Why the AI gold rush is hitting a reality check

The industry is currently split on how to handle the immense cost of AI. Companies like Amazon are continuing to invest heavily in data centers and specialized computer chips, betting that demand for AI services will justify the high price tag. However, the financial pressure is showing elsewhere. Reddit is seeing its search traffic become unpredictable as AI summaries change how people find information. Meanwhile, companies like Apple are looking toward subscriptions—essentially asking users to pay for extra AI access—to help cover the costs of their increasingly powerful AI features. Even LinkedIn has bucked the trend by announcing it will keep its hardware footprint flat for the next year, choosing to optimize its existing technology rather than buying more equipment.

At the heart of this tension are the specialized computer chips known as GPUs, or graphics processing units. These are the engines that train models—the complex mathematical systems that handle tasks like writing text or recognizing images. Because training these models requires massive amounts of data and constant calculations, companies have been rushing to assemble thousands of these GPUs into massive data centers. When these chips sit idle, they are wasted money, much like a plane parked on a runway while it could be flying. LinkedIn’s recent success came from better allocating their projects to ensure their chips stay busy, and using a technique called distillation, where engineers train a smaller, more efficient model by having it learn from a larger, more expensive one. By making the smaller model just as capable for specific tasks, they avoid the need for more energy-intensive, expensive hardware.

The fundamental question facing the industry is whether the demand for AI will actually keep growing enough to justify these massive investments. For companies selling cloud hosting, AI is a double-edged sword: they want to sell as much space and power as possible, but if their customers—the startups and apps building the AI—find the costs unsustainable, those revenue streams could dry up. We are seeing a transition from an era where businesses simply bought more to solve every problem, to a new phase where success is defined by who can get the most work out of the hardware they already own. The companies that thrive won't necessarily be the ones that spent the most, but the ones that learned to build with discipline.