Why AI Will Cause One of the Worst Crashes in Market History
By: Farzad Hussain – Investment Associate
Contributing Editors: Nathan Li, Caden Warner
The market has realized that AI is a bubble.
The market is overvalued because it is pricing in a future that current technology cannot realistically reach. The entire AI super cycle is built on the belief that today’s Large Language Model (LLMs) (think Chat GPT and Gemini), built on next-work prediction, will somehow evolve into full Artificial General Intelligence (AGI) fast enough to justify the trillions of dollars pouring into GPUs, data centers, and power infrastructure.
The first warning sign is spending. McKinsey estimates about $5.2 trillion in AI related data center spending by the end of the decade. But the technology is not keeping pace with the costs. The newest models, such as GPT-5, are only showing incremental improvements to their predecessors. As energy costs rise, hyper-scaling requires more cash for ever diminishing returns in capability. At the current trajectory, the marginal trillion in CapEx buys better auto complete, not AGI.
The financial demand around the AI boom is structured in a bubble-formed manner. The industry has drifted into circular financing, where vendors help fund the purchasing power of their own customers. When Nvidia signs a deal with Open AI they then give that money to a hyperscaler like Oracle, who in turn buys chips from Nvidia, Nvidia then invests in a different company and the cycle repeats itself. On the surface, it looks like revenue for all three companies involved is increasing when in reality it's just the same money cycling around inflating their numbers. This mirrors the telecom bubble, when major equipment makers invested in carriers, who then used the money to buy more equipment. Revenues looked great, right up until they popped. AI is showing the same pattern.
As the capability of current models flattens out and AGI remains out of reach, the industry is already starting to shift toward Edge AI. Edge AI runs models directly on devices: think phones, cars, wearables and medical devices, rather than on cloud clusters (which is what they do now). Computing happens on the device itself, reducing latency, costs, and power consumption. Analysts now expect Edge AI devices to take a large share of AI workloads by 2030. If that happens, the current economics for today's enormous cloud-based data centers weakens because many of the workloads driving their buildout will migrate off the cloud entirely.
But none of these factors matter as much as economics. According to Bain, to sustain the expected $500 billion per year in AI-related datacenter CapEx, the industry needs to generate $2 trillion a year in AI revenue. Even under bullish pricing and adoption assumptions, Bain still sees an $800 billion annual shortfall, only shored up by enormous federal subsidies fueled by geopolitical rivalry and false hopes.
It's a clear picture. Record breaking CapEx is being deployed to scale next-world prediction systems that are slowing down in improvement. Circular financing is inflating demand. A structural pivot towards Edge AI threatens to cannibalize cloud workloads. And the revenue required to justify the investment simply will not materialize. This bubble is built on the false hope of imminent AGI—when reality catches up with valuations, what we see today will become one of the worst crashes on record.