The economy’s new life support: AI data centers
AI infrastructure spending refers to the large-scale investment in data centers, networking, and power systems required to train and run advanced AI models, and it has become a central engine of economic growth even though there is little public visibility into how heavily this new infrastructure is being used or how sustainable the current pace of investment will prove to be over time. A Columbia Business School economist argues that this surge in data center building is now the single biggest driver of US economic growth and that without it, the economy would already be in recession. In other words, the construction of AI facilities is functioning as macroeconomic life support. That should make investors, policymakers, and workers uneasy, not triumphant. Growth pinned on one opaque boom is fragile by definition.
Bigger than railroads, powered by debt
According to Stijn Van Nieuwerburgh, AI infrastructure investment could average roughly 2.8% of GDP during the current buildout, exceeding even the historical peak of the railroad era. That comparison is not marketing spin; it is a warning about scale and concentration of risk. These facilities are far costlier than traditional data centers because AI racks draw vastly more power and require liquid cooling and dedicated substations, turning each site into an industrial project rather than a warehouse upgrade. The worrying part is how this is being financed. One Meta-linked campus was funded with a maze of special-purpose vehicles designed to keep debt off the company’s balance sheet. When an entire macro expansion depends on economic growth data centers financed in this opaque way, we are not building resilience; we are hiding fragility.
The AI capex boom is flying blind on traffic data
The most unsettling feature of the AI capex boom is not its size but its ignorance. Hyperscalers and their financiers are pouring resources into AI infrastructure spending without reliable public numbers on the traffic these systems will carry or the revenue that traffic will support. A review of more than 30 technical papers found no comprehensive public study of AI traffic volumes or patterns, and hyperscalers do not share such data. Cisco has begun publishing live AI inference traffic measurements, but even that is an early baseline and not a foundation for confident long-term planning. Meanwhile, estimates of annual AI-related capex already exceed US$600 billion and may be higher. This is infrastructure utilization transparency in name only: investors see soaring capex lines, not whether these data centers run full, half-empty, or at a loss.
Opacity all the way down: who holds the risk?
The opacity problem does not stop at traffic data; it extends into the financial plumbing that underwrites the buildout. Many hyperscalers prefer leasing AI campuses rather than owning them, which shifts the debt onto pension funds, private credit vehicles, and structured-finance entities that rarely appear in headlines. As Van Nieuwerburgh notes, that makes it harder to see where risk is accumulating or to time the buildout so that capacity arrives close to demand. At the same time, credit markets are flashing concern: spreads for major AI builders have widened, and one large neocloud provider now carries a market-implied default probability that would worry any long-term saver. When infrastructure utilization transparency is missing and the debt is scattered across hidden holders, a slowdown in AI enthusiasm would not be a tech story — it would be a systemic one.
A boom that needs proof, not hype
Two opposing risks now sit side by side. If AI demand disappoints, underused data centers and stressed borrowers could amplify a downturn. If AI adoption overshoots expectations, labor markets may face abrupt displacement — and we still will not know which facilities are essential and which are speculative because the industry’s core metrics are opaque. Both outcomes are worsened by the same flaw: a lack of reliable data on AI traffic, tokens processed, and unit economics. The reasonable path is not to halt AI infrastructure spending but to demand measurement and disclosure that match its macro importance. Policymakers, regulators, and large asset owners should push for standardized reporting on utilization, traffic, and financing structures. An economy propped up by AI data centers can be defensible, but only if we stop treating ignorance as a strategy.
