The custom AI ASIC state of play (May 2026) (www.tomshardware.com)

Nvidia continues to dominate the AI chip market with a share of around 70%, but this is expected to decline as Google, Amazon, Meta, Microsoft, and OpenAI direct significant investments toward custom ASIC (Application-Specific Integrated Circuit) chips tailored to their unique workloads. ASIC shipments are projected to capture 27.8% of the market by 2026, marking the highest share since 2023, with a robust year-over-year growth rate of 44.6%—nearly three times that of merchant GPUs. This shift is largely facilitated by TSMC's advanced fabrication capabilities, enabling hyperscalers to leverage highly customized chip designs. Broadcom has emerged as a central player in this ecosystem, reporting a 106% increase in AI semiconductor revenue and targeting $100 billion in annual AI chip revenue by 2027, backed by a $73 billion AI backlog. Google's TPU v7 represents a leap in custom AI silicon architecture, achieving high utilization rates and lower total cost of ownership compared to rival offerings. Similarly, AWS's recent Trainium3 launch highlights the competitive race for performance, with each new iteration promising substantial improvements in compute power and memory bandwidth. As major companies increasingly adopt custom silicon solutions for AI workloads, the implications for Nvidia’s pricing power and market share are significant, shifting the landscape of AI hardware toward bespoke solutions that better cater to the burgeoning demands of AI inference workloads.