AI Hardware Architecture

A new generation of AI-native chips — Nvidia's Vera Rubin, AMD's EPYC Venice, and Astera Labs' Scorpio X-Series — is racing to market as hyperscalers accelerate infrastructure buildouts. Leading-edge ASIC design now costs hundreds of millions of dollars, locking out newcomers and concentrating market power among incumbents. Supply chain fragility and U.S.-China semiconductor rivalry add structural risk to an already volatile transition.

NVIDIA is ramping Blackwell production and targeting Rubin Ultra architecture for 2027 while facing intensified competition from AMD's Meta partnership and custom silicon from Google and Baidu. Simultaneously, Apple and TSMC are bringing advanced chip manufacturing to the US, driven by geopolitical pressures and Trump administration infrastructure requirements for AI hardware.

Aehr Test Systems projects $60-80M in bookings for the second half of FY2026, driven by AI chip burn-in testing demand. The company secured orders for high-power Sonoma systems capable of testing devices at 2,000 watts, with a lead customer forecasting shipments starting Q1 FY2027.