On June 10, 2026, U.S. May CPI came in at 4.8% year-over-year—well above the consensus forecast of 4.3%. The data reignited speculation about a potential Federal Reserve rate hike in December and triggered an immediate sell-off across the semiconductor sector. Shares of onsemi (ON), Teradyne (TER), and Penguin Solutions (PENG) plunged 7.2%, 6.5%, and 9.1% respectively. While this appears to be a textbook reaction to macroeconomic tightening, a closer look reveals deeper structural vulnerabilities that leave these companies disproportionately exposed in the current phase of the AI investment cycle.
Take onsemi first. Despite its leadership in automotive and industrial power semiconductors, less than 15% of its revenue is directly tied to AI workloads. The company has committed heavily to silicon carbide (SiC) capacity expansion, with 2025 capital expenditures reaching $2.8 billion—60% allocated to new fabs in Arizona and the Czech Republic. These are long-cycle investments with payback horizons of 5–7 years. In a high-rate environment, the cost of debt servicing rises sharply. As of Q1 2026, onsemi’s net debt-to-equity ratio stood at 42%, and free cash flow had declined by 19% year-over-year. When investor sentiment shifts from “growth at all costs” to “capital discipline,” asset-heavy players with limited AI exposure become prime targets for de-rating.
Teradyne’s position is more nuanced. As one of the world’s top two semiconductor test equipment vendors, its traditional logic chip testing business faces dual pressure: advanced-node chips now achieve higher yields, reducing per-unit test time, while leading AI chipmakers like NVIDIA and AMD increasingly rely on built-in self-test (BIST) architectures to minimize dependence on external ATE (automatic test equipment). Although Teradyne has diversified into system-level testing through its acquisition of MiTAC and is building a presence in AI server burn-in validation, this segment contributed only 11% of total revenue in 2025. More critically, its largest customers—TSMC and Samsung—have slowed their 2026 capex growth to single digits, directly compressing near-term equipment order visibility.
Penguin Solutions represents a different risk profile altogether. Once celebrated for its liquid-cooled NVIDIA HGX infrastructure solutions, the company operates essentially as a high-end engineering services firm dependent on a handful of mega-customers. Its 2025 financials show that its top three clients—including an unnamed North American cloud giant and SpaceX—accounted for 78% of total revenue. Such engagements are inherently project-based and non-recurring. The timing proved disastrous: just days before the market selloff, SpaceX filed for a $1.77 trillion IPO—the largest tech offering in history. The announcement forced hedge funds to rebalance portfolios and meet margin calls, triggering forced outflows from peripheral tech names like Penguin.
It’s crucial to clarify that SpaceX’s IPO wasn’t the root cause of the semiconductor decline. Rather, it acted as a catalyst that accelerated the market’s reassessment of “AI adjacency.” Investors are now sharply differentiating between core enablers of the AI stack—companies like TSMC, SK Hynix, and Broadcom—and secondary beneficiaries. onsemi, Teradyne, and Penguin fall squarely into the latter category.
I judge that this correction is not yet complete. Should the Fed signal a more hawkish stance at its September FOMC meeting, semiconductor firms with low capital efficiency and weak AI integration will face continued pressure. Those still executing large-scale capacity builds without long-term customer take-or-pay agreements are especially vulnerable to both valuation compression and fundamental downgrades.
The broader question lingers: has the market overgeneralized the equation “semiconductors = AI”? As liquidity tightens and capital becomes scarce, only companies with irreplaceable technology nodes, binding customer contracts, and resilient balance sheets will endure. For onsemi, Teradyne, and Penguin, the next 12 months will test whether they can prove they are not merely AI-adjacent—but indispensable.