When the AI boom meets the real economy

For the past two years, conversations surrounding AI have primarily centered on software development. This has led to debates about models, benchmarks, agents and the competitive race between technology companies. However, an often-overlooked story is emerging as the AI boom progresses. The growth of AI is not just about creating digital products; it is also driving physical investment in infrastructure.

Data centers are being constructed in various regions, utilities are expanding capacity, and water systems are becoming part of planning discussions that previously seemed unrelated to software development. AI is increasingly viewed as infrastructure, rather than just an application layer. This comparison to the dot-com era often focuses on valuations and whether technology stocks have become detached from reality.

However, a more pertinent question may be where the money is being directed. The dot-com boom generated optimism, creating companies, websites and fortunes based on expectations of a digital future that had not yet materialized. The AI boom is also encountering physical constraints that could lead to a similar outcome. Historical technological revolutions, such as railways, electrification and the internet, all required significant physical resources and faced constraints that slowed their progress.

The AI boom is beginning to confront similar challenges, particularly with energy, water, and capital requirements that are slower to adapt than software. Energy is a critical factor, as AI models increasingly rely on electricity generation and distribution systems. Water consumption and emissions from data centers are also becoming externalities that extend beyond the companies developing AI models, impacting utilities, communities, and society at large.

Capital investment in AI is substantial, with data centers, chips, and network infrastructure requiring large upfront costs. The assumption is that future demand will justify these expenditures, but historical trends show that assumptions about future demand can be vulnerable. During periods of rapid growth, attention often focuses on user adoption and market share rather than profitability.

Many AI tools are priced as if the full cost of running them can be deferred, which may not accurately reflect the true cost of compute, energy, chips, and infrastructure. Consumers may be subsidizing these costs through investor-funded funding gaps. The industry's pricing mechanism through token consumption may not align with customer value, especially when scaling usage.

As models become more capable, customers may begin to scrutinize costs at a different level, potentially leading to a disconnect between pricing and actual business value. The AI economy is transitioning from a software-centric model to one that is more dependent on physical systems like power grids, construction, and resource allocation.

While this does not necessarily mean the AI boom will fail, it highlights the importance of timing and distinguishing between a technology that will eventually change the world and an investment cycle that has outpaced reality.