Chip Equipment Stocks as a Lead Indicator of the AI Buildout

This series assumes the AI buildout (the ongoing construction of data centers and the chips that fill them) is real and runs for several years. That assumption is worth checking against market data periodically. During the week of June 9, 2026, AI hardware stocks fell sharply, so I looked at which part of the supply chain the selling hit.

The decline concentrated in the chip designers (companies that design processors but contract out the manufacturing). For the week: Arm fell 25%, Qualcomm 24%, Broadcom 22%, and AMD 17%. Demand had not dropped. Broadcom reported record quarterly AI revenue, and its stock fell because management held its AI revenue target flat rather than raising it, which investors read as a limit on growth. The chip-equipment makers (companies that build the machines used to manufacture chips) did not fall the same way. ASML, Applied Materials, Lam Research, and KLA stayed in a slow upward trend through the same week.

The two groups report on different timelines, which is why I track the equipment makers separately. A chip designer’s quarterly results describe the current cycle: orders already booked and product already sold. A chip-equipment maker’s order book describes a later cycle, because its customers (the chip factories, or fabs) order a machine like an EUV lithography system (the equipment that prints the finest circuit patterns onto a chip, priced around $200 million each) one to two years before the factory using it produces chips at volume. That lead time means the equipment makers’ orders reflect demand roughly four to six quarters ahead of the designers’ results.

For planning purposes at AWS, the equipment-maker trend is a useful check on whether the buildout is a single cycle or a multi-year one. A sustained decline in ASML and Applied Materials orders while the designers are still selling would be an early signal that the buildout is slowing, and a reason to revisit plans that assume inference (running trained AI models in production) keeps getting cheaper and more abundant. As of mid-June 2026, those equipment makers continued to trend upward through the broader selloff, so the assumption underneath Bedrock capacity planning and frontier-model economics still holds. I will note in a future entry if that changes.