Is the massive spending on AI actually paying off?

The companies building the most powerful AI, such as Microsoft, Google, and Meta, are investing heavily in data centers. These are giant facilities filled with thousands of specialized computer chips—the brains that process information for an AI model. An AI model is essentially a massive, complex software program trained on vast amounts of data to recognize patterns and generate new content. Companies are betting that these models will transform businesses and boost the global economy. However, recent data shows that revenues are not yet matching the staggering costs of building and maintaining this hardware. In fact, many smaller AI startups are already failing as they struggle to find paying customers or get outpaced by the larger platforms that fold similar features directly into their own products.

Building an AI prototype is relatively easy, but bringing that technology to the public requires a different level of engineering. Think of a prototype like a chef’s experiment in a home kitchen: it proves a dish is possible. Moving to production is like opening a global restaurant chain. You need to ensure the dish tastes the same every time, can be made thousands of times a day, and is safe and reliable. In AI, this means ensuring the software works consistently outside of a controlled lab. If the system is too expensive to run, or if users find it confusing or unhelpful, the business fails. Additionally, the chips inside these data centers become outdated every few years, requiring constant, expensive upgrades just to stay competitive. If these companies cannot make their systems significantly more efficient and productive, they risk owning a collection of expensive, obsolete machines.

The central question for the economy is whether this AI buildout will lead to a new era of growth or become a historic financial miscalculation. If these companies cannot generate enough revenue to cover their massive debts, the risks could ripple through the broader economy. We are at a stage where the novelty of AI is wearing off, and the real-world utility—the ability for these tools to actually improve work and profit—is being tested. Whether this era of AI succeeds will depend on whether these machines can move from being impressive experiments to essential, profitable tools that the average person and business can rely on every day.