Is AI Infrastructure Dot-Com 2.0?

Outlier Capital · Outlier Capital · July 20, 2026 at 12:46 · ⏱ 12 min read  | Read on Substack ↗
Summary
The AI infrastructure buildout is not a repeat of the dot-com bubble because it is funded by record corporate profits, driven by verifiable demand from hyperscalers already monetizing AI, and constrained by supply-side bottlenecks (CoWoS, HBM). The viral crash narrative comparing SMH to Cisco's 90% collapse is rejected on earnings, capex economics, and institutional analysis from Goldman Sachs, making the multi-year setup asymmetric for investors in the physical-layer enablers of AI.
  • US corporate profits as a share of GDP hit 12.4% in Q1 2026, near all-time highs, contrasting with the profit deterioration before the 2000 bust.
  • Goldman Sachs models hyperscaler capex at ~$740B in 2026 and ~$996B in 2027, with cumulative AI-related spend reaching $7.6 trillion through 2031.
  • Hyperscalers (Microsoft, Amazon, Alphabet, Meta) are funding capex with internal cash flows and describe data center clusters as 'already pre-sold' with multi-year customer commitments.
  • Supply constraints persist in HBM, CoWoS advanced packaging, optics, and networking, with NVIDIA disclosing over $119B in forward supply commitments to secure capacity through at least 2027.
  • Goldman Sachs Portfolio Strategy explicitly states the dot-com comparison 'looks more reassuring than not' because profits are expanding and balance sheets are stable, unlike the late 1990s.
  • Valuation risks exist (elevated multiples, capex-to-revenue ratios) but the base case of a 90% sector collapse is unsupported by fundamentals; the setup is asymmetric over a 3–7 year horizon.
Read time 12 min
Length 12,253 chars
Category finance
Ideas
Outlier Capital Substack author, Outlier Capital
Article highlights persistent supply tightness for NVIDIA GPUs through 2027 and the company's $119B forward supply commitments as evidence of robust demand visibility, directly supporting NVIDIA's rev
Article highlights persistent supply tightness for NVIDIA GPUs through 2027 and the company's $119B forward supply commitments as evidence of robust demand visibility, directly supporting NVIDIA's revenue and margin outlook despite valuation concerns. Risk: Elevated multiples and customer concentration among a few hyperscalers could lead to sharp de-rating if AI monetization pace disappoints or a macro shock hits.
Outlier Capital Substack author, Outlier Capital
Article specifically calls out advanced packaging (CoWoS) as a persistent bottleneck, with management and supply-chain data indicating tightness likely through 2027 — directly benefiting TSMC's revenu
Article specifically calls out advanced packaging (CoWoS) as a persistent bottleneck, with management and supply-chain data indicating tightness likely through 2027 — directly benefiting TSMC's revenue from CoWoS capacity expansion and pricing power. Risk: Geopolitical risks (Taiwan stability) and cyclical semiconductor downturns could disrupt the buildout timeline.
Outlier Capital Substack author, Outlier Capital
Article identifies HBM (high-bandwidth memory) as another supply-constrained area, with tightness persisting through 2027; Micron is a leading HBM supplier directly leveraged to the hyperscaler GPU cl
Article identifies HBM (high-bandwidth memory) as another supply-constrained area, with tightness persisting through 2027; Micron is a leading HBM supplier directly leveraged to the hyperscaler GPU cluster buildout. Risk: Commodity memory pricing cycles and potential oversupply if demand growth decelerates faster than expected.
More from Outlier Capital

This newsletter, published July 20, 2026, features Outlier Capital discussing NVDA, TSM, MU. 3 trade ideas extracted by AI with direction and confidence scoring.

Speakers: Outlier Capital  · Tickers: NVDA, TSM, MU