AI infrastructure, investment pressure, and social acceptance

The most interesting question about AI is increasingly not whether the technology works.

It is whether, and when, the economics work.

Data centers, chips, energy, valuations, geopolitical dependencies: what started as a software debate has become an infrastructure debate. That is where it becomes uncomfortable.

The large tech companies are investing hundreds of billions into new capacity. At the same time, many enterprise surveys show that measurable business value is arriving more slowly than the investment story suggests.

Resistance is also growing. At US universities, pro-AI commencement speakers have been booed. Local communities are pushing back against new data centers because electricity, water, and land are becoming very concrete costs. And export restrictions around new Anthropic models show that AI is not just software, but geopolitical infrastructure.

The parallel to 1999 is therefore not the technology itself. The internet was not an illusion either. What was overestimated was not the internet, but the speed at which technical possibility would turn into durable business models.

AI is now facing a similar pressure. Infrastructure is being pre-financed at massive scale, while productive use in many companies still lags behind. As long as capital is cheap, expectations are high, and strategic fear is strong enough, that gap can be covered up. It will not disappear permanently. It will only close through measurable value and social acceptance.