The current artificial intelligence boom is characterized by an unprecedented surge in infrastructure capital expenditures (CapEx), which Goldman Sachs predicts will grow from $765 billion this year to $1.6 trillion by 2031. Amidst high computing costs and market concentration, questions arise regarding the actual economic efficiency of these investments.
What Happened
Investments in AI infrastructure, including chips and data centers, are showing exponential growth. There is high market concentration: just 41 AI-related stocks account for nearly half of the S&P 500's market value. Meanwhile, corporate adoption of AI technologies has already reached approximately 80%, and model capabilities are doubling every four months.
Context
The US economy shows a critical dependence on the technology sector: in the first half of 2025, the IT industry accounted for 92% of total GDP growth. This creates a situation where any changes in AI demand or delays in data center construction could trigger significant volatility across the entire financial system.
Why It Matters for the Industry
The AI industry is entering a phase of extremely high CapEx, where the rate of spending growth outpaces proven return on investment (ROI). High inference costs, for example, in GPT-5.5 ($5 per million input tokens and $30 per million output tokens), create pressure on margins. Developers urgently need to optimize pipelines through RAG, agentic workflows, quantization, and distillation to justify high operating expenses.
Why It Matters for Users
For a wide range of users and businesses, this situation means that implementing complex LLMs will remain expensive, limiting their use in low-margin products. Furthermore, the high dependence of global markets on AI investments increases the risk of market turbulence for private investors.
What Remains Unknown / Limitations
There is an expert debate regarding the risks: while engineers and architects point to the threat of a bubble forming, product developers and enthusiasts emphasize the need to adapt to the rapid pace of model capability growth.
Sources
Author
Look at AI, Editorial Team