The stock market has begun killing its darlings, sliding into a brutal selloff that has rippled across the technology sector. Nvidia, the chipmaker synonymous with the AI boom, has shed around 17 percent of its value since its October high and hit a year-to-date low on February 5. Enterprise software stocks have lost roughly $1 trillion in market capitalization since late January, while in early February the Nasdaq was on track for its worst stretch since April. Amazon’s stock plunged around 10 percent in early February after the company forecast $200 billion in capital expenditures for 2026—exceeding Wall Street expectations of roughly $146 billion.
Yet the market has been telling two seemingly incompatible stories about artificial intelligence. The first story is about wasted spending. Four tech giants alone—Alphabet, Meta, Amazon and Microsoft—announced plans to spend around $650 billion on AI infrastructure this year. Amazon’s capex bombshell was only the latest. Alphabet said its 2026 spending could reach $185 billion, nearly double last year. Increasing skepticism over whether these colossal outlays will ever earn back their costs has hammered the stocks of the companies making the bets. Oracle, which has taken on roughly $250 billion in long-term data center leasing commitments, has seen its stock cut in half.
Photo illustration by Newsweek/Getty
But then there was a new story to emerge after Anthropic’s release of new plugins for Claude Cowork—an AI agent capable of automating complex knowledge work in legal, financial and marketing functions—triggered immediate selloffs across enterprise software companies. Asana has shed more than two-thirds of its value since its most recent high. Docusign and ServiceNow have each fallen more than half.
Stephen Yiu, lead portfolio manager at Blue Whale Capital in London, calls it “the equivalent of the DeepSeek moment” for software companies. “For a lot of companies that thought maybe AI is not coming into their territory, basically it’s mind-blowing. ‘Oh s***, maybe we now have to start seriously thinking that AI capabilities might fundamentally undermine existing business models,’” Yiu told Newsweek.
If AI agents can perform the functions that enterprise software currently handles—managing workflows, processing information, coordinating teams—but better and cheaper, the entire software as a service, or SaaS, industry faces an existential question. Why pay subscription fees for Salesforce or ServiceNow when AI agents can accomplish the same outcomes?
There’s a glaring apparent contradiction here: on one hand fears that demand for AI compute won’t materialize on the scale the hyperscalers are building; on the other a fear that the demand for AI will be so great as to level an entire sector. But perhaps this confusion isn’t irrational. It’s the predictable response to a specific phase of technological transformation: While everyone knows AI will be big, nobody knows yet who will win.
“The market hates uncertainty,” Bryan Wong, portfolio manager of the Osterweis Opportunity Fund, told Newsweek. Software companies “were priced on just the ability to have recurring subscription revenue forever and growth associated with that. And so now people are questioning terminal values.”
Consider a thought experiment. It’s August 12, 1994. You open your copy of The New York Times—still a print-only publication—and read the headline, “Attention Shoppers: Internet Is Open,” which announced “the first retail transaction on the internet”: $12.48 plus shipping for Sting’s latest CD. A light bulb goes off: This e-commerce is going to be huge, so huge it will disrupt all of brick-and-mortar retail! What could you do to make money on this prescient insight?
Very little, it turns out. The company featured in the article, NetMarket of Nashua, New Hampshire, didn’t grow into a tech giant. The next Times article to mention e-commerce, three weeks later, spotlighted early leaders America Online, Home Shopping Network and QVC—none of which proved to be good long-term bets. HSN lost more than 70 percent of its value from its 1994 price by 2000 and is now worth a fraction of even that.
To really profit from betting on e-commerce in 1994, you would have needed to know that a man named Jeff Bezos who worked at a hedge fund had weeks earlier started a company in his garage in suburban Seattle called Cadabra, that would eventually become Amazon—as well as also having the foresight to steer clear of Kozmo, Pets.com, eToys, Boo.com and the countless other startups that went belly up in the dot-com bust.
