Broadcom reports $73B AI custom-ASIC backlog with $100B annual revenue target by 2027 — custom shipments growing 44.6% YoY vs 16.1% for merchant GPUs confirms structural shift
Broadcom's $73B backlog and 44.6% custom-ASIC growth rate (nearly 3x merchant GPU growth) confirms the structural shift in hyperscaler procurement away from NVIDIA for inference. The custom-silicon path now has a measurable revenue trajectory through 2027, not just a capacity story.
The substantive piece is the growth-rate differential. Through 2024, the custom-ASIC vs merchant-GPU competitive narrative was qualitative — hyperscalers preferred custom silicon for inference cost optimization, but specific revenue commitments weren't publicly visible. Broadcom's $73B backlog disclosed at 44.6% YoY growth provides the first hard-number anchor: custom ASICs are growing ~3x faster than merchant GPUs, scaled to billions in revenue. The H2 2026 hyperscaler-procurement narrative now has a structural underpinning rather than a directional speculation.
The connection to TSMC's HPC-overtakes-mobile crossover is that both data points triangulate the same structural shift: AI-compute demand growth at TSMC plus custom-ASIC growth rate at Broadcom together confirm that hyperscaler capex allocations through H2 2026-2027 will accelerate toward custom-silicon designs. NVIDIA holds the training-tier dominance; inference-tier procurement is structurally restructuring toward custom-ASIC alternatives.
Tom's Hardware — Custom AI ASICs Examined: From Broadcom to MTIA → · Heygotrade — AI Semiconductor Stocks 2026: NVIDIA, TSMC, Broadcom → · Finviz — Taiwan Semiconductor and Meta Hidden Bull Case for Broadcom →