Lattice Semiconductor, Amkor, Himax, Teradyne, and Kulicke and Soffa don’t make the chips everyone argues about. That’s precisely why their sudden drop matters.
None of these companies design the GPUs that dominate AI headlines. Lattice sells low-power programmable logic. Amkor handles outsourced assembly and test. Himax makes display drivers. Teradyne builds test equipment, and Kulicke and Soffa specializes in the packaging and bonding machinery that turns bare silicon into something usable. Together they form the unglamorous connective tissue of the semiconductor industry. When that tissue twitches, it’s worth paying attention.
The selloff wasn’t isolated to a handful of mid-caps. SK Hynix and SoftBank shares fell sharply in Asia as the weakness tracked westward, a reminder that semiconductor equities now trade as a single global organism. A tremor in one time zone becomes an earthquake in the next. The proximate cause, as is often the case lately, was anxiety about whether AI infrastructure spending can keep climbing at its current slope.
That anxiety has been building for months. Every earnings cycle brings a fresh round of questions about whether hyperscalers are overbuilding, whether model training costs will eventually plateau, and whether the revenue from AI services can justify the capital being poured into data centers. Those questions rarely get satisfying answers, so the market does what it always does: it oscillates.
What’s interesting here is the composition of the decline. If investors were simply souring on AI accelerators, you’d expect the pain to concentrate in the companies selling directly into that market. Instead, the damage spread to equipment vendors and packaging specialists, the companies whose order books depend on fab construction and capacity expansion rather than any single product cycle. Test and assembly suppliers sit downstream of decisions made years in advance. When their stocks fall, the market is pricing in something slower and more structural than a single bad quarter.
There’s a genuine tension in the data. On one hand, advanced packaging has become one of the most constrained parts of the AI supply chain. Stacking memory and logic into a single package, whether through techniques like chiplet integration or high-bandwidth memory stacking, requires exactly the kind of equipment and test capacity these companies provide. Demand for that work has been climbing, not falling. On the other hand, the broader semiconductor cycle remains cyclical. Automotive, industrial, and consumer end markets have been soft, and those markets still account for a meaningful share of revenue at companies like Lattice and Himax.
Macro concerns aren’t helping. Trade policy uncertainty, currency fluctuations, and the persistent question of how much capacity China will build domestically all weigh on sentiment. Taiwan’s outsized role in the ecosystem remains both a competitive advantage and a systemic risk, a fact that analysts have flagged for years without any real resolution. Every geopolitical flare-up sends the same shiver through the same stocks.
Meanwhile, the technology itself keeps moving. Research into quantum computing continues to attract government money, with the Pentagon recently committing significant funding to the field. Quantum error correction, long the field’s central obstacle, is inching toward practical demonstrations. None of that threatens classical semiconductors this decade, but it does shape how investors think about the very long arc of computing. And AI agents, the current darling of the software world, ultimately run on the same silicon everyone is now second-guessing.
It’s also worth noting what the selloff doesn’t tell us. Stock prices reflect sentiment as much as fundamentals, and semiconductor equities have always been volatile. A sharp pullback after a long run is not the same thing as a demand collapse. The companies caught up in this wave still hold real positions in supply chains that take years and billions of dollars to replicate. Amkor’s packaging capacity doesn’t appear overnight. Teradyne’s test platforms don’t have obvious substitutes. Kulicke and Soffa’s bonding tools sit inside fabs that cost more than most companies are worth.
The more useful question isn’t whether these stocks fell. It’s whether the underlying orders are slowing. That answer will show up in earnings calls and capex guidance over the next few quarters, not in a single trading session. If hyperscaler spending holds and advanced packaging demand keeps growing, this looks like noise. If guidance starts to soften across equipment and materials suppliers, it’s an early warning that the AI buildout is entering a more mature, less breathless phase.
For now, the market has voted for caution. The supply chain, as always, will have the final word.