The numbers are staggering. Microsoft, Amazon, Google and Meta are spending over a hundred billion dollars combined on AI data centres this year alone. That is more than the entire global semiconductor industry spent on fabrication plants in 2023. The justification is that demand for AI inference will eventually absorb all that capacity. But the revenue numbers tell a different story. OpenAI is burning cash. Anthropic is burning cash. Every major AI lab relies on investor money or parent company subsidies to pay for compute. The only company making real money from AI chips is Nvidia, and even its growth rate is slowing.
Signs of strain are appearing. Some hyperscalers are already delaying data centre builds. Venture funding for AI startups has dropped from its peak. The cost of running a large language model at scale is still higher than the value it generates in most applications. Enterprise adoption is real but shallow. Most companies are experimenting, not deploying at scale. The assumption that demand will grow exponentially forever is a extrapolation of a hype curve, not a fundamental law.
The bubble will burst not because AI is useless but because the spending has overshot the actual market need. Infrastructure built on borrowed time and inflated expectations will eventually face a reckoning. When the next earnings cycle shows weaker than expected cloud revenue growth, the cuts will begin.
Paul