or cuts real labor. (fool.com) (goldmansachs.com) That is why the same artificial intelligence boom can make one balance sheet stronger and another weaker. The platform that owns the server farm can spread costs across cloud contracts, advertising, search, and subscriptions, while a smaller app company may be paying retail prices for the same computing power. (rbcwealthmanagement.com) (goldmansachs.com) So the headline is not just that spending is rising. It is that artificial intelligence is starting to look less like a broad software wave and more like a utility business, where a few firms build the grid and everyone else decides how much expensive electricity they can afford to use. (cnbc.com) (goldmansachs.com)