Investors question AI valuations

AI investing is starting to look less like one giant theme and more like a sorting exercise. That is the real shift. Investors still want exposure to artificial intelligence, but they are getting pickier about which part of the stack deserves huge multiples. Over the last few weeks, that split got hard to miss — with CoreWeave landing another massive infrastructure contract, Anthropic drawing eye-popping private-market offers, and strategists warning that software valuations are now unusually exposed to AI disruption. (money.usnews.com) ### Why are people suddenly talking about valuations again? Because the easy phase of the AI trade is over. For much of the last two years, investors could buy almost anything tied to AI and assume the market would reward the story. Now the questions are more basic — who actually gets paid, how durable that revenue is, and whether today’s price already assumes a near-perfect future. That change is showing up in both public stocks and private funding rounds. (money.usnews.com) ### What does the market still love? Infrastructure. The cleanest example is CoreWeave. On April 9, Meta expanded its agreement with the company by about $21 billion, on top of an earlier $14.2 billion deal. That is the kind of contract investors understand immediately — long duration, huge customer, visible demand for compute, and a direct link between AI spending and revenue. In other words, the picks-and-shovels case still feels tangible. (money.usnews.com) ### Why do infrastructure names get more slack? Because demand is easier to see than end-user monetization. The hyperscalers are still spending aggressively on data centers and chips, and that keeps the revenue base for GPU, networking, and cloud-capacity vendors looking real even when multiples wobble. Bloomberg’s tech analysis framed the recent pullback in AI semiconductor (money.usnews.com)y. (bloomberg.com) ### So where are investors getting nervous? Applications and software. Goldman said this week that AI is making investors rethink how much of U.S. equity valuations rests on long-term growth assumptions, especially in software. The worry is simple: if AI changes pricing power, product differentiation, or labor economics faster than expected, then some soft(bloomberg.com)gine and a margin threat. (money.usnews.com) ### What about the giant private valuations? That is where the tension gets loudest. Anthropic reportedly received offers valuing it at as much as $800 billion, more than double its February valuation level. You can read that two ways. One reading says frontier-model leaders are becoming the most valuable companies in the world before they are eve(money.usnews.com) be true at once. (finance.yahoo.com) ### Is this a bubble call? Not quite. It is more like a repricing debate inside the boom. The market is no longer treating all AI exposure as interchangeable. Compute suppliers with contracted demand look safer. Frontier labs with strategic scarcity can still command enormous prices. But companies selling thinner AI features on top of expensive models face a tougher question — can they keep enough gross margin after compute costs, competition, and falling prices? That is where reratings usually start. (money.usnews.com) ### What is the bottom line? Investors have not stopped believing in AI. They have stopped pretending every AI story deserves the same valuation. The new rule is harsher but healthier — show the contract, show the margin path, or expect the multiple to do the talking.