Until recently, investors viewed the AI trade through a simple lens: it’s an Anthropic and NVIDIA world, and everyone else is living in it. With that logic, the market steadily sold off tech services, market research, SaaS and data providers. Hundreds of billions of dollars in value evaporated on the theory that AI would eat knowledge work.
However, in the past few months, something very different is afoot. Those old-line sectors are healing, and it’s the AI infrastructure buildout itself that is showing cracks. The firms getting eaten, it turns out, own the pipes.
Last week CoreWeave, the poster child of AI infrastructure, lost a quarter of its value in five days, and its bonds traded like distressed debt. Jensen Huang vouched for it personally, but the stock fell again the next day. Meanwhile, Accenture had its best day ever this month and is up about 70% in three months. Cognizant is up 76% from its June lows, and Gartner is up nearly half since July.
This fits a historical pattern. Phase 1 of an innovation boom centers on the new infrastructure or power source (and often too much money flows in). Phase 2, which creates more durable value, is about applying new solutions on top of it.
A century ago, Samuel Insull built America’s mightiest electric utility empire…and watched it collapse. In 1999, the fiber barons laid so much glass that most of it sat dark for years (remember Global Crossing and Worldcom?). The electricity was real, and so was the Internet. But the lasting fortunes went to the second wave: the people who figured out what to plug into it. For electricity, that meant the assembly line, household appliances, radio and television. For the Internet, it meant Amazon, Netflix, and the Cloud.
Today’s game has now shifted to bending AI’s power into offerings that are custom and hard to copy: new workflows, new products, new experiences. And (surprise surprise) that requires insight, consulting, and solutions.
Big buyers are driving the shift. Like industrial leaders a century ago, the Fortune 493 have realized that the new, yet ubiquitous, power source only yields proprietary advantage through proprietary application. So they are being judicious about letting generic AI and agents into their businesses. For example:
JPMorgan won’t let outside agents touch customer accounts.
Schwab and Amazon have blocked Meta’s Muse.
Delta Airlines refuses third-party agents.
Yelp admits only agents that pay.
Most are building their own agents instead. They have worked out that the advantage isn’t the model anyone can rent. It’s the custom application no one else can copy. They intend to own the new means of production. And they need insight, consulting, and data to navigate the way.
The opportunity was never just the new energy or information source. It’s in what you do with it.
The week's most important AI news, curated for you:
