Saturday, April 18, 2026
Search

AI Chip Testing Firms Land $60-80M Orders as ASIC Demand Outpaces GPU Supply

Aehr Test Systems projects $60-80M in bookings for AI chip burn-in equipment despite Q2 revenue falling 27% to $9.9M. The company secured orders for high-power Sonoma systems testing chips at 2,000 watts per device, with its lead customer forecasting volume shipments starting May 2026. Semiconductor packaging rival Amkor is expanding Arizona capacity while memory makers push HBM3e bandwidth to address AI's computational bottleneck.

AI Chip Testing Firms Land $60-80M Orders as ASIC Demand Outpaces GPU Supply
Image generated by AI for illustrative purposes. Not actual footage or photography from the reported events.
Loading stream...

Aehr Test Systems forecast $60-80M in equipment bookings for the second half of fiscal 2026, driven by AI chip manufacturers ramping wafer-level and packaged-part testing. The Fremont-based firm reported Q2 revenue of $9.9M, down 27% year-over-year, but posted $18.3M in effective backlog including $6.5M in orders received during the first six weeks of Q3.

The company's Sonoma systems are testing AI accelerators at power levels up to 2,000 watts per device, reflecting the extreme thermal demands of next-generation chips. Aehr's lead production customer provided a "very large forecast" with shipments expected to begin in Q1 fiscal 2027, starting May 30, 2026. CEO Gayn Erickson stated production capacity exceeds 20 systems per month for both wafer-level and packaged configurations.

NVIDIA maintains GPU dominance with Hopper and upcoming Blackwell architectures, but supply constraints have pushed hyperscalers and enterprise customers toward custom ASICs. SambaNova's Reconfigurable Dataflow Units represent one alternative approach, while memory manufacturers address the "memory wall" limiting AI performance. Samsung's HBM3e high-bandwidth memory shipments target this bottleneck.

Amkor Technology is expanding semiconductor packaging capacity in Arizona, part of a broader U.S. manufacturing buildout. The facility will handle advanced packaging techniques required for chiplet-based AI processors and 3D-stacked memory integration. Packaging and testing represent critical steps between wafer fabrication and final deployment, with burn-in testing catching defects before chips reach data centers.

Credo Technology, which supplies high-speed connectivity for AI clusters, guided Q3 gross margins between 63.8% and 65.8%, indicating healthy pricing power in the interconnect market. Ensurge Micropower is developing microbatteries for AI-enabled edge devices, addressing power delivery challenges as inference moves closer to end users.

The hardware infrastructure race extends from chip design through manufacturing, testing, and assembly. Aehr's $2M in delayed gallium nitride WaferPak shipments—pushed from Q2 to Q3 due to high-voltage protection circuit redesigns—illustrates technical challenges in scaling production. The company closed its Incal facility in May 2025, consolidating operations to Fremont while maintaining claimed capacity of 20+ systems monthly.