1Q26 AI Channel Checks: Pt2 - GPUs vs ASICs

Research evidence excerpt

1Q26 AI Channel Checks: Pt2 - GPUs vs ASICs

Technology | Semiconductors and Semiconductor Equipment

May 18, 2026

Mark Lipacis 1Q26 AI Channel Checks: Pt2 - GPUs vs

214-459-4007

Mark.Lipacis@evercoreisi.com ASICs

Joseph McCormack We asked more than 25 AI engineers and experts working with a broad

212-446-9448 range of companies in the AI industry to share their thoughts on the

joseph.mccormack@evercoreisi.com competitive landscape for AI infrastructure. Main themes include: 1) AI

workloads are shifting from training-led regime toward inference-led

regime by end of 2026, 2) shift is increasing focus on cost-per-token,

Upcoming Industry Events ROI and TCO, which is accelerating hyperscaler interest in homegrown

ASICs and alternative accelerators, 3) NVDA remains broad ecosystem

Apr-17-2026: 1Q26 AI Channel Checks – Pt2: CPU Renaissance leader and market share is insulated in N-T given supply-chain capacity

Evolution investments, but moat is increasingly being tested at Hyperscaler /

Mar-13-2026: 4Q25 AI Channel Checks: Part 3 – LT Themes & Frontier-Lab scale, 4) Heterogeneous compute environments seem

Constraints increasingly likely for Inferencing, 5) “Inference at the Edge” interest

increasing due to lower cost/token, but need better small language

Feb-23-2026: 4Q25 AI Channel Checks: Part 3 – ASIC Partner models.

Competitive Positioning

Inference Becoming Main Workload. Checks point to shift from still

Feb-22-2026: 4Q25 AI Channel Checks: Part 2 – Networking & training-heavy 2025 towards majority-inference compute by Eo26, w/

Optics several contacts sizing inference at 50% to 60% by late 2026, materially

higher over time. Training still grows, but inference outgrows. (See CPU

Feb-17-2026: 4Q25 AI Channel Checks: Part 1 – GPU vs Custom Renaissance Evolution)

ASICs