Signal Custom AI Accelerators (ASICs) Outpace GPUs — Hyperscalers Pursue In-House Silicon at Scale

Summary

TrendForce projects 44.6% growth in custom AI accelerator (ASIC) shipments in 2026 against 16.1% growth for merchant GPUs, marking the first year ASIC growth has outpaced GPU growth since the two categories began being tracked separately. HPCwire describes 2026 as the "ASIC acceleration point," with Google's TPU v7, Amazon's Inferentia/Trainium, Microsoft's Maia/Cobalt, Meta's MTIA and an OpenAI-Broadcom co-designed accelerator all reaching production volume. Broadcom serves as the design partner behind Google's TPU, Meta's MTIA, Microsoft's Maia, and, as confirmed in early 2026, the Titan accelerator program for OpenAI and Anthropic. Midjourney reported cutting its monthly compute costs from about $2.1 million to $700,000, a 65% reduction, after moving inference workloads from Nvidia GPUs to Google's seventh-generation TPUs. Nvidia currently holds more than 90% of the AI accelerator market, but New Street Research projects its share of inference-specific compute could fall to 20-30% by 2028 as hyperscaler ASICs reach scale, even as the training market remains Nvidia's stronghold due to CUDA and NVLink advantages.