Semiconductor correction: reality check after strong rally
Semiconductor stocks have come under significant pressure in recent days. However, the correction follows an exceptional rally. Even after this week’s decline, the MSCI Semiconductor index1 remains around 45% higher year-to-date, although it is now more than 10% below its 2026 high. So far, this looks more like a reset in expectations than evidence of a fundamental deterioration in the semiconductor cycle or underlying AI demand.
What are the main reasons for the sharp downturn in semiconductor stocks? Several concerns have converged at the same time. First, AI returns and financing are increasingly becoming central concerns. Investors wonder whether the enormous capital committed to AI infrastructure will generate sufficient returns. Higher interest rates make this more challenging, as they lead to higher financing costs for investments in data centres and computing capacity.
In credit markets, investors also increasingly worry about financing risks related to AI. CoreWeave, one of the more highly leveraged AI infrastructure providers, has seen the cost of insuring its debt against default rise sharply (much higher spreads on credit default swaps). In other words, credit investors are demanding much more compensation for the risk of lending to the company. If financing becomes structurally more expensive, highly leveraged AI companies may have to moderate their data centre expansion and, ultimately, chip purchases. The cost of insuring against defaults by other AI-related companies has also increased.
Safety of AI called into question
AI safety is emerging as another potential constraint. The rise of ‘agentic AI’ – AI systems that can autonomously plan and carry out tasks – raises concerns around cybersecurity, liability and regulation. Anthropic, the company behind chatbot Claude, recently warned that autonomous AI agents could become increasingly difficult to control. Tighter safeguards are unlikely to stop AI development, but they could slow the rollout of commercial applications and delay returns, meaning current AI investments could take longer to pay off.
Competition from China
Chinese competition is amplifying the correction. Advances by semiconductor company CXMT, a Chinese rival of other memory chip makers, have raised concerns over future memory chip supply and competition. If supply increases faster than formerly expected, memory chip prices could fall quicker, putting pressure on manufacturers’ margins and profits.
Finally, the upcoming US midterm elections could add another layer of uncertainty. Political scrutiny of the rapid expansion of datacentres is increasing, particularly because of their energy use and the pressure they place on existing infrastructure. While unlikely to derail the AI build-out altogether, this could result in permitting delays, higher costs or a slower pace of capacity expansion.
Positive view, but more clarity needed
Still, there is little evidence of a fundamental deterioration in demand for semiconductors. This week, ASML mentioned its currently planned capacity will not be sufficient to meet customer demand in 2027 and 2028. This indicates that demand for AI-related chips remains strong in the near and medium term. Beyond 2028, demand is more uncertain. For demand to remain strong after 2028, today's enormous AI investments will have to translate into sufficient revenues and returns.
Be selective and monitor developments
For now, we see the correction in semiconductor stocks as a reset in expectations rather than the start of a downturn: in some areas, expectations of sustained, exceptionally high growth are being revised. For investors, however, this does not make every correction a buying opportunity. After the strong rally that preceded the correction, selectivity matters more.
We favour semiconductor companies with a strong order outlook , strategic positions in the semiconductor value chain and robust balance sheets. We are more cautious towards businesses that heavily depend on continued rapid growth in AI investment, as the focus of investors is increasingly shifting from AI investment to AI returns. Evidence that AI revenues are catching up with AI infrastructure spending would strengthen the case that today’s expansion in semiconductor capacity can be sustained well beyond 2028. Conversely, if AI generates lower profits than expected, financing costs remain persistently high, or material cuts are made to data centre investment, the risk of overcapacity and a weaker semiconductor investment cycle later in the decade would increase.
Our conclusion is therefore not to abandon the semiconductor theme, but to be more selective. The 2027-28 outlook remains strong; the case for growth beyond 2028 has yet to be proven.
Joost Olde Riekerink
Equity research and advisory expert