AMD closed a 6-gigawatt GPU deal with Meta for AI training infrastructure, marking one of the largest single deployments globally as NVIDIA targets a 2027 launch for its Rubin Ultra platform. The contract's power allocation equals the electricity consumption of a mid-sized European city, reflecting the massive compute resources tech giants now dedicate to AI development.
The competition has shifted beyond chip specifications to integrated ecosystems spanning fabrication, power delivery, and data center infrastructure. Apple and TSMC deepened their US manufacturing partnership to build domestic production capacity for advanced AI chips, addressing supply chain sovereignty concerns that have intensified across Western markets amid geopolitical tensions with China.
Specialized hardware suppliers are capturing growth from the global AI buildout. California-based Aehr Test Systems reported a major production customer provided "very large forecast" with shipments starting Q1 FY2027. The company expects $60M to $80M in bookings for second half FY2026, driven primarily by AI wafer-level and packaged-part burn-in testing. Sonoma system orders reached $5.5M in Q3, exceeding Q2's entire total.
Credo Technology Group forecasts GAAP gross margins between 63.8% and 65.8% for Q3 FY2026, powered by demand for high-bandwidth interconnects that prevent bottlenecks as GPU clusters scale to thousands of accelerators. Taiwan-based TSMC remains the dominant foundry for cutting-edge AI chips, though US and European governments are funding domestic alternatives to reduce reliance on Asian manufacturing.
Aehr Test's new Sonoma configurations handle up to 2,000 watts per device, addressing thermal demands of next-generation accelerators. The company partnered with South Korea's ISE Labs and Taiwan's ASE to expand testing services for semiconductor customers developing high-performance computing and AI chips across Asia, North America, and Europe.
The AI hardware stack now requires capabilities from silicon design through data center integration. Companies winning contracts must demonstrate expertise across power infrastructure planning, thermal management, and advanced packaging as AI workloads consume unprecedented compute resources across global markets.