Intel just laid off 15 percent of its workforce. That is not a company about to sell billions of dollars of AI chips. It is a company admitting its core business is dying while the AI boom passes it by.
The narrative says hyperscalers will spend a trillion dollars on GPUs and data centres. The reality says corporate IT budgets are finite. Every dollar spent on Nvidia is a dollar not spent on Oracle, SAP or VMware. Those companies are reporting slowing growth. The displacement is real.
Meanwhile, the basic unit of AI economics is broken. Training a frontier model costs hundreds of millions. Running inference at scale costs more per query than the revenue it generates. OpenAI is reportedly spending more on compute than it makes in total revenue. Microsoft is embedding Copilot into everything at a loss to keep usage numbers up.
The public cloud providers are the canary. AWS, Azure and GCP all reported slower growth in their last quarters. Enterprises are optimising existing workloads, not launching new AI projects. The pilot phase is ending. The deployment phase is not beginning at the expected scale.
Capex commitments get cut when revenue misses. If you think this cycle is different, look at the debt loads on the companies building the infrastructure. They are levered to a demand curve that has not materialised.
The bubble will not pop with a bang. It will deflate as successive quarterly reports show the gap between spending and return on that spending.
Paul