The world’s largest technology companies are spending hundreds of billions of dollars building AI infrastructure. New data centers, specialized chips, and massive electrical systems are essential to power the enormous amount of computing AI requires. No company wants to be left behind, so the spending has become a race in which participation seems mandatory.
There’s no question AI is useful. Millions of people use it every day to search for information, summarize documents, brainstorm ideas, write emails, generate software, and answer questions. Businesses are finding ways to automate routine tasks and improve productivity. AI has clearly earned a place in our digital toolbox.
But usefulness alone doesn’t create a viable business.
The promises surrounding AI have been extraordinary. We were told it would approach human reasoning, replace large numbers of knowledge workers, and fundamentally reshape the economy. While today’s large language models are remarkably capable, they still require human oversight. They hallucinate, confidently present incorrect information, and can rarely be trusted without verification.
Even so, AI can provide tremendous value in the right applications. A friend who works with home builders uses AI to analyze construction plans and quickly generate detailed material lists and cost estimates. What once took hours now takes minutes. That’s real productivity.
The larger question, however, is whether that value is enough to support the industry’s staggering investment.
Oddly, that discussion has been largely absent.
Every successful technology product has eventually had to answer a simple question: Can the revenue exceed the cost of delivering the service? Personal computers, smartphones, and streaming services all developed business models that customers were willing to support.
AI is different. Every prompt consumes expensive computing resources, and the cost of delivering those answers are very substantial. Most consumers, meanwhile, expect AI to be free or are willing to pay only a modest monthly subscription. I’m an enthusiastic user of ChatGPT, Gemini, and Claude, yet I doubt I’d pay much more than $20 a month for the value they all currently provide.
That leads to the key question: If customers won’t pay enough to cover the enormous cost of building and operating the AI infrastructure, where do the profits come from?
While never answering this question, the industry continues to invest at an unprecedented rate. Companies are committing hundreds of billions of dollars to new data centers under the assumption that the economics will eventually take care of themselves.
In more than 50 years in the technology industry, I’ve never seen anything like this. The enthusiasm is understandable—AI may ultimately transform computing as profoundly as the internet did. But enthusiasm isn’t a business model.
Before we celebrate AI as the next great economic revolution, investors should ask the question that has determined the success of every major technology product:
Has anyone actually demonstrated how this business will earn an attractive return on the extraordinary capital being invested today?
This subject recalls the early 1990s, when businesses were debating whether to invest in creating websites. “What will we do with a website?” one boardroom exec would ask. “We don’t know,” answered another, “but if we’re not well positioned when we figure it out, we’ll be too late.” I’m not sure that was sound logic then. Or now.