Broadcom lost 19.5% between June 3–5, 2026, as the US 2-year Treasury yield surged to a 16-month high — pulling down technology stocks from New York to Tokyo.1 NVIDIA dropped 6% on June 5 alone. Meta shed 6% across the same stretch. The leveraged FNGU ETF, tracking mega-cap tech, collapsed 22% over five days.1
Rising short-term US yields tighten financial conditions globally. When American Treasuries offer higher risk-free returns, capital rotates out of high-multiple growth equities worldwide — a mechanism felt as acutely in Frankfurt and Seoul as on Wall Street.
Broadcom's own guidance undermines any AI-demand explanation for the drop. The company raised its Q3 FY26 AI semiconductor forecast during the selloff.1 Customers pulling back on chip orders do not produce upward guidance revisions. The data points in opposite directions.
NVIDIA faces the same disconnect. AI infrastructure spending indicators held firm through the period, yet shares fell with the broader sector. When leading semiconductor companies decline in unison on a bond yield print — not on earnings warnings — macro rate sensitivity is the cleaner explanation.
Dell's Q1 results reinforce that picture. The company reported $24 billion in AI server order bookings and raised its full-year revenue guidance to approximately $60 billion.1 Global hyperscalers and sovereign AI initiatives across the Gulf, Asia, and Europe are sustaining that pipeline. Order backlogs of that scale do not accumulate during a genuine AI spending freeze.
The distinction carries real consequence for international investors. A rate-driven selloff is mean-reverting if yields stabilize or retreat. A fundamentals-driven decline — customers canceling AI infrastructure budgets — requires earnings downgrades and a longer recovery. Current evidence favors the former.
AI earnings revisions across the sector have not turned negative. Until they do, sharp price declines during yield spikes are valuation compression, not business deterioration. Investors reading a structural AI slowdown into this data are misreading the signal.