What if instead you decided to bet against the brick-and-mortar retail giant Walmart? That would have been a massive money loser: Since August 11, 1994, Walmart’s stock price has increased a mammoth 3,125 percent, and the company, after years of learning to master e-commerce, recently swelled its market capitalization to a tech-giant-like $1 trillion. As Yiu observes, Walmart “managed to defend themselves very successfully” against the e-commerce threat—and at 40 times forward earnings, the stock now trades at a higher multiple than Nvidia.
An investor reading the August 12, 1994 Times article with perfect understanding that e-commerce would revolutionize retail had no mechanism for identifying the actual winner, because the actual winner barely existed yet. But even more fundamentally, most of the people who invested in e-commerce during the dot-com boom missed what turned out to be the bigger digital business of the early 2000s: digital advertising, which powered companies like Google and Facebook, which also didn’t yet exist. As Wong noted, “You can go back to examples of Netscape and AOL, which were kind of the pioneers in their space and not necessarily the horses that won in the long run.”
The recent AI market turbulence reflects the same dynamic playing out in real time. AI capabilities are real and advancing rapidly, threatening to disrupt established software companies. But the path to monetizing those capabilities—and the infrastructure required to deliver them—remains fundamentally unclear.
Ben Barringer, an analyst at Quilter Cheviot, drew a useful distinction: The threat to software isn’t that these companies will vanish tomorrow, but that the risk of disruption “has gone from zero to something—whether it’s 30 percent or 50 percent.” For a sector whose valuations depended on the assumption of perpetual recurring revenue, even a modest increase in the probability of disruption demands a significant repricing. “You start playing around with your long-term growth rate,” he said, “and that can often mean big changes in your valuation.”
But Barringer also offered a critical nuance: Not all software is equally vulnerable. A company like Adobe, which builds tools to help marketing departments produce graphics, is fundamentally different from an enterprise resource planning system that runs a company’s core accounting and compliance functions. The deeper software is embedded in mission-critical operations and the more it provides not just functionality but risk mitigation, regulatory compliance and security, the harder it is to replace. “What I’m trying to say,” he told Newsweek, “is, are companies just selling you software, or are they selling you risk mitigation?” The market, for the moment, is not making these distinctions. It is, as Barringer put it, “shooting first, asking questions later.”
The two stories—excessive infrastructure spending and disruptive AI capability—aren’t actually contradictory. They’re different manifestations of the same uncertainty about who captures value in a massive wave of transformation that everyone can already see coming. A Cassandra scenario in which both sets of fears are true is also entirely plausible: the centralized, massive server farms don’t generate the returns their investors expect, and the technology proves supremely disruptive to incumbents. Something similar happened with the internet itself. As Barringer recalled, “Cable companies put billions of dollars into fiber optics and the network. They never got that money back. But the winners were the companies that used those rails to make money.”
The warnings from some AI researchers about current architectures potentially being dead ends add another layer of complexity. Significant voices among those who built today’s generative AI argue the future lies in fundamentally different approaches that will surpass current large language models. Companies building empires on today’s technology might face obsolescence just as they achieve dominance—the equivalent of betting everything on dial-up infrastructure in 1998. As I argued recently, AI may still be in the stage of the internet before HTML and web browsers were widely available.
Markets are pricing in this multilayered uncertainty by discounting everything. Nvidia might be the essential infrastructure provider for the AI age, or AI might not require the specific chips and scale Nvidia provides. Microsoft might successfully integrate AI and dominate the next era, or AI-native companies might make Microsoft’s stack obsolete. Oracle’s data center investments might generate enormous returns or represent $250 billion in stranded assets.
“If you think about 100 companies coming in to try to disrupt the other side of the fence, ultimately that 100 would be consolidated to 10,” said Yiu. A third or more of the capital pouring into AI will likely be wasted. That’s not a bug of the current moment—it’s the defining feature of every major technological transition. From an investor’s perspective, the challenge isn’t predicting whether AI will transform industries, it’s picking the survivors from a field that hasn’t finished forming. As Wong put it: “We’re going to have to assess business models and management teams and see how it plays out.” That is not a statement of ignorance. It is the only honest assessment of a market that has just realized what it doesn’t know.
