As shown in the chart, that remains below prior peak infrastructure cycles. Railroad capex reached ~6% of GDP at peak buildout, and internet infrastructure spending approached roughly ~4% during the late 1990s. Relative to those precedents, today’s AI buildout appears earlier sitting at ~1% of GDP.
Why this matters
Large capital cycles rarely move in straight lines. Early phases often feature concentrated investment, uneven returns across participants, and meaningful dispersion in outcomes. Periods of heavy infrastructure buildout can lead to over earning in certain segments, misallocation in others, and ultimately significant divergence between durable value creation and short term enthusiasm.
Importantly, infrastructure intensity is occurring alongside measurable enterprise adoption. As discussed in our recent webinar, early evidence on return on investment has been constructive. According to Accountable Acceleration: Gen AI Fast-Tracks Into the Enterprise (Wharton Human-AI Research and GBK Collective, October 2025), approximately 72% of enterprises are formally measuring Gen AI ROI, and roughly three quarters report positive returns. Four out of five leaders expect positive ROI within two to three years.
Historically, productivity gains from new infrastructure cycles take time to diffuse. The fact that enterprises are already reporting measurable returns at this stage of the buildout is notable.
The key is that paid business usage grows at about an equal or greater rate than hyperscaler capex. As the chart below demonstrates, that’s currently the case. As long as that continues, the investments make sense.
At Patient Capital, we work diligently to sort through the details of how business fundamentals compare to market expectations. We still like Alphabet, Amazon, Meta, and Nvidia.
Amidst the current turmoil, we are working hard to find the best opportunities. As you might expect, software is a current area of focus.
Share