For months, skeptics have warned that the artificial intelligence boom is headed for a crash, arguing that runaway capital expenditures will eventually lead to a glut of computing power and plummeting rental rates. However, recent data suggests the exact opposite is happening. Rather than reaching a peak, the AI compute market appears to be entering a massive new growth cycle characterized by skyrocketing prices and an insatiable appetite for hardware that defies traditional economic gravity.
Evidence of this surge is most visible among the major providers of merchant AI compute. Companies like CoreWeave and Nebius are reporting significant price increases even as they aggressively expand their capacity. In several cases, short term deals are fetching premium rates far above previous benchmarks, while long term contracts are seeing strong prepayments from customers eager to lock in access. Industry leaders note that customers are finally starting to generate real revenue from their AI products, making them more than willing to pay higher premiums for the processing power required to scale their operations.
The scale of current planning indicates that the industry is preparing for a world measured in gigawatts rather than megawatts. Hyperscalers are pushing projected investments toward staggering sums, with some estimates suggesting global AI investment could exceed one trillion dollars by 2026. This expansion isn’t limited to established tech giants either. Former outliers like SpaceX and Meta are evolving into hyperscale operators themselves, shifting from potential suppliers to some of the largest buyers in the ecosystem. This creates a feedback loop where every new attempt to solve the capacity shortage only drives further demand for infrastructure.
Beyond the chips themselves, this cycle is triggering a fundamental reengineering of how capital flows through the tech sector. Because data center shells last decades while the silicon inside is replaced every few years, these facilities act as long term assets that become increasingly profitable over time. The financial world is reacting accordingly, with GPUs being securitized into tradable assets and new futures markets emerging for compute power. With lead times for critical components like transformers stretching into several years, it is becoming clear that the bottleneck isn’t lack of interest, but physical reality. All signs point to a build out that is still accelerating rather than rolling over.