The April 15, 2026 Bloomberg newsletter tentatively revives discussion of an AI bubble, noting that "there’s still a lot of investor hype." While factually accurate, the coverage remains surface-level, focusing on valuation multiples without examining the extreme market concentration, absent productivity translation, or the intertwined geopolitical vulnerabilities that distinguish this cycle from prior manias.

Primary documents paint a clearer picture. The Stanford AI Index 2026 reports private AI investment exceeded $250 billion in 2025, with over 65% flowing to just seven firms. NVIDIA’s latest 10-Q filing shows data-center revenue growing 180% year-over-year, yet BLS productivity statistics through Q1 2026 reveal only marginal gains outside the tech sector itself. The Bank for International Settlements Annual Economic Report 2025 explicitly flags "signs of exuberance in select technology sub-sectors" and warns of "crowded trades" where correlated selling could amplify shocks.

What mainstream cheerleading continues to miss is the feedback loop between concentrated capital expenditure and geopolitical exposure. U.S. Commerce Department export control updates from October 2025 tightened restrictions on advanced AI chips to China, yet the same documents acknowledge that domestic data-center buildouts remain dependent on TSMC fabrication in Taiwan. A correction in AI-related equities would not merely deflate valuations; it could constrain the very capital needed to sustain U.S. technological primacy at a moment when Beijing’s state-directed semiconductor program (outlined in its 14th Five-Year Plan progress report) accelerates.

Two perspectives emerge from primary records. Optimists, including testimony by OpenAI executives before the Senate Banking Committee in December 2025, frame current spending as infrastructure for transformative general intelligence, comparable to electricity rollout. Skeptics, citing Meta’s publicly released scaling-law research updated in March 2026, demonstrate flattening returns on additional compute, suggesting the marginal dollar is increasingly inefficient. Neither view is dispositive, yet both are underweighted in favor of earnings-call optimism.

Patterns from the dot-com era are instructive but imperfect. SEC post-mortem studies of the 2000-2002 crash showed that while many companies vanished, underlying network technologies endured. Today’s AI infrastructure, however, is far more capital-intensive and geographically concentrated. The dominant 2025-2026 market theme is precisely this narrowing of risk into a handful of balance sheets. Should returns disappoint, the policy consequences—potential bailouts, tightened export controls, or fractured international tech standards—extend well beyond financial markets into national security and global competitiveness.

Credible discussion of an AI bubble therefore cannot stop at hype. It must address how concentrated investment creates single points of failure that link U.S. market stability to semiconductor chokepoints in East Asia. Policymakers reviewing Federal Reserve Financial Stability Reports from 2025 would be wise to model scenarios where AI capex reversal coincides with heightened cross-strait tensions. The Bloomberg piece opens the conversation; the deeper record shows why it matters far more than Wall Street narratives suggest.