AI infrastructure trade loses steam amid earnings miss and competition
In brief
- Samsung reported record Q2 earnings but missed revenue estimates, triggering semiconductor selloff
- Micron and SK Hynix declined sharply on hyperscaler AI infrastructure spending slowdown concerns
- Investors debate whether next-phase AI demands more GPUs or efficiency gains will reduce chip demand
- Domestic hardware alternatives emerge as lower-cost competitors to cutting-edge US semiconductors
Earnings miss triggers selloff
Samsung Electronics reported record second-quarter earnings but missed revenue estimates. The miss rattled the sector immediately. Semiconductor and memory stocks including Micron Technology and Sandisk fell nearly 7% on Tuesday following the report. The sell-off wasn't isolated—it reflected a broader anxiety about whether the AI infrastructure narrative still holds.
The damage is significant when you zoom out. Sandisk gained more than 525% in 2026, Micron gained over 120%, and SK Hynix climbed roughly 225% in 2026. Those gains are now under pressure. SK Hynix is down 25% from its all-time high ahead of its U.S. listing this week.
The spending question
Two concerns are colliding. Concerns are growing that hyperscalers could slow AI infrastructure spending. At the same time, investors are increasingly questioning whether the next phase of AI will require ever more GPUs and high-bandwidth memory, or whether more efficient models will reduce demand for infrastructure.
This isn't idle speculation. China's Zhipu AI, one of the country's leading artificial intelligence startups, is exploring a custom AI chip as demand for its open-source GLM models surges, highlighting a structural shift: lower-cost AI ecosystems are being built around domestic hardware rather than cutting-edge US chips.
The thesis that drove the trade—infinite GPU demand, ever-rising capex—is being tested.
