TrendForce projects ASIC shipment surge

Why it matters

TrendForce’s latest AI server forecasts point to a narrower claim than “GPUs are losing.” The firm said cloud service providers’ in-house AI ASIC shipments are expected to grow 44.6% in 2026, versus 16.1% for GPUs, even as GPUs remain the largest category in AI servers. That matters because the forecast is about growth rates and shipment mix, not an outright handoff in market leadership. TrendForce said GPUs are still expected to account for 69.7% of AI server shipments in 2026, while ASIC-based systems rise to 27.8%, the highest share since 2023. ### So what is actually growing faster here? TrendForce’s October 2025 analysis said “CSPs’ in-house ASICs” are the segment projected to grow 44.6% in 2026, compared with 16.1% for GPUs. (trendforce.com) That means custom chips designed or commissioned by large cloud providers are expanding faster than merchant accelerators sold broadly into the market. The January 20, 2026 TrendForce server outlook added that shipment growth for ASIC AI servers is projected to outpace GPU-based systems, with North American cloud spending, sovereign cloud projects and edge AI inference among the drivers. (trendforce.com) ### Does that mean Nvidia’s position is breaking? TrendForce said no. Its October note said the faster ASIC growth “does not mean NVIDIA’s dominance will collapse overnight,” because the contest is shifting beyond chip performance into interconnects and software ecosystems. (trendforce.com) Nvidia’s own response, as described by TrendForce, includes opening NVLink for customer ASIC integration and introducing Rubin CPX/VR200 NVL144 CPX architectures aimed at inference workloads. (trendforce.com) TrendForce framed those moves as an attempt to extend Nvidia’s position from accelerators into the interconnect and platform layers. ### Why are cloud providers pushing custom chips now? (trendforce.com) TrendForce said the mix is changing as inference becomes a larger commercial workload. Its January forecast said the market from 2024 to 2025 centered on training large language models with GPU-heavy AI servers, but that the second half of 2025 brought more focus on inference services tied to AI agents, LLaMA-based applications and Copilot upgrades. (trendforce.com) That shift changes the economics. TrendForce said inference workloads now run not only on dedicated AI server racks but also on general-purpose servers handling pre- and post-inference computing and storage, which broadens the hardware stack around AI deployment. ### Which companies are named in the forecast? TrendForce named Google and Microsoft as likely leaders in expanding procurement of general-purpose servers for inference traffic from Gemini and Copilot services. (trendforce.com) It also said Google and Meta are expanding their own ASIC efforts, with Google “positioned to become a leading market player.” The same forecast said the combined capital expenditures of Google, AWS, Meta, Microsoft and Oracle are expected to rise 40% year over year in 2026. (trendforce.com) TrendForce tied part of that spending to infrastructure buildouts and part to replacing general-purpose servers bought during the 2019-2021 cloud investment cycle. ### What should readers take from the numbers? (trendforce.com) The clearest takeaway is that AI hardware demand is becoming more mixed, not less. TrendForce’s numbers show merchant GPUs still dominating shipment share, while custom ASICs gain ground faster as hyperscalers chase cost, efficiency and inference-specific designs. For buyers and software teams, the practical implication is compatibility pressure. (trendforce.com) TrendForce said the battle has moved toward interconnects, switches, software and ecosystem support, which means hardware choice increasingly depends on whether workloads can move across more specialized systems. TrendForce’s next public benchmark in this story is its 2026 AI server shipment outlook, published January 20, with Google, Meta, Microsoft, AWS and Oracle named among the companies shaping demand. (trendforce.com 1) (trendforce.com 2)

Key numbers

• TrendForce said custom AI ASIC shipments are projected to grow faster than merchant GPU shipments in 2026, as cloud providers expand in-house chip programs.
• TrendForce put 2026 growth for CSP in-house ASICs at 44.6%, versus 16.1% for GPUs, while GPUs still lead 69.7% of AI server shipments.
• TrendForce’s January 20, 2026 server forecast said Google and Meta are among North American firms expanding ASIC efforts.
• TrendForce said GPUs are still expected to account for 69.7% of AI server shipments in 2026, while ASIC-based systems rise to 27.8%, the highest share since 2023.

What happens next

• TrendForce said GPUs are still expected to account for 69.7% of AI server shipments in 2026, while ASIC-based systems rise to 27.8%, the highest share since 2023.
• Its October note said the faster ASIC growth “does not mean NVIDIA’s dominance will collapse overnight,” because the contest is shifting beyond chip performance into interconnects and software ecosystems.
• TrendForce’s next public benchmark in this story is its 2026 AI server shipment outlook, published January 20, with Google, Meta, Microsoft, AWS and Oracle named among the companies shaping demand.

Sources

• trendforce.com
• trendforce.com

Quick answers